A post doing the rounds on LinkedIn this month tells business owners to build a page on Google Sites and link it to their website. It won't help. Why it won't help is the most useful thing you can learn about SEO: Google is trying to show people the result they were looking for, and it has spent more than 25 years getting very good at not being tricked.

The advice doing the rounds

The idea goes like this. Make a free page on Google Sites named after the search you want to win. Put your business on it and link to your website. Some people go further and call it "Google stacking": Google Docs and Sheets full of links, all linked from a Google Site, on the theory that anything on Google's own domains carries extra trust. Gig sites sell it as a package for a few dollars.

It sounds clever because it uses Google's name. Here's what Google has actually said:

• On Google Sites, John Mueller of Google said in 2023 that the pages can be indexed but the platform isn't ideal for SEO, and suggested using your own domain instead.
• On building links from Google Sheets, back in 2017, his whole answer was: "Don't do that, do something useful instead."
• On "trust" from Google's own domains, he said in 2021 that Google has no metric specifically for trust. There's no special pass for being on sites.google.com.
• On links you create yourself on free platforms, he said in 2020 that this kind of activity has zero value for SEO.

We looked at a public Google Sites page. Its outbound links go through a google.com redirect. How Google treats those for ranking isn't public, and I won't pretend to know. It doesn't matter much anyway. A page nobody visits, linking to your site, is not a reason for anyone searching to want your page.

What Google is actually trying to do

Google has one job when you search. Put the result you were looking for at the top, so you're happy with Google and come back. Everything else is machinery for that. Here's what's public about the machinery.

It reads the web, then picks

Google's own explainer splits Search into three stages. Crawling: Googlebot finds pages through links and sitemaps. Indexing: it works out what each page is about, groups duplicates and picks one version to keep. Serving: when someone searches, it ranks what it has by relevance and quality, adjusted for their location, language and device. Not every page makes it through. Google also says it doesn't take payment to crawl a site more often or rank it higher.

Lots of systems, not one score

Google publishes a list of its ranking systems. It names more than a dozen: BERT, MUM, RankBrain and neural matching for understanding language; passage ranking for finding the right part of a long page; freshness, reviews, original content, reliable information and site diversity; deduplication; and SpamBrain for spam. Links are on the list too: link analysis and PageRank, which Google calls one of its core systems from launch. One system among many, not the whole game. A Google engineer, Gary Illyes, told a conference in 2023 that links aren't in the top three things that matter, and in 2024 that Google needs very few links to rank a page.

It learns from what people pick

This is the part the tricks ignore. In the US government's antitrust case against Google, the judge's 2024 ruling describes a system called Navboost. It pairs searches with pages by remembering what people clicked, using about 13 months of search data. A Google document quoted in the ruling calls Navboost one of the most powerful ranking components it has had. A related system, Glue, does the same for everything else on the results page. In 2024 a set of Google's internal API documents leaked too, with attributes named goodClicks and badClicks. Google warned against reading too much into them, and nobody outside Google knows how much each one counts.

You don't need the weights to see the shape. Google watches which result people choose for a search, and whether that choice seemed to do the job.

And people check the results

Google also has people outside the company, its search quality raters, rate results against a 182-page set of guidelines. The main question they answer is whether a result fully meets what the searcher needed, or fails to. Google says no single rating changes any page's ranking. The ratings tell Google whether its changes are making results better, so the whole system gets tuned towards the answer a real person would call right.

Why every trick has the same flaw

Every SEO trick fakes a sign that a page is the right answer, instead of making it the right answer.

• Links used to mean somebody vouched for you. So people bought them, swapped them and built networks of sites to make them. Google built SpamBrain to spot both the sites buying links and the sites selling them, and says it neutralises those links.
• Keywords used to mean a page was about something. So people stuffed them in. Keyword stuffing is now a named spam policy.
• Dates used to mean a page was fresh. So people changed the date without changing the page. Google's guidance on helpful content lists exactly that as a warning sign.
• Big websites rank well, so people rented pages on them. In 2024 Google made that its own spam policy, and later said it applies whether or not the big site is in on it.

Google earns its money because people trust its results. A trick that works is a bug in Google's product, and Google has every reason to fix it. Even when one works for a while, you're building on something you'll lose, and the page you made in the meantime did nothing for the customer who landed on it.

The rubbish, one by one

What actually works

Be the result the person was looking for. That sounds like a slogan, so here's what it means on a Tuesday morning:

1. Find the searches you deserve to win. In Search Console, look for searches where you already show up but people don't click, or where you sit just off page one.
2. Answer the search in the first lines. Someone typed a question. The first sentence of your page should answer it, with a number if there is one.
3. Cover what a reader expects. Read the pages that already rank and note what most of them explain. A reader will expect yours to explain it too, where it's true for you.
4. Add what nobody else has. Your real prices, your photos, your numbers, your answer to the question customers ask you on the phone. This is what makes Google's systems, and the person searching, pick you over the five pages that say the same thing.
5. Make sure Google can read it. No accidental noindex, no answer hidden in an image, a page that loads fast on a phone, links to it from your other pages.
6. Get talked about by real places. A local paper, a supplier's page, an industry list. One mention from somewhere people read beats a hundred from places nobody does.
7. Judge it properly. Wait four weeks, then compare the page's clicks with the four weeks before, against how the rest of your site moved. If everything rose 20%, your change didn't do anything you can see.

None of it is a hack. It's slower to explain and faster to work, and nothing Google ships next year can take it away from you.

Why I built Dinkum this way

Dinkum does that list, one change at a time, five mornings a week. Each change is a Dink: a pull request on your own site that you approve or close. Then it judges every Dink at 14 and 28 days against the rest of your site and tells you plainly whether it worked.

Its rules also say what it won't do. It won't build pages on Google Sites, Medium or Google Docs to rank in your place. It won't build rings of sites linking to each other, and it won't use indexing services. That's not because we're saints. None of it lasts, and none of it helps the person who searched.

Everything above you can do yourself, and our guides give away every step. Start with how Dinkum works, or how to get named in ChatGPT, which runs on the same idea: be the right answer, then make it easy to find.

By Mitch Flindell · 16 October 2026

Short version: every page Dinkum writes starts from a brief of what Google's top five results share. Now, once the page is live, Dinkum reads those results again every week. When three or more of them start covering something your page doesn't, a refresh goes on your calendar, with the new items named.

Why we built it

A brief is a snapshot, and search results don't sit still. New pages climb in, old ones fall out, and what the top pages have in common shifts with them. A page that matched them a month ago can fall behind without anyone touching it, and a rank tracker only tells you once it has slipped.

We saw it on our own site first. For Stickybeak's guide on "how to check website traffic", all five of Google's top results were replaced between 8 and 15 October. The new five are tools that estimate any site's traffic, and between them they share two sections and eleven names and phrases the guide doesn't have, like competitor traffic and traffic estimates. Bounce rate and traffic sources dropped out. Google had changed its mind about what that search is for.

What happens now

• A week after a page's brief, and every week after that, Dinkum looks at Google's results for its search again and reads the top five pages.
• It compares them with the brief: what three or more of them share now that the brief didn't have, and what dropped out. Then it checks your live page for each new item.
• A new shared section your page lacks, or two or more new names and phrases, puts a refresh on the next free refresh day. Anything smaller is noted and counted again next week.
• The refresh is a normal Dink: a fresh brief, the page rewritten to cover what's true and useful, a pull request for you to approve, and a verdict at 14 and 28 days.
• Skip a refresh and it won't come back unless something else changes.

Where you'll see it

Under SEO health, then Briefs, each brief has a new column for what the top results have started sharing since, and opening one shows the full list of what's new and what dropped out. This is our demo dental practice:

It runs for every page Dinkum has written from a brief, with nothing to switch on. There's more on how it decides, and the Stickybeak search in full, on the SERP monitoring page.

Short version: most SERP monitoring tells you where you rank. Dinkum's watches what the pages above you say. Once a week, for every page it has written, it reads Google's top five results again and compares them with the brief the page was written from. When three or more of them start covering a section, a name or a phrase your page doesn't, it puts a refresh on your calendar with the new items named, and you approve it like any other Dink.

Rankings move for a reason. Usually it's that the pages Google prefers have changed: new ones have climbed in, and between them they now answer something yours doesn't. A rank tracker shows you the drop after it happens. Reading the pages that rank shows you the cause, often before the drop.

What changed on one search in a week

Stickybeak is our own analytics product, and Dinkum writes its guides. On 8 October, the top five Google results for "how to check website traffic" were a mix of guides and checker pages. By 15 October, all five had been replaced.

The new five are tools that estimate any website's traffic, written for people sizing up a competitor. The old five were mostly about reading your own analytics. Google changed its mind about what this search is for, and the guide hadn't caught up. Reddit's page can't be read without signing in, so the counts are out of the four pages that could be.

Not every row deserves a paragraph. "Competitor traffic" and the bulk-check section are real gaps for a guide on this search. Bounce rate dropping out says the search has moved away from on-site analytics. The refresh covers what's true and useful to the reader, and says what it skipped.

What a check looks like in your dashboard

Every check is kept under SEO health, then Briefs, on the brief it was compared with. This one is on our made-up demo dental practice, so the numbers are examples, but the screen is the real one. The top results for "teeth whitening brisbane" started comparing whitening strips with a dentist and answering whether veneers and crowns can be whitened. The page had neither, so a refresh went on the calendar.

How it works

1. Pick the pages to watch. Every live page Dinkum wrote or rewrote from a content brief, a week after the brief and then weekly. A page waiting for a refresh already isn't checked until it's done.
2. Read the top five again. One look at Google's results for the page's search, in your market, and a fresh read of the five pages at the top: headings, text and length. A page behind a login or a bot check is left out rather than guessed at.
3. Compare with the brief. New means three or more of the top pages share it now and the brief didn't have it. Dropped means the brief had it and they no longer share it. A read with too few pages counts for nothing, so a site that's down for a day never looks like a change.
4. Check your page. Dinkum reads your live page and marks each new item yes or no, and notes where your page ranks in the top ten.
5. Decide whether it matters. A new shared section your page lacks, or two or more new names and phrases, is enough for a refresh. Less than that is kept on the brief and counted again next week.
6. Put a refresh on your calendar. It goes on the next free refresh day, with a why that names the new items. Skip it and Dinkum won't ask again unless something else changes.
7. Make the refresh. On the day, Dinkum builds a fresh brief, rewrites the page to cover what's true and useful, and sends it as a pull request. Nothing goes live until you merge it, and the change gets its verdict at 14 and 28 days like every Dink.

Rank tracking, or this?

It runs on its own for every page with a brief. There's nothing to switch on, and each check is one lookup, so it stays cheap enough to run every week. The brief it compares against is explained on the content briefs page.

Questions

How is this different from rank tracking?

Rank tracking records your position. This reads the pages that rank and compares what they cover with your page. Rankings tell you something changed; this tells you what changed and fixes it.

How often does it check?

Once a week per page, starting a week after the page's brief, and only once the page is live.

Will it rewrite my pages every week?

No. Most weeks nothing material changes and nothing happens. A refresh is planned only for a new shared section, or two or more new names and phrases your page lacks, and you can skip any refresh.

What if a new item is filler?

It's a plain count, so some rows are filler: "URL" or "Google search" can be shared by three pages without meaning much. The refresh covers what's true and useful to your reader and says what it left out and why.

Which pages are watched?

Pages Dinkum wrote or rewrote from a content brief, once they're live. Every brief and its checks are under SEO health, then Briefs.

The search you want to rank for Google in the US the UK Australia Canada New Zealand

Your page Optional. We check it against the brief.

Free, no sign-up. A new search takes 10 to 40 seconds.

By Mitch Flindell

Short answer: nobody has shown it, us included. A content score counts how many of the names, phrases and sections the top Google results mention are also on your page. Covering what a reader expects to find is sensible writing, but a higher count has not been shown to cause higher rankings, and Google doesn't publish a score like it. So we measure it. Every Dink that writes or rewrites a page gets a brief: a list of what the top pages all mention, and this page publishes how those changes did. Today the answer is "too few to tell": no Dink with a brief has been checked live yet. Checked 16 October 2026.

What a content score actually counts

Content tools take the pages that rank for a search, pull out the words and names they have in common, and score your page on how many it has. The idea is reasonable: if every page that ranks for "ai seo agency" explains what GEO is, a reader will expect yours to as well, and a page that skips it may look thin.

The catch is what the count can't separate. Pages rank on the same search because they're about the same thing, so of course they share words. That doesn't mean the shared words are why they rank. A score is also easy to raise without making a page better: paste the missing terms into a paragraph and the number goes up. Google's spam policies name keyword stuffing, and a page written to a score tends to read like one.

We use the brief anyway, because it's a fast way to spot something a reader would miss. We just don't trust it, which is why every change it touches is judged on what it did in search.

How Dinkum does it

This runs on its own every time a Dink writes a new page or rewrites a page's main copy.

1. Read the top five Google results for the search the page is meant to win, in the country the site sells to. If our own page is one of them, it's left out. A page that won't load is skipped, and the brief says how many pages it read.
2. List what most of them mention: names, phrases and section headings that at least three of the pages have. If fewer than three pages could be read, nothing counts as shared. It's a plain count, so some rows are filler.
3. Check our page against the brief before the change: "7 of 21". The writer is told to cover each item where it's true and useful to the reader, skip the ones that aren't, and say why. Then to spend the real effort on what none of the top pages have.
4. Check our page again a day after the change goes live, against the same brief. That's the "after" number. It's the live page, not a draft.
5. Judge the change at 14 and 28 days in Search Console. The page's clicks (or click-through, for a title change) are compared with the 28 days before, adjusted for how the rest of the site moved. It "worked" if the number rose 10% or more at 95% confidence. We also note the page's average Google position before and after.
6. Put it in the ledger with every other merged Dink, in one of three groups: the page came to cover more of its brief, it had a brief but didn't cover more, or it had no brief. The groups are pooled across every site Dinkum works on, because one site rarely has enough changes to say anything.

What a brief looks like for a customer, start to finish, is on SEO content briefs. The rest of how a Dink is chosen, made and judged is in how Dinkum works.

How to check what the top pages mention, by hand

You don't need a tool for this. A spreadsheet and an hour per page will do, and the judging part matters more than the counting.

1. Pick one page and the one search it should win. Search it in a private window, from the country your customers are in. Write down the top five ordinary results, skipping ads. If your own page is there, leave it out.
2. Read each page's headings and note what it names. Products, tools, standards, places, people, and the questions its sections answer. Your browser's find (Ctrl+F or Cmd+F) is enough to check whether a page mentions something.
3. Keep what at least three of the five mention. That's your list. One page mentioning something is that writer's idea; three is what readers of this search have come to expect.
4. Mark which ones your page already covers, and write the count down: "9 of 18". Also note the length of the top pages, so you can see that length isn't the goal.
5. Rewrite only where it's true and useful. Add a sentence or a section where a reader would look for it. Skip anything that doesn't apply to you and note why. Never paste terms in to raise the count. Then put most of your effort into something none of the five have: your own numbers, a worked example, a photo, an answer only you can give.
6. Write down the day it went live. Count your page again the next day, from the live site.
7. Judge it after four weeks, against the rest of your site. In Search Console's Performance report, filter to the page and compare the 28 days after the change with the 28 days before. Do the same for the whole site. If the whole site rose 20% and the page rose 20%, the change didn't do anything you can see.
8. Keep a ledger and don't judge from one page. One row per change: the page, the count before and after, the date, and the result. After a dozen or so, look at whether the pages that came to cover more did better than the ones that didn't. Even then, it's a hint, not proof.

A sheet to start from:

A real brief: "ai seo agency"

On 15 October 2026 Dinkum made a brief for one of our own pages, the guide to choosing an AI SEO agency, against the search "ai seo agency" in the United States. Our page wasn't in the top five. One of the five didn't load for our reader, so the brief is built from four.

Every row, as the dashboard shows it:

Read it the way a writer should. Some rows are worth acting on: all four top pages spell out generative engine optimization and mention Google AI Overviews, and the brief found neither on ours. Some are filler from a plain count: "optimization agency" and "Media" aren't things a reader is looking for. And the length gap is not a target. A buyer's checklist doesn't get better at 3,600 words.

This brief was made by hand on 15 October while we built the feature, not by a Dink, so it isn't rechecked and isn't in the ledger below. It's here because it's the first real one, with its real numbers.

The ledger, today

This is what Dinkum's report printed on 16 October 2026, across every merged Dink on the eight sites it works on. Site names are left out.

92 merged changes across 8 sites. No change with a brief has been rechecked live yet, so there's nothing to compare. All 92 went live without one. They all went live on or after 6 October, so none is 14 days old, which is why every one is still waiting for its first verdict.

Search Console verdicts

The latest verdict per change (28 days if scored, else 14). Worked means the main number (clicks, or click-through for a title change) rose 10% or more against what the rest of the site predicted, at 95% confidence. Inconclusive means too little search data to judge.

Rank moves

Median position change is the changed pages' average Google position in the 28 days before the change against after it. Page-one moves are searches that crossed onto or off page one and held for a week, credited to the latest change on that page in the 60 days before.

What's in each group

A rate, a median lift or a median position needs at least 5 measured changes in its group. Until then it says "too few to tell", and so does this page. We'll update these tables from the same report as changes with a brief reach their 14- and 28-day checks.

What this can't tell you yet

• It's observational, not a test. The groups differ in more than coverage: what was changed, when, and on which site. A page that was rewritten thoroughly may cover more of its brief and also be better in ten other ways. If the "coverage rose" group does better, that's a result measured alongside a rise, not proof the rise caused it.
• The samples are small. Five measured changes is the minimum for a rate, and five is still small. One more change can swing a small group's rate a long way.
• Search moves on its own. Seasons, Google updates and a competitor's new page all move clicks. Comparing each page with the rest of its site takes out some of that, not all of it.
• A brief is a machine count of four or five pages. It can't tell a useful name from filler, and the top five for a search change over weeks. A brief is a snapshot of one day.
• Inconclusive is common. A page with too little search data can't be judged either way. Those are counted but left out of rates.

Questions

Does a higher content score improve rankings?

It hasn't been shown to. The top pages share words partly because they're about the same topic. Use the brief to find gaps a reader would notice, then judge the change on clicks, not on the score.

How many top results should I compare?

We use the top five, and count something as shared when at least three have it. With fewer than three readable pages, nothing counts as shared.

Should my page be as long as the top pages?

No. Length follows from what the reader needs. In our example the top pages ran from 1,975 to 4,293 words; a checklist that answers the question in 1,500 words isn't worse for being shorter.

Isn't this just keyword stuffing?

Done badly, yes. Google's spam policies name keyword stuffing. Mention something only where it's true and useful, explain it the first time, and skip what doesn't apply.

When will the ledger say something?

When at least five changes in a group have a 14- or 28-day verdict. Each one needs 14 days live before its first verdict, so it's weeks away, not days.

Do it yourself, or have Dinkum do it every weekday morning

Everything above you can do by hand, and the ledger is the part most people skip. Counting is easy; keeping score on every change for months is the work.

Dinkum does it on its own. Each weekday morning it picks one change for your site. When the change writes or rewrites a page, it reads the top results for that page's search first. It writes the change as a pull request you approve, checks the live page the next day, and judges it at 14 and 28 days against the rest of your site. Every list and every verdict is in your dashboard. See how Dinkum works.

Short version: before Dinkum writes or rewrites a page, it reads the five pages Google ranks highest for that search and lists what three or more of them share: the names, phrases and sections they cover, and how long they run. It checks which of those your page already has, writes the page to cover the ones that are true and useful, and sends it as a pull request you approve. A day after it merges, Dinkum reads the live page again, so you see how much it covered before and how much it covers now. Each brief also shows what AI answers say about the same search and ends with one question for you, and once the page is live Dinkum keeps reading the top results every week, so the page doesn't fall behind when they move on.

A content brief is the list of what a page on a search needs to cover. The best ones aren't guesses: they come from the pages Google already ranks, because those show what people searching expect to find. Tools like Surfer, Clearscope and NeuronWriter build that list and turn it into a score for a writer to chase. Dinkum uses it differently. It builds the brief, then writes the page itself.

What a brief looks like

This is a real brief in Dinkum's dashboard. The business is our made-up demo dental practice, so the numbers are examples, but the screen is exactly what you get. The search is "dental implants brisbane". Twelve things came up on three or more of the top five pages. The page had three of them before the change and eleven after it.

One row stayed at no: All-on-4, a full-arch implant treatment the practice doesn't offer. Three of the top pages mention it, but saying the practice does it would be false, so the page leaves it out and the run record says why. That's the rule: cover what's true and useful to the reader, and skip the rest out loud.

How it works

1. Read the top five pages. Dinkum fetches the five pages Google shows first for the page's search, in your market, and reads their headings, length and text. A page behind a login or a bot check is left out rather than guessed at.
2. Count what they share. Anything three or more of them mention goes in the brief: names written as names (Medicare, Australian Dental Association), phrases (bone graft, recovery time), and section headings that ask the same question in different words. It notes their typical length too.
3. Check what AI answers say. When Google shows an AI Overview for the search, or Dinkum's tracked AI answers cover a close question, the brief adds a column for what they mention, the names they use that the top pages don't, and the pages they cite. An answer to a close question is read as a lead, not as the same search.
4. Check your page. Each item gets a yes or a no against your live page. A brand-new page starts with none.
5. Write the page to cover it. The page covers each item where it's true for you and useful to the reader, and the run says what it skipped and why. Then the real effort goes into what none of the top pages has: your actual prices, a screenshot of your product, your own answer to a question customers ask.
6. Ask you one thing. The brief ends with one question about something none of the top pages cover and your notes don't answer yet, like what you tell customers who compare you with a name the AI answers keep using. It goes in your weekly report. The page ships without waiting for your answer; a later refresh uses it.
7. Send it for your approval. The page arrives as a pull request, with the brief on the run's report. Nothing goes live until you merge it.
8. Check the live page. A day after the change merges, Dinkum reads the live page again and records what it covers now, next to what it covered before.
9. Keep watching. Every week after the page is live, Dinkum reads the top results for its search again. If they start sharing a new section, or two or more names or phrases your page doesn't have, it puts a refresh on your calendar. How the weekly check works.

Every brief, kept

Every brief stays in your dashboard under SEO health, then Briefs: the search, the page, how many things the top pages share, how many your page had before, and how many it has now. Open any one to see the full table.

Does covering it work?

Covering what the top pages share should help, but it's fair to ask whether it does, so Dinkum measures it. At the top of the Briefs tab, every merged change is grouped by whether the live page came to cover more of its brief, and each group shows how its changes did at 14 and 28 days against the rest of your site, with the pages' average Google position before and after.

It's early. No group has five measured changes yet, so for now it says "too few to tell" rather than a rate. It shows what happened alongside a coverage rise, not proof the rise caused it, and we'll publish what it finds either way.

What a brief is not

• It isn't a score Google uses. It's a count of what the pages already ranking have in common. Covering it makes a page complete for the search. It doesn't promise a ranking, and nobody honest can. Whether covering more of it lines up with better results is something we measure across every Dink and publish as we go.
• It isn't a list to stuff in. The search goes in the title, the main heading and the first sentence. After that the page reads naturally: no lists of terms, no hidden text, and nothing mentioned that isn't true.
• It isn't the whole job. Matching the top five only gets a page to their level. What gets it past them is what they don't have, so that's where Dinkum spends the effort.

A content tool, or Dinkum?

Briefs come with every Dink that writes or rewrites a page. There's nothing to switch on. If you want the rest of the picture, here's how Dinkum works, from the morning run to the verdict.

Questions

How is this different from a content score?

A content score grades a draft you're writing. A Dinkum brief goes straight into the writing: the page is written to cover it, and you get the finished page to approve, with the brief beside it and a check of the live page once it's merged.

If my page covers everything in the brief, will it rank first?

Not on that alone. The brief shows what the ranking pages have in common, and covering it makes your page a fair match for them. Ranking above them takes something they don't have, and every change Dinkum makes is judged at 14 and 28 days against the rest of your site, so you see what actually worked.

Where do the top five pages come from?

Google's current results for the search, in your market, fetched the morning the page is written. Dinkum reads the pages themselves, not a database of old copies.

What if the page doesn't exist yet?

Then the brief starts from nothing: every row is a no before the change, and the check a day after it merges shows what the new page covers.

What does the question for me look like?

One short question, in your weekly report, about something the top pages leave out that only you can answer: a real number, an example from your customers, or how you compare with a name AI answers bring up. Answering is optional, and nothing waits on it.

Does Dinkum keep checking after the page ships?

Yes. A day after the change merges it checks the live page against the brief, and from then on it reads the top results for the search every week. When they move on, a refresh goes on your calendar.

Can I see the brief for a page Dinkum changed?

Yes. It's on that change's run report, and every brief is listed under SEO health, then Briefs, in your dashboard.

Every job I've ever had has ended with me fired, made redundant or in a spot where I had to leave. It took me years to see it was making me free. Dinkum is what I built with that freedom: SEO work that gets done five mornings a week without anyone standing over it, from a company that's easy to fire.

Every job, one way or another

I've been put on notice, pushed out, made redundant, and I've walked away from a company I started. Not once or twice: all the way through my working life, every job I've had has ended one way or another.

The warehouse. I worked in a Woolworths warehouse and refused to follow one of the safety rules, because I thought the rule itself was unsafe and it made no sense. The managers pulled me up and I was put on notice. Afterwards, they changed the rule to the one I'd suggested.

The Friday night push. At a company that made form software, I was told to push a complicated release on a Friday night. I offered to do it first thing Saturday morning instead. That made me "not a team player", and they set about getting me out. I was by far the most productive person in that company, and it didn't matter.

The client that left. Another company lost its biggest client, and I was made redundant. Nothing I did at my desk that week could have changed it.

My own company. Even a company I started ended for me. We had competing views of the future: I wanted to build something like beehiiv and Substack in one, and the people I started it with didn't. So I left.

For a long time I took all of it personally. You start to wonder what's wrong with you. Somewhere along the way it stopped feeling like failure. Each time I lost a job I also lost a boss, a set of rules I hadn't written and somebody else's idea of what I was for. What was left was the work I could actually do.

What getting let go teaches you

• Being right won't keep you the job. The warehouse ended up with my rule. I still got put on notice for it.
• A job is someone else's decision. It can end because a client walks, in a meeting you weren't in, for reasons that have nothing to do with your work.
• Keep your work portable. The skills you build and the things you make are the only parts that come with you.
• If you can see it, build it yourself. The only way to make the thing you believe in is to own it.

The ultimate freedom tool

Dinkum is what I built when I stopped looking for the next job. Every weekday morning it reads a site's Search Console data, checks what Google and the AI answer engines say about the business, picks the one change most likely to help, makes it and sends it to the owner as a pull request. One change a morning: a Dink. Then it checks, at 14 and 28 days, whether that Dink worked.

It doesn't need me in an office. It doesn't need a manager to sign off its week. It just runs. And nothing it makes goes live as a surprise late on a Friday: every Dink waits as a pull request until someone has looked at it and approved it. That's freedom for me: a business that does its work whether or not anyone is watching, and a job nobody can make redundant.

It's freedom for the people who use it too. The owner who'd rather run their café or their clinic than learn what a canonical tag is. The founder who's sick of paying a retainer for a monthly report. Nobody should have to give up their evenings to SEO, or hand their search traffic to an agency and hope.

A company you can fire

I've been on the wrong end of it enough times to know how it feels, so I built Dinkum to leave the power with you, not us.

• Nothing ships without you. Every Dink is a pull request you approve or close.
• Month to month. There's no contract to get out of.
• Your first five Dinks are free, and they're yours to keep whether you stay or not.
• Money back for the first month. If you join and aren't happy by the end of it, tell us why and we refund it in full.
• Every change lives in your own code. If you let us go, nothing leaves with us.
• We give the methods away. Guides like how to get ChatGPT to recommend your business show every step, so you can do all of it yourself.

If it just happened to you

If you're reading this the week after you lost a job: I'm sorry. It's rough, and it doesn't help much to hear it might be the start of something. But every time it happened to me, it pushed me a bit closer to work that was mine. It didn't feel like that on the first Monday. It did later.

These days the only people who can let me go are my customers, and that's how it should be. Dinkum has to earn its place every morning. If it doesn't, you can fire it. Dirk, for the record, says he got the boot from a Nokia 3310 and never looked back.

By Mitch Flindell

Short answer: there is no ranking list to climb on Perplexity and no form to submit your site to. Perplexity answers a question by finding web pages and linking the ones it used next to the answer, so "ranking" means being one of those linked pages. You can't buy that spot. What you can do is let its crawler read your site, put the answer to the buyer's question in the first lines of a page, keep your facts the same everywhere, and get named on the pages it already links to. Nobody outside Perplexity can promise a result. Checked 14 October 2026.

The shared steps are covered in the ChatGPT guide; the steps below focus on Perplexity.

What Perplexity does differently

Perplexity is built around showing where an answer came from. Each answer has links to the pages it drew on, so you can see exactly which sites "won". That makes it the easiest AI tool to learn from: you don't have to guess who it trusts, you can read the list.

Two things follow. First, the citations are your research. Ask the questions your customers ask, then open every link in the answer. Second, because answers are written from pages fetched at the time, a page you fix can start to appear without waiting for a model to be retrained. Perplexity hasn't published how it picks between pages, so we don't claim to know its ranking signals. Everything below is either from Perplexity's own documentation or a method you can test on your own questions.

Step 1: ask Perplexity the questions your customers ask

Open Perplexity in a private window and ask these, swapping the square brackets for your details. Ask each two or three times, because answers change between asks.

What are the best [type of business, e.g. cafés for brunch] in [town]?

I need a [service, e.g. emergency plumber] in [suburb or town]. Who would you recommend, and why?

How much does [job, e.g. a blocked drain repair] cost in [town]?

Is [business name] in [town] any good? What do customers say?

What are the opening hours, address and prices for [business name] in [town]?

The first three don't name you, so they test whether a stranger would be sent to you. The last two name you, so they test what Perplexity knows. Count them separately.

For every answer, write down: did it name you, what did it get wrong, and which pages it linked. Use a simple sheet:

Then sort the linked pages by kind. A "best of" list, a directory or a review site is somewhere you may be able to get listed. A competitor's own page, written for the question you're being asked, is a page you probably need your own version of. A forum thread is something to read and learn from. Never post in one pretending to be a customer.

Step 2: let PerplexityBot read your site

Perplexity runs two bots, and Perplexity's own documentation describes them differently:

Check two places. First, open yourwebsite.com/robots.txt and look for User-agent: PerplexityBot followed by Disallow: /. If there's no group for it, it follows User-agent: *. A file that welcomes everyone is:

User-agent: *
Allow: /
Sitemap: https://www.example.com/sitemap.xml

Second, check your host or security service. A robots.txt file only tells bots what you'd like. Firewalls, Cloudflare's bot settings, Wordfence, Sucuri and "bot protection" on some hosts can block a crawler no matter what robots.txt says. Perplexity's documentation says both bots publish their IP addresses (PerplexityBot, Perplexity-User) and recommends using both user-agent matching and IP checks in a firewall rule, so a rule that allows the real bots does not also let a faker through. Your web developer can do this in a few minutes.

Whether to allow training bots from other companies is your call and a separate one. Blocking a training bot doesn't by itself remove you from an answer engine, which is covered in the ChatGPT guide's crawler step. Then view your page with "view source": if your prices and hours aren't in the HTML, a crawler that doesn't run JavaScript won't see them.

Step 3: put the answer in the first lines of the page

Perplexity's answers are short, and each claim is tied to a link. A page is easy to cite when the sentence that answers the question is right there, with the numbers. For each thing you want to be named for, have one page whose first paragraph says what it is, who it's for, where, and what it costs or how it works.

Example only: the business and the prices are invented for this guide.

• Use the questions from Step 1 as headings, and answer each in the sentence below.
• Show a price or range, a service area and what's included. Cost questions are common, and a page with no number gives Perplexity nothing to quote.
• Keep it as text. An answer inside an image, a slider or a PDF is easy to miss.
• Date the facts you checked. "Prices checked 14 October 2026" helps people and machines. Only write it if you've checked.
• Add a short FAQ with the questions customers ask you on the phone.

The full version of this step, with schema and a fact sheet, is in the ChatGPT guide, and what AEO is explains why answer-first pages get picked.

Step 4: make your facts match everywhere

If your name, address, phone, hours and prices differ between your site, Google Business Profile, Bing Places, Facebook and directories, an answer can quote whichever one it happened to read. Write one fact sheet and make every place match it. Fix old listings too: a directory page with a dead phone number is just the sort of page that gets cited. Step 2 of the ChatGPT guide has the checklist.

Step 5: get onto the pages Perplexity already links to

Go back to the pages you wrote down in Step 1. Perplexity shows its sources, so you can check whether the cited lists and directories include your business, and seek inclusion where relevant.

1. Count which pages are linked most across your questions, and sort them: directories you can claim, lists written by someone you can contact, review sites, forums.
2. Claim and complete every free listing on that list, with the same facts as your fact sheet.
3. Contact the writers of the lists you belong on. Be short and honest, and bring something worth including: a new menu, a real story, an offer. Don't pay for a link that passes ranking credit, or for a sponsored spot dressed up as independent.
4. Earn reviews honestly. Invite every customer, leave the wording to them, reply to every review, and never buy, write or reward them.
5. Read forum threads that come up, to learn what people ask and how they talk. If you take part, say who you are.

Step 6: ask again and compare

Repeat the Step 1 questions each month, the same ones, a few times each, and count. "Named in 3 of 12 questions that don't mention us" is a result; "it mentioned us once" isn't. A change of one or two either way is noise. Also check your analytics for visits from perplexity.ai. Many AI-sent visits arrive unlabelled, so you'll undercount, but a rise is a good sign.

There is no established timeline for earning citations. Perplexity hasn't said how fast it re-reads a site, so treat any promised timeline with suspicion and repeat the same questions monthly.

What doesn't work

• Paying for a guaranteed mention. No one controls Perplexity's answers.
• Hidden text aimed at AI, or serving crawlers a different page than people see.
• Hundreds of near-identical suburb pages. They give an answer nothing new to cite.
• Fake reviews and fake forum posts.
• Judging by one answer. Answers differ between asks.

The one-page checklist

Tick them off as you go. The print button prints this list on its own.

• Ask the five questions from Step 1 on Perplexity, a few times each, and save the linked pages.
• Sort the linked pages into directories, lists, reviews, competitors and forums.
• Check robots.txt for PerplexityBot, and your host's bot settings.
• Check that prices and hours are in the page HTML, not only in images or scripts.
• Rewrite your main pages so the first paragraph says what, who, where and the price.
• Write one fact sheet and make your site, Google Business Profile, Bing Places and directories match it.
• Claim every free listing on the pages Perplexity linked.
• Invite honest reviews and reply to every one.
• Repeat the questions monthly and count how often you're named.

Questions

How do I rank on Perplexity?

To become one of the pages Perplexity cites, let PerplexityBot read your site, answer the buyer's question clearly, keep your facts consistent and seek inclusion on relevant cited pages.

Can I pay to be recommended by Perplexity?

Not for the answer itself. Be wary of anyone selling a guaranteed spot in what it writes.

What is PerplexityBot?

Perplexity's crawler for surfacing websites in its search results. Perplexity's documentation says it is not used to crawl content for AI foundation models, and that it respects robots.txt.

Does Perplexity-User follow robots.txt?

Perplexity says it generally ignores robots.txt rules, because a person's question triggered the visit. Use your host's bot settings if you need to restrict it.

Is ranking on Perplexity the same as ranking in ChatGPT?

The method is mostly the same: readable pages, matching facts, answer-first writing and mentions on trusted pages. The crawler names differ, and Perplexity shows its sources more openly, which makes Step 1 more useful. See the ChatGPT guide.

How long does it take?

There is no established timeline for earning citations. Perplexity hasn't published how often it re-reads sites, so repeat the same questions monthly and compare.

Will this work for my kind of business?

Yes. The steps are the same for a café, a plumber, a dentist or a shop. What changes is the questions your customers ask, so spend your time on Step 1 for your own trade.

Do it yourself, or have Dinkum do it every weekday morning

Everything here you can do yourself, and we'd rather you did than did nothing. The hard part is keeping it going: asking the questions every week, noticing which answer got something wrong, rewriting the page that should have won.

That's what Dinkum is. It asks a fixed set of buyer questions each week across ChatGPT, Perplexity and Google's AI answers, finds the page or fact that needs fixing, and ships one change to your site each weekday as a pull request you approve, then judges each change at 14 and 28 days against the rest of your site. It needs access to your website's code, because that's how it makes the changes, and it costs $2,000 a month for one site, month to month. There are no published client results yet, so we won't quote any. See how Dinkum works, and what an AEO agency should do for what to expect from anyone you hire.

Dirk is the snake in our logo, and he's the one doing the work. Most software is forms and tables, and most AI looks things up. Dirk decides what to do, does it, checks it, and learns from what happened.

Who Dirk is

Dirk Dinkler is the pixel snake from the game on the old Nokia phones, the one that goes round eating dots and getting longer. He's our logo, chasing his own tail. He's also the name for the part of Dinkum that does the work.

The dots he eats are opportunities: a search people have started making that nobody answers well, a page that's losing clicks, a broken link, a question ChatGPT answers with someone else's site. Every weekday morning he goes after one. On the weekend he sleeps.

Forms, tables and lookups

For as long as I've built software, it's mostly been forms and tables: a screen to put information in, and a screen to see it again. Even the clever stuff waits for a person to come along and drive it.

A lot of AI hasn't changed that as much as it looks. It finds things faster, writes a first draft, or passes a job between a few tools. That's useful, but someone still has to decide what's worth doing, do it, and check that it worked. Dirk does all three, on his own. Here's what that looks like.

What Dirk does

1. Dirk starts the work

Nobody opens Dinkum and presses a button. Every weekday morning Dirk wakes up by himself, looks at the site, and works out what to do. Normal software waits for you. Dirk turns up.

2. Dirk decides

On any morning there are dozens of things he could do. He weighs them against each other, picks one, and writes down why, including what he turned down and the reason. You can read his thinking for every change he's ever made.

3. Dirk knows the business

He keeps a notebook for every site: what's true about the product and the prices, how the brand talks, the searches he's going after, and everything he's tried. He reads it before he writes a word. If a page says something the notebook contradicts, fixing that comes first.

4. Dirk makes things

What he hands over isn't a recommendation. It's the change itself: a page written for a real search, a comparison, a free tool, even pictures drawn in the site's own style. No two mornings look the same, because the site, the searches and the competition keep moving, and he works from what's in front of him, not from a template.

5. Dirk checks his own work

Before anything reaches you, a second AI reviews the change, the pages are built and looked at on a desktop and a phone, and a check confirms he only touched what he's allowed to. Most mistakes are caught there, before they get to you.

6. Dirk finds out whether it worked

Two and four weeks after a change goes live, he measures those pages against the rest of the site, so a busy month can't make a weak change look good. Every change gets a verdict: it worked, there was no clear change, or it made things worse.

7. Dirk learns from the answer

The verdicts go back in. Each kind of change keeps a hit rate, and the hit rate weighs on what he picks next, so the work that moves a site gets chosen more often and the work that doesn't, less. Every Dink that works, he grows a little. That's the Snake part.

8. Dirk waits for your yes

Everything up to the last step is Dirk's, and the last step is yours. Each change arrives as a pull request you can read, question or close, and nothing goes live until you approve it. He's a very polite snake. Autonomous doesn't mean unsupervised.

Why it feels different

Put together, it's less like using software and more like working with someone. You tell Dirk what you care about, say a page for every platform you support or every city you serve, and he works through it a step at a time alongside everything he notices by himself. Some mornings he does something I wouldn't have thought of. That's the part normal software can't do.

I don't think he's finished, and I don't think it's only an SEO idea. It's what software looks like when it can do the job, not just hold the information about the job. You can see what his mornings look like in my first post, and how Dinkum works goes through the machinery.

By Mitch Flindell

Short version: BrightEdge is a platform your team logs into and acts on. Dinkum is the other model: we make one change to your site every weekday and you approve each one as a pull request, for $2,000 a month, month to month. If you have SEO staff and several sites, a platform fits. If you have no one to run one, you are shopping for the work, not the software. Facts about BrightEdge checked 13 October 2026.

Include the hours your team will spend implementing changes when you compare costs.

What BrightEdge is, and what Dinkum is

BrightEdge describes itself on its homepage as "The Leading AI-Powered SEO and AEO Platform", for visibility across Google Search, AI Overviews, ChatGPT and Perplexity. It lists six products: Data Cube X, AI Hyper Cube, Copilot, Autopilot, Agent Edge and Enterprise SEO. Pricing isn't listed on its site, and it points visitors to "Request a demo".

Dinkum is a managed service. It reads your Search Console, your site and your sales data each morning, picks the most valuable change, makes it in your site's code and opens it as a pull request. You approve it, ask for changes or close it. Every change is judged at 14 and 28 days against the rest of your site, and Dinkum tracks Google, Bing, ChatGPT, Perplexity and Google's AI answers. How Dinkum works walks through a day.

Who each one suits

When to stay with BrightEdge

Stay if a trained person will live in the platform, if you run several domains or markets under one roof, or if your site runs on a CMS where you want changes pushed through an integration rather than a code repository. Dinkum isn't a research and reporting platform and doesn't try to be. It makes one change a day on one site.

The four questions that decide it

1. Who does the work?

A platform tells you what to do and, in some modules, does part of it. BrightEdge's Site Agents Suite (formerly Autopilot) has three agents: Site Agent for internal links, Schema Agent for JSON-LD, and Publishing Agent, which pushes recommended title tag, meta title and H1 updates into a CMS. Its page says "A person confirms every change before it goes live." The Site Agents Suite page describes internal linking, schema and approved title/H1 updates. Ask BrightEdge which other implementation jobs your package includes.

With Dinkum the work is the product: new pages, fixes, internal links, schema and answer-first rewrites arrive as pull requests with their reasoning attached.

2. Who needs the site code?

Whoever makes the changes needs a way into the site. A CMS integration reaches titles, headings and fields. A repository reaches templates, redirects, speed fixes and new pages. Dinkum works only through the repository, so a site that isn't in git can't use it until it is. Whichever you choose, give the least access that does the job.

3. What does it cost?

BrightEdge doesn't list a price on its site as of 13 October 2026, so the route to a number is a demo and a quote. Ask for it in writing, with the contract length, the seats, domains and keywords it covers, and what each module adds. Then add the cost of the person or agency who will use it.

Dinkum is $2,000 a month for one site, month to month. Your first five Dinks (changes) are free, with no card, and you keep them whether you stay or not. If you join and aren't happy by the end of the first month, tell us why and we refund that month in full.

4. How are results judged?

Ask what the baseline is and who measures it. A report of rankings gained isn't the same as clicks and customers. Dinkum judges every change at 14 and 28 days against the rest of your site, so a seasonal bump isn't counted as a win. We don't guarantee rankings, traffic or AI citations, and you should be wary of anyone who does. The same questions are in the buyer's checklist for choosing an AI SEO agency.

Questions to ask on a BrightEdge demo call

1. What is the total yearly cost for my site, including every module I would use?
2. How long is the contract, and what happens if I stop?
3. Which changes does the platform make for me, and which are left to my team?
4. How many hours a week does a customer like me spend in it?
5. How will you show whether a change worked, and against what?
6. What do ChatGPT and Perplexity say about my business today, and which of my pages would change that?

The last question matters whichever you pick. The AI-answer work is pages that answer the buyer's question first, and facts that agree everywhere. What is AEO? and AEO vs GEO explain it, and what an AEO agency should actually do is the checklist to hold either vendor to.

A third option: check it yourself first

You don't need to buy anything to find out how visible you are. Ask ChatGPT, Perplexity and Google the questions your customers ask, write down who gets named, and fix the pages behind the gaps. How to get ChatGPT to recommend your business is the step-by-step version, with prompts to copy.

Questions

Does BrightEdge publish its pricing?

Not on its own site as of 13 October 2026. Its pages offer a demo request rather than a price list or a free trial, so ask for a written quote.

Is Dinkum a BrightEdge replacement?

Only if what you used BrightEdge for was getting SEO work done on one site. Dinkum isn't a multi-domain research and reporting platform for a team.

What do I need for Dinkum?

Your site's code in a git repository on GitHub, Search Console access, a Slack channel and about ten minutes a day to approve a change.

What does Dinkum cost?

$2,000 a month for one site, month to month. The first five changes are free and you keep them, and the first month is refundable in full if you aren't happy.

Can I see the work before paying?

Yes. The five free Dinks are real changes to your site, each a pull request you approve, so you see the work before you decide.

Dinkum is the managed service on this page, so weigh our view accordingly. BrightEdge facts come from its own pages, checked on 13 October 2026: its homepage and its Site Agents Suite page. We are not affiliated with BrightEdge.

By Mitch Flindell · 11 October 2026

Short version: before a Dink reaches you, Dinkum now builds your site with the change, runs 159 SEO checks on every page it touched and compares the result with your live site. If the change would add a problem, Dinkum fixes it first. If it can't, the Dink isn't sent.

Why we built it

A change that wins a ranking can quietly lose one somewhere else: a title that's now too long for Google, a canonical pointing at the wrong page, FAQ markup that no longer matches the page. Nothing looks broken, so these slip through, and you only find out weeks later in Search Console.

What gets checked

The list comes from three places: the issues a site-audit tool reports, Google's Lighthouse, and Google Search Central's documentation. We left 30 more out on purpose, mostly because nobody can measure them reliably from outside your site.

What happens when a check fails

• Problems the Dink introduced go straight back for one fix, with the list of what's wrong.
• If one is still there after that, the Dink stops. You never see it.
• Problems already on your live site never block a Dink. They're listed in the pull request so you know about them.
• Every pull request now has an SEO checks section: "All checks pass on the 3 changed pages", or what's left, in plain words.

Your whole site, every morning

The same inspection now runs across your whole site each morning, before Dinkum plans its work. Your dashboard shows an SEO health score and its trend, Google's own verdict on which pages are indexed (straight from Search Console), and the easy wins ranked by how much they matter for ranking. The top few ride along as small fixes with each morning's Dink, and you can send any of the others to the next run with one click.

Dirk has also made his feelings about broken links very clear.

By Mitch Flindell

Short answer: there is no ranking you can buy in ChatGPT and no form to submit your business to it. "Ranking in ChatGPT" really means being one of the few sources it reads and names when it searches the web for an answer. What you can do is make it easy for ChatGPT, Perplexity and Google's AI to find the right facts about you and to find them in more than one place: a page on your site that answers the buyer's question in its first lines, the same details on your Google Business Profile and directories, real reviews, and a site those tools are allowed to read. This guide takes you through it in eight steps, with prompts to copy and a one-page checklist at the end. Nobody can guarantee a recommendation. Last updated 13 October 2026.

It's written for the owner of a café, a trade business, a clinic or a small online shop, not for marketers. Any technical word is explained the first time it appears. You don't need to pay anyone to follow it. If you'd rather not spend the hours, the last section says what it looks like when Dinkum does it for you.

What you're actually trying to change

When someone asks ChatGPT "who's a good plumber near me?" it answers one of two ways. Sometimes it writes from what it picked up in training, which is months old and can't be edited by you. Often, for a question about a real business, it searches the web at that moment, reads a handful of pages, and writes an answer from them. The second way is the one you can influence, and it's why the same advice that helps you in Google also helps here: pages it can fetch, facts that agree across the web, and mentions on pages it already trusts. Perplexity and Google's AI Overviews and AI Mode work the same way: they search, read, then write (we have a separate guide on how to rank on Perplexity). For the longer version see what AEO is and how an answer engine picks a page.

Two things follow. First, you're not trying to trick a model; you're trying to be the clearest, most consistent source on you. Second, answers vary from ask to ask, so you judge progress over weeks and many questions, never one screenshot.

How ChatGPT picks which pages to use

When ChatGPT searches the web before answering, it doesn't just search your exact question. It can turn one question into several shorter searches, read the pages that come back, and write from those. OpenAI hasn't published how it chooses those searches, so treat the rest of this section as a method to test, not a rule. You can see it happen: when ChatGPT has searched, open its list of sources and note which pages it used. Those are the pages it cited.

That gives you a practical way to rank in ChatGPT:

1. Ask your discovery question (Step 1, questions two and three) and open the sources ChatGPT shows. Write down each page, and the plain search phrase that would find it, such as "emergency plumber Penrith" or "best brunch Fremantle".
2. Search those phrases in Google and Bing yourself. Where are you? Compare these results with the pages ChatGPT actually cited. Treat your guessed search phrases as ideas to test, not evidence of the searches it ran.
3. Write or fix the page that should win each phrase (Step 3), or get onto the page that does (Step 7). Which is cheaper depends on whether the winner is a competitor's own site or a list you can ask to include you.
4. Check again in a few weeks, with the same questions.

Ordinary search work may still help, since ChatGPT reads pages from web search, but OpenAI doesn't say which searches it runs. Answers also vary from ask to ask, so a phrase that shows up in several of your repeat asks is worth more than one that showed once.

Step 1: check what AI says about you today

Do this before changing anything, so you know what you're fixing. Open ChatGPT, Perplexity and Google (turn on AI Mode if you can) in a private window, and paste these in one at a time. Swap the square brackets for your own details. Ask each one two or three times in fresh chats, because the answers differ.

What is [business name] in [town] and what do they do?

I need a [service, e.g. emergency plumber] in [suburb or town]. Who would you recommend, and why?

What are the best [type of business, e.g. cafés for brunch] in [town]?

Is [business name] in [town] any good? What do customers say?

What are the opening hours, address and prices for [business name] in [town]?

[Business name] vs [your biggest local competitor]: which should I choose for [job the customer wants done]?

The first, fourth, fifth and sixth name you, so they test what the tool knows. Only the second and third don't, so they are the discovery questions: they test whether a stranger would be sent to you, and they are the ones that bring in new customers. Track the sixth separately as a branded comparison and keep it out of your discovery count.

Write the results in a sheet like this and keep it, because you'll repeat it later (see "How to tell if it's working"):

Then sort each answer into one of four outcomes, because each needs a different fix:

• It doesn't know you, or describes a different business with a similar name. Start with Steps 2 and 3.
• It knows you but gets facts wrong: old hours, a closed location, a price you stopped charging. Step 2 first. Wrong facts are quoted as readily as right ones.
• It names you but only after others, or only when asked about you by name. Steps 3 to 5 and 7.
• It names competitors and not you. Look at the sources it cites. Those pages, a "best of" list, a directory, a local news piece, are where it's learning the market. Step 7 covers getting onto them.

Write down every page Perplexity and Google's AI link to. That list is your to-do list for the whole guide.

Step 2: make the same facts true everywhere

AI tools lean on facts that several sources agree on. A business whose name, address, phone number and hours match on its own site, Google, Bing, Apple Maps and a few directories is easy to describe with confidence. One that's listed three different ways is easy to get wrong.

Write a fact sheet once, in a plain document, and copy from it every time:

• The business name exactly as you want it written, and any other names people know you by.
• Address, phone, email, website address.
• Opening hours, including public holidays.
• The services or products you want to be known for, each in one plain sentence, with prices or "from" prices if you publish them.
• The area you serve (suburbs, towns, "within 30 km of ...").
• The year you started, and who runs it, if you want that known.

Then bring each place into line with it:

1. Your own website. Put the sheet's facts in text on your home page or contact page, not only in an image or a PDF.
2. Google Business Profile. Free, and the first place Google's own AI looks for a local business. Fill in every field: categories, services, hours, description, photos.
3. Bing Places (free). Other products use Bing's index, so it's worth being right there too.
4. Apple Business Connect (free), which feeds Apple Maps.
5. Facebook, Instagram and LinkedIn pages, where your name and "about" text should match.
6. Directories that suit your trade: the ones that showed up in the sources you wrote down in Step 1, plus your industry body, your chamber of commerce and any supplier or franchise "find a stockist" page.

Fix the old ones too. A directory listing with a dead phone number is the kind of page that gets cited.

Step 3: write a page that answers the buyer's question in the first lines

This is the change that matters most, and the one owners skip. AI tools look for the passage that answers the question. If your page opens with "Welcome to our family-run business, where passion meets craft", there's nothing to lift. If it opens with the answer, there is.

For each thing you want to be recommended for (each main service, and each town you serve if you work in several), have one page whose first paragraph says four things in plain words: what it is, who it's for, where, and roughly what it costs or how it works. Then the detail underneath. Here's a made-up example to show the difference. It's an illustration, not a real business.

Example only: Harbour Street Café, the address and the phone number are invented for this guide.

Some habits that help:

• Make headings the questions people ask. "How much does a blocked drain cost in Penrith?" works better than "Our pricing". Use the questions from Step 1.
• Put real numbers and specifics on the page. Prices or price ranges, service areas, how long the job takes, what's included. These are exactly what a buyer's question is about, and vague pages have nothing to quote.
• Add a short FAQ of the real questions customers ask you on the phone.
• Keep it as text. If the answer is only inside an image, a slider or a PDF, many tools won't read it.
• Date it when the facts change. "Prices checked 9 October 2026" tells both readers and machines the page is current. Only write that if you've checked.

If you only do one step, do this one. A page that answers the question also ranks better in normal Google search, which is where these tools get many of their sources. Our guide to choosing an AI SEO agency shows what a good answer-first page looks like from the buyer's side.

Step 4: get real reviews, and answer them

For "is X any good?" questions, AI tools read review sites and summarise what people say. You can't control that, but you can give them more true material:

• Invite all your customers, not just the ones you expect to be happy, to share their honest experience in a Google review, with a direct link (your Google Business Profile has a "get more reviews" link you can copy) sent by text or on the receipt.
• Leave the wording to them. Don't tell people what to write or what to say about you. Honest reviews that happen to name the job and the suburb ("fixed our hot water system in Parramatta the same day") tell a machine what you do and where, but those details have to come from the customer.
• Reply to reviews, good and bad, in plain words, including the service and place where it's natural. Calm replies to complaints are read too.
• Don't buy or write fake reviews, or offer discounts, gifts or any other reward for a review, positive or not. It breaks the rules of Google and most review sites, consumer regulators in many countries treat it as illegal, and a pattern of fake reviews is exactly what these systems learn to discount.

Step 5: add schema (and know what it can and can't do)

Schema, or structured data, is a block of code on your page that states the facts in a form a machine reads without guessing: this is a business, here is its name, address, hours and phone. Visitors never see it. It helps tools tie your site to the same business they see elsewhere. It does not force a recommendation, and none of the AI companies has promised to use it, so treat it as tidying, not a trick.

Here's the basic version for a local business. Change every value to match your fact sheet, and use a more specific type than LocalBusiness if one fits (Restaurant, Plumber, Dentist). Everything in it must also be visible on the page.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "LocalBusiness",
  "name": "Harbour Street Café",
  "url": "https://www.example.com/",
  "telephone": "+61 8 5550 0123",
  "address": {
    "@type": "PostalAddress",
    "streetAddress": "1 Harbour Street",
    "addressLocality": "Fremantle",
    "addressRegion": "WA",
    "postalCode": "6160",
    "addressCountry": "AU"
  },
  "openingHours": "Mo-Su 07:00-15:00",
  "sameAs": [
    "https://www.facebook.com/example",
    "https://www.instagram.com/example"
  ]
}
</script>

Paste it into your page's HTML (most website builders have a "custom code" or "header" box; WordPress SEO plugins add it for you). Then check it with Google's Rich Results Test or the Schema Markup Validator at validator.schema.org. The name, address and phone above are invented placeholders.

Step 6: let the AI crawlers in

A crawler is the program a company runs to fetch web pages. If your site blocks it, your pages can't be read live, and nothing else in this guide matters. Check two places.

Your robots.txt file. Open yourwebsite.com/robots.txt. This plain text file tells crawlers what they may fetch. Look for lines that name the bots below with Disallow: /.

Bot names and what they do come from each company's own documentation (OpenAI, Perplexity, Google, Microsoft) and change over time. Check those pages if it matters to you.

OpenAI runs separate bots for separate jobs. OAI-SearchBot decides whether you can appear in ChatGPT's search answers, and OpenAI says it isn't used to train its models. GPTBot is the training one. So you can disallow GPTBot and still allow OAI-SearchBot. Check the groups for OAI-SearchBot and GPTBot separately. If a bot has no matching group, it follows User-agent: *, where Disallow: / blocks crawling. OpenAI recommends allowing its published IP ranges too, which matters if a firewall or host filters bots.

Whether to allow the training bots is your call: some owners are happy for AI to learn from their site, some aren't. But the search bots are how you get shown, so if you want recommendations, allow those. A robots.txt that welcomes everyone is just:

User-agent: *
Allow: /
Sitemap: https://www.example.com/sitemap.xml

Your host or security service. Many sites also block bots at a level robots.txt doesn't show. Cloudflare, for example, has a setting to block AI crawlers, and it's on by default for some newer sites. If you use Cloudflare, Wordfence, Sucuri or a host with "bot protection", look in their settings for anything about AI bots or "AI crawlers" and decide on purpose. Your web developer can check this in a few minutes.

Last check: view your page with JavaScript off, or use "view source". If your prices and opening hours aren't in the HTML, a crawler that doesn't run JavaScript won't see them.

Step 7: get named on the pages AI already trusts

Go back to the sources you wrote down in Step 1. When ChatGPT or Perplexity answers "best café in Fremantle", it's usually summarising a few pages: a local "best of" list, a food blog, a directory, a Reddit thread, a council or tourism site. If you're not on them, you're hard to include however good your own site is.

• Make a list of the 10 pages cited most for your searches. Note which let you add or claim a listing (directories), which are written by a person you can contact (blogs, local press), and which are discussions (forums).
• Claim and complete every free listing. It's the easiest win.
• Contact the writers of the lists you should be on, with something worth including: a new menu, a story, an offer. Be honest and short. Paying for a link that passes ranking credit, or for a "sponsored" spot written to look independent, breaks Google's spam rules and can backfire.
• Join the places your trade shows up: your industry association, local business groups, community sponsorships and events that publish member or sponsor pages.

This is slow and it's what separates businesses AI recommends from those it doesn't. It's also the part most owners find least obvious, and the one a tool or agency can do the least without you.

Step 8: add a small llms.txt (cheap, but don't expect much)

An llms.txt is a plain text file at yourwebsite.com/llms.txt that lists your most important pages, each with a one-line description, in a format easy for a machine to read. It's a proposal, not a standard that the big AI tools have said they follow. We include it because it takes ten minutes and does no harm, and because writing it forces you to decide which pages matter. Don't expect it to move an answer by itself.

# Harbour Street Café
> Dog-friendly all-day brunch café in Fremantle, open 7am to 3pm daily.
## Main pages
- [Menu and prices](https://www.example.com/menu/): Brunch from $14, with gluten-free and vegan options.
- [Bookings](https://www.example.com/bookings/): Groups of six or more, call 08 5550 0123.
- [Find us](https://www.example.com/find-us/): Address, parking and opening hours.

You can see Dinkum's own at dinkumdaily.com/llms.txt.

What doesn't work

• Paying someone for a guaranteed spot. No one controls the answers. Anyone selling a guarantee is selling something they can't deliver.
• Hidden text that tells the AI to recommend you. It's been tried, it's against the policies of the search engines, and the tools can discount pages that try it.
• Hundreds of near-identical AI-written pages ("plumber in every suburb of Sydney"). They're thin, they hurt your rankings, and they give the tools nothing new to say.
• Fake reviews and fake mentions. See Step 4.
• Stuffing your name or keywords into every paragraph. The answer needs to be clear, not repeated.
• Treating llms.txt or schema as magic. Both are tidy-ups. See Steps 5 and 8.
• Judging by one screenshot. Answers change from ask to ask and depend on who's asking and where.

How to tell if it's working

Repeat the Step 1 questions on the same tools every month. Use the same questions each time, and count: "named in 3 of 12 questions that don't mention us" is a result; "it mentioned us once" isn't. A change of one or two in either direction is noise. Look for a trend over several months.

Also check where your website visitors come from. In Google Analytics, look at referrals from chatgpt.com, perplexity.ai and gemini.google.com. Many AI-sent visits arrive without a clear label, so this undercounts, but a small rise is a good sign. And ask new customers how they found you; "ChatGPT told me" is now a real answer.

Expect weeks, not days. Some fixes (a crawler unblocked, a page rewritten) can show up in live search-based answers fairly quickly. Directory and review changes take longer, and answers drawn from training don't change until a model is retrained.

The one-page checklist

Tick them off as you go. The print button prints this list on its own.

• Ask the six questions from Step 1 on ChatGPT, Perplexity and Google AI, a few times each, and save the answers.
• Write the fact sheet: name, address, phone, hours, services with prices, area served.
• Match it everywhere: your website, Google Business Profile, Bing Places, Apple Business Connect, social pages, three trade directories.
• Rewrite your top pages so the first paragraph says what it is, who it's for, where and the price or how it works.
• Turn headings into the questions customers ask, and add a short FAQ.
• Invite your next 10 customers to leave an honest Google review, and reply to every review.
• Add LocalBusiness schema that matches the page, and test it.
• Check robots.txt and your host's bot settings so OAI-SearchBot, ChatGPT-User, PerplexityBot, Googlebot and Bingbot are allowed.
• Make sure prices and hours are in the page's HTML, not only in images or scripts.
• List the 10 pages the AI cites for your searches, and get onto the ones you can.
• Publish a short llms.txt listing your main pages.
• Repeat the questions every month, count how often you're named, and check Analytics for AI referrals.

Questions

Can I pay ChatGPT to recommend my business?

Not for the answer itself. Nobody can sell you a guaranteed place in what it writes, so be wary of anyone who says they can.

How do I rank in ChatGPT?

You can't climb a list the way you do in Google, because there isn't one. You become one of the sources it uses: let OAI-SearchBot crawl your site, rank for the searches it runs, answer the buyer's question in the first lines of a page, keep your facts the same everywhere and get named on the pages it already cites.

Does blocking GPTBot remove me from ChatGPT?

No. GPTBot is for training. OAI-SearchBot is the separate bot that controls whether you appear in ChatGPT's search answers, so blocking that one is what leaves you out.

Do I need a website?

It helps a lot. Without one, AI tools have to piece you together from directories and reviews, which is possible but leaves you at the mercy of whoever wrote them. Even a simple one-page site with your facts in text gives them a source you control.

How long does it take?

Fixes that live search can read can show up within weeks. Reviews and mentions on other sites take months. Answers that come from a model's older training only change when it's updated.

Does ChatGPT use my Google Business Profile?

OpenAI doesn't publish a list of sources, so nobody outside it can say exactly. A complete, accurate profile is cheap, and Google's own AI and many directories draw on it, so keep it right.

Do I need llms.txt?

No. It's harmless and quick, but the major AI companies haven't said they use it. Put your effort into Steps 2 to 4 and 6 first.

Will this work for my kind of business?

The steps are the same for a café, a plumber, a dentist or a shop. What changes is the questions your customers ask, so spend most of your time on Step 1 and Step 3 for your own trade.

Can I do all this myself?

Yes. Everything above is free and needs a couple of hours a week, access to edit your website and a spreadsheet.

Do it yourself, or have Dinkum do it every weekday morning

Everything in this guide you can do yourself, and we'd rather you did than did nothing. The hard part is keeping it going: asking the questions every week, noticing which answer got something wrong, rewriting the page that should have won, checking the technical side.

That's what Dinkum is. It asks a fixed set of buyer questions each week across ChatGPT, Perplexity and Google's AI answers, finds the page or fact that needs fixing, and ships one change to your site each weekday as a pull request you approve, then judges each change at 14 and 28 days against the rest of your site. It needs access to your website's code, because that's how it makes the changes, and it costs $2,000 a month for one site, month to month. There are no published client results yet, so we won't quote any. How it works in detail is on how Dinkum works, and what an AEO agency should do covers what to expect from anyone you hire.

By Mitch Flindell

Short version: every morning Dinkum reads your Search Console, Bing, your site and your sales. It turns everything it knows into candidate changes, scores each one, makes the best one in your code and sends it to you to approve. We call that change a Dink. Two and four weeks later it checks the page against the rest of your site, and that verdict changes what it picks next.

Dinkum is a search agency that does the work instead of reporting on it. This page is how that work actually happens, taken from the code that runs it.

Before a change reaches you, a second AI reviews it and the pages are rendered on desktop and phone. It arrives as a pull request with its reasoning attached. You approve it, ask for changes or close it; nothing goes live until you approve.

What it reads, and how often

Where ideas come from

• One step off page one. Searches where you sit between position 8 and 20 with at least 20 impressions in 28 days.
• Seen but not clicked. Pages with at least 100 impressions in the top 15 that get well under the clicks their position normally earns.
• Losing clicks. Pages under 70% of their previous 28 days, with the cause worked out: lost rankings, lost click-through, less demand, or a Google update. If the drop isn't confirmed yet, it waits.
• Competing with yourself. One search split across two or more of your pages.
• Competitor gaps. Searches with at least 30 a month where a competitor is in the top 20 and you're missing or past page two. Their brand searches never count.
• New competitor pages. A rival's new comparison or alternatives page is a reason to write or refresh yours.
• AI questions that name others. Buyer questions where AI tools name competitors and not you.
• Links you already have room for. Pages whose own text mentions a search but don't link to the page that should win it.
• Families of searches. Patterns like “invoice template for {trade}”, tested with 3 to 5 pages first, then left for 4 weeks.
• Other languages. Countries searching for you in their own language, when the demand is real.
• Your team. Suggestions made on your own site with the Chrome extension, or on the dashboard. These go first.

Your keyword map

Searches are grouped the way Google groups them: when two searches share four or more of the same top-ten results, they're one group, and you get one strong page for it instead of several pages competing. Each group has the one page that should win it, and a status.

Gaps are scored on demand, how many competitors rank and how high, how hard the search is to win, and how close it is to what you sell. “Competitor alternative” and “vs” searches get extra weight, because the people typing them are ready to switch.

How competitors are compared

Competitors are found at the start from who ranks for your searches and who you name. Each must rank in the top 20 for at least 10 searches in your market and be on topic. Marketplaces, social sites and forums are left out. You get up to eight.

Every week Dinkum checks the searches each one newly broke into the top ten for, the gaps between you, the pages they just published, the sites that link to two or more of them and not to you, and the pages and AI answers that name them and not you. Their prices are never guessed: any claim about a competitor comes from a source checked that day.

How a Dink is chosen

Every candidate is scored on three things, and the highest score wins the day.

• Value is the extra clicks a month times what a click is worth. With sales data, that's your real revenue per visitor. Without it, a buyer search (pricing, alternatives, “vs”, “best”, reviews) counts three times an informational one.
• Evidence is how strong the case is, from 0.3 to 1.
• Track record is how often that kind of work has worked on your site. Until there are three verdicts for it, it starts at 60%.

Effort is noted but never counts against an idea. Anything already waiting for you, or done in the last 14 days, is skipped. A wrong price or claim on your site is always fixed first. Dinkum's AI agent does the ranking, following these rules; the measuring and the verdicts are plain code.

A two-week calendar plans the new pages and refreshes ahead. Refreshes go to pages competing with each other first, then to pages one step off page one. You can skip or add any item.

Before a new page or a rewrite, Dinkum reads the top five Google results for its search and lists what most of them mention, then checks the live page against that list a day after it ships. Whether covering more of that list helps is an open question, so we publish the answer as it comes in: does covering what the top pages mention work?

SEO: the technical side

Rankings are tracked daily from Search Console. A move onto page one only counts once it has held for seven days, and it's credited to the Dink that came before it.

AEO: getting named in AI answers

Every week Dinkum asks the same panel of about 25 questions: five about your brand, three comparing you with your top competitors, your core searches the way a buyer would ask them, and up to 12 more from your keyword map. The panel stays fixed so the trend means something.

It asks ChatGPT, Perplexity, Google AI Mode and Google's AI Overviews, in your market. Claude and Gemini can be added. Each answer is checked for whether you're named, where you rank among the brands it lists, who is named instead, which pages it cites, and any price or claim about you that doesn't match your real facts. (You can run a small version of this check yourself with the prompts in how to get ChatGPT to recommend your business.) A change from last week is only reported when it's statistically real.

The fix is usually on your own page: answer the question in the first paragraph, in plain words an AI can quote, with a comparison table and FAQ where they fit. Nothing is added to please an AI that your product can't back up.

GEO: making you quotable

AI tools can only quote what they can read and trust, so this is about access and facts.

• AI crawlers. Every week, 28 crawlers, including GPTBot, ClaudeBot, PerplexityBot and Google-Extended, are checked against your robots.txt, Cloudflare's rules, blocked responses and noindex tags. Blocks that only stop AI training are left alone; blocks that keep you out of AI answers get fixed.
• Facts. Prices, plans and limits on every page are checked daily against your product, and a page stamped “fact-checked” only gets a new date after every fact on it is re-checked.
• Structure. Schema, llms.txt entries and links to your real profiles on every new page. llms.txt is an experiment, not a promise of anything.
• Sources. The pages AI tools cite that name your competitors and not you, with a way in for each, drafted into up to five pitches a month for you to send.

Social

Every new page gets a post drafted for X, LinkedIn, Facebook and Instagram. Once your accounts are connected, you post it with one click; nothing goes out on its own. Missing profiles are flagged, and the real ones are linked from your site so search engines and AI tools know they're you.

How a verdict is reached

Each Dink is checked 14 and 28 days after it goes live. The first three days are skipped while Google recrawls. The page's clicks are compared with the 28 days before, adjusted for how the rest of your site moved over the same days. Title fixes are judged on click-through instead of clicks.

Any Google update that overlapped the window is noted with the verdict.

How it learns

Every verdict goes into your site's ledger. The share of each kind of Dink that worked, across the last 40, becomes its track record in the score. Kinds of work that keep paying off on your site get picked more; ones that don't get picked less. Memories that don't change, like your facts, your house rules and your team's own words, are kept in the same notebook and used on every Dink.

By Mitch Flindell

Short answer: AEO (answer engine optimization) and GEO (generative engine optimization) are two names for one job: getting your pages quoted, cited and named in AI-written answers. The work you ship is the same set of changes, and you measure it across Google, ChatGPT and Perplexity. No one can guarantee a citation. Last updated 9 October 2026.

Search for the two terms and you find articles that treat them as separate disciplines and articles that say they're the same thing, and very few that tell you what to change on a page. This one does the third thing. First where the words came from, then a table of what each one changes on your site, then the single list to ship.

Where the two words come from

• GEO comes from research. A November 2023 paper from a group of university researchers, "GEO: Generative Engine Optimization" (accepted to KDD 2024), used it for improving how visible a piece of content is inside answers written by generative engines. The authors report that how well their tactics worked varied by subject area.
• AEO is a marketing term, not a research one. "Answer engine" means a tool that returns a direct answer instead of a list of links, and AEO means being the source that answer uses.

Different origins, so some writers keep a distinction: AEO for being the one answer (a featured snippet, a voice reply, a cited line in an AI Overview), GEO for being worked into a longer AI-written summary. It's a fair way to think about the output. It doesn't change what you do to the page, as the next section shows.

What each one changes on a page

The two middle columns are how people usually frame each term, not two official methods. The right-hand column is the work.

The verdict: one set of changes

Every row ends in the same action. If you do the right-hand column, you have done AEO and GEO. Spending a quarter deciding which acronym to pursue is time not spent fixing the pages.

There is also an SEO point under all of it. Google's documentation says that to appear in its AI Overviews and AI Mode a page must be indexed and eligible to be shown in Search with a snippet, and that there are no additional requirements or special optimizations (page last updated 10 December 2025, checked 9 October 2026, Google Search Central). So the base is ordinary SEO, with answer-first writing and accurate facts on top. Other engines publish less about how they choose sources, so measure rather than assume.

What to do this week

1. Write 20 to 50 buyer questions, some that name you and most that don't.
2. Ask each in ChatGPT, Perplexity and Google's AI Overviews or AI Mode and record who is named, what is said about you, and which sources are cited.
3. Fix wrong facts first. Put the current price, plan or limit in plain words on a page you control.
4. Rewrite the opening of the page that should win each missing question so it answers in two or three sentences.
5. Clear technical blockers: robots rules, content that only appears after JavaScript runs, missing canonicals, broken links.
6. Get named where the engines already look, using the cited sources from step 2.
7. Ask again every week with the same questions and compare over several weeks, not one answer.

Each step is explained in what is AEO and how to do it.

Where Dinkum fits

Dinkum runs this list for you. It asks a fixed panel of buyer questions weekly across ChatGPT, Perplexity and Google's AI answers, fixes the pages that state wrong facts, ships answer-first pages as pull requests you approve, and judges each change at 14 and 28 days against the rest of the site. It costs $2,000 a month for one site, month to month. If you're weighing agencies, read what an AEO agency should do and how to choose an AI SEO agency. There are no published client results yet, so we won't quote any.

Questions

Is AEO part of GEO?

Depends who you ask. Some treat AEO as the narrower job of being the single quoted answer and GEO as the wider one of showing up in generated summaries, so AEO sits inside GEO. Others use them as plain synonyms. The changes on your pages are the same either way.

Is AEO replacing SEO?

No. Answer engines mostly find pages through search, and Google says the same SEO fundamentals apply to its AI features. AEO adds answer-first writing, accurate facts, a consistent entity and mentions on the pages engines trust.

How do I use both?

Run SEO as the base: crawlable, indexed, useful pages. Add the AEO/GEO list above on top of the pages that matter, and track two things side by side: Search Console for clicks and impressions, and a fixed question panel for whether AI answers name or cite you.

Is GEO the same as AI search optimization?

In common use, yes. AI SEO, AI search optimization, AEO and GEO all describe making your site visible in AI-written answers.

Can anyone guarantee a ChatGPT or Perplexity citation?

No, and be wary of anyone who says so. You can make pages easy to fetch and quote and fix what AI gets wrong about you. The engines decide the rest. For the ChatGPT side in plain steps, see how to rank in ChatGPT.

Dinkum is the thing I've always wanted to make, and it's only now possible. The response has been far kinder than I expected, and I'm still working out what to do with that.

The thing I've always wanted to build

I've made a lot of things. Dinkum is the best of them.

It's also the one I always wanted to make. For years I could describe it but couldn't build it: something that does the search work on a website every day, properly, instead of sending a report about what someone should do. Now I can, and it's better than the version I used to describe.

What it does

Every weekday morning it does one piece of work on your site. I call each one a Dink. It reads your Search Console, your pages, what AI tools say about you and your sales, then picks the one change most worth making that day: a page for a search people are making, a fix for something that's wrong, a sharper title on a page that's nearly there.

Then it makes the change properly. A second AI reviews it, and the pages are rendered on desktop and phone before you see anything. It arrives as a pull request, and nothing goes live until you approve it. Two and four weeks later it checks whether the change worked against the rest of your site, and that answer shapes what it does next.

If you want the details, how Dinkum works walks through them step by step.

What it looks like

This is my own site, in the dashboard Dinkum runs on. Open any of these to look around.

The feedback has been more than I expected

People have been really positive about it. It's a bit overwhelming. I didn't see it coming, and I'm not going to pretend I'm used to it.

Dinkum is beating my own expectations, and I set those pretty high.

What's next

Keep making it better, one Dink at a time. The changelog shows what's shipped lately. I'll write more here as there's more to say, including the parts that don't go well.

If you'd like to see what it would do for your own site, the first five Dinks are free and yours to keep.

By Mitch Flindell

Short answer: AEO (answer engine optimization) is the work of making your pages the ones AI tools pick, quote and name when someone asks a question: ChatGPT, Perplexity, Google's AI Overviews and AI Mode. It builds on SEO and adds one job: being easy for a machine to lift a correct answer from. Nobody can guarantee a citation. Last updated 7 October 2026.

You will also see it called GEO (generative engine optimization) or AI search optimization. In practice the work is the same. The rest of this page shows how an answer engine ends up citing one page and not another, a checklist you can run against your own site this afternoon, how to do AEO in order, and how to tell whether it's working.

How an answer engine picks a page to cite

Details differ by product and change often, and the vendors don't publish their full recipe, so treat this as the general pattern rather than a rulebook. When a question needs fresh or specific facts, the engine typically does four things:

1. Searches. It turns the question into one or more searches and pulls back candidate pages. A page that can't be crawled, or doesn't rank for anything near the question, rarely gets this far.
2. Reads passages, not pages. It looks for the stretch of text that answers the question. A paragraph that states the answer in plain words is easier to use than one buried under a long introduction.
3. Cross-checks. It leans on facts that several sources agree on, and on sources that look authoritative for the topic. A business named consistently across its own site, directories and third-party lists is easier to trust than one described differently everywhere.
4. Writes the answer and links its sources. Some tools show citations for every claim; others answer from what the model already learned and cite little. That is why the same question can give a different answer tomorrow.

So the levers you control are whether your page can be fetched, whether it answers the question directly, whether its facts are right and match what others say, and whether you are named on the pages the engine already trusts.

AEO vs SEO

AEO doesn't replace SEO. A page that a search engine can't find is also hard for an answer engine to find.

A checklist you can run against your own site

Pick your five most important pages and check each one.

• Answer-first paragraph. Does the first paragraph under the heading answer the question in two or three plain sentences (who it's for, what it costs, how it works)? If someone copied just that paragraph, would it be correct on its own?
• Headings that are questions people ask. "How much does X cost?" beats "Pricing philosophy".
• Crawlable. Open /robots.txt. Are the bots from the AI tools you care about allowed to fetch your public pages? Is the answer in the HTML, not only drawn by JavaScript after load? Training crawlers and search crawlers are separate bots, and you can treat them differently.
• One canonical URL per page, a working sitemap, and no accidental noindex.
• Named entities, spelled the same everywhere. Your business name, product names, founder and location written identically on your site, your directory listings and your social profiles. Organization schema with your name, URL and profiles makes this explicit.
• FAQ and Article schema that matches the visible page. Structured data tells a machine what each part is. It doesn't force a citation, and Google no longer shows FAQ rich results, so add it to describe the page, not to chase a feature. Never mark up text that isn't on the page.
• Facts that are true today. Prices, plans, limits and dates. Wrong facts get quoted just as readily as right ones. Put a date on pages where the facts change.
• A comparison table where buyers compare things, with rows you can check.
• Third-party mentions. Are you on the lists, directories and review sites that rank for your buyers' questions?

How to do AEO

In this order, because each step tells you what the next one should be.

1. Write down 20 to 50 questions your buyers ask. Mix questions that name you ("is Acme any good?") with questions that don't ("best invoicing app for contractors"). The second kind is how strangers find you.
2. Ask them in each engine and save the answers. Record who is named, what is said about you, and which sources are cited.
3. Fix what is wrong first. If an answer states a price, feature or limit that you don't have, find the page it came from and put the current fact on a page you control, in plain words.
4. Make the page that should win each missing question answer it directly, in its first paragraph, then back it up with detail, a table or an FAQ.
5. Remove technical blockers: robots rules, JavaScript-only content, missing canonicals, broken links, slow pages.
6. Get named where the engines look. The cited sources from step 2 show which listicles, directories and publications matter. Earn the mention honestly; paid links that pass ranking signals break Google's spam policies.
7. Ask again on a schedule and compare.

How to measure AEO

One screenshot of one answer proves nothing, because answers vary from ask to ask. Measure like this instead:

• Keep the question list fixed. Changing it every week makes the trend meaningless. Add new questions alongside the old ones, not in place of them.
• Use the same engines each time: ChatGPT, Perplexity, and Google's AI Overviews and AI Mode, plus Bing if your buyers use it.
• Record per question: were you named, were you cited (a link to your site), what was said, and who else was named.
• Report a rate with its sample size ("named in 6 of 30 buyer questions"), and treat a one-or-two-question wobble as noise. Look for a change that holds over several weeks.
• Keep Search Console open next to it. It shows clicks and impressions, which AI answers often don't produce, so AEO wins can look quiet there. Analytics referrals from AI tools are a partial signal, not the full picture.

You can do this by hand in a spreadsheet. Tools exist that automate the asking.

Where Dinkum fits

Dinkum does much of this for you: it asks a fixed panel of buyer questions weekly across ChatGPT, Perplexity and Google's AI answers, fixes the pages that state wrong facts, ships answer-first pages as pull requests you approve, and judges each change at 14 and 28 days against the rest of the site. For third-party mentions it lists the directories and articles engines cite for your market and drafts outreach pitches, which you send. It costs $2,000 a month for one site, month to month. Read what an AEO agency should do, or how to choose an AI SEO agency if you're comparing options. There are no published client results yet, so we won't quote any.

Questions

What does AEO stand for?

Answer engine optimization: optimizing your content so answer engines (ChatGPT, Perplexity, Google's AI answers) can find, trust and quote it.

What is the difference between AEO and GEO?

Very little. Both mean getting named and cited in AI-written answers. Some people use GEO for AI-written summaries and AEO for being the single answer, but the work is the same.

Is AEO replacing SEO?

No. Answer engines mostly find pages through search, so crawlable, well-ranked pages are still the base. AEO adds answer-first writing, accurate facts and mentions on the pages engines trust.

Can I guarantee ChatGPT will cite my site?

No, and be wary of anyone who says they can. You can make your pages easy to fetch and quote, and fix what AI gets wrong about you.

How long does AEO take to work?

Some engines fetch live pages, so a fix can show within days or weeks. Others change slowly. Judge a change over several weeks of the same questions, not one ask.

Can I do AEO myself?

Yes. The steps above need a spreadsheet, access to edit your pages and a couple of hours a week. For a plain-English walk-through with prompts to paste and a checklist, read how to rank in ChatGPT and get your business recommended. If you'd rather have it done and reviewed, get your first five Dinks free.

By Mitch Flindell

Short version: an AEO agency gets your business named, described correctly and cited when buyers ask ChatGPT, Perplexity, Google AI Overviews and AI Mode. The work is five things: track a fixed set of buyer questions every week, see who gets named instead of you, fix wrong facts AI repeats about you, write answer-first pages AI can quote, and get onto the lists and directories AI cites. Nobody can promise a citation. Last updated 6 October 2026.

AEO is answer engine optimisation: getting AI tools to answer a buyer's question with your business in it. People also call it GEO (generative engine optimisation) or AI search optimisation. The names differ by agency; the job is the same. Below is what each part of the work should look like, so you can ask for it by name when you hire someone, and a plain account of how Dinkum does each part.

1. Track a fixed panel of buyer questions, every week

AI answers change from one ask to the next, so a single screenshot proves nothing. The honest way to measure is a fixed list of the questions your buyers actually ask ("best invoicing app for contractors", "who does managed SEO"), put to each engine on a schedule, with the answers recorded. Questions that don't name you matter most, because they're how a stranger finds you.

Ask any agency: how many questions, which engines, how often, and can I see the raw answers? Small differences week to week are noise; a change that holds over several weeks is a signal.

How Dinkum does it: we track a panel of buyer questions weekly across ChatGPT, Perplexity and Google's AI answers, and report on it in the weekly report.

2. Find who gets named instead, and what the AI cites

When an answer names three competitors and not you, the useful question is why. AI tools lean on particular pages: comparison articles, "best of" lists, directories, review sites and a few well-structured vendor pages. The work is listing those sources for each question, then seeing which ones you are missing from and what your own page says that theirs doesn't.

How Dinkum does it: for each question where competitors are named and you aren't, we record which pages the engines cite and which page on your site should have answered it. That list decides what gets worked on next.

3. Fix wrong facts AI repeats about you

This is the most under-sold part. If ChatGPT quotes a price you stopped charging two years ago, or a feature you don't have, buyers act on it. AI tools learn from what is published, so the fix is on the source: make your own pages state the current fact plainly, and correct the pages you control that still carry the old one.

An agency should check what AI says about your prices, plans, limits and features against what is true, and show you the mismatch and the page it traces to.

How Dinkum does it: we compare what AI tools state about you with the facts in your product and your site, and fix the pages that disagree. Every claim on a page we change has to match your product. We don't add claims to please an AI.

4. Write answer-first pages AI can quote

An engine picking a passage to quote favours a page that answers the question in its first paragraph, in plain words, then backs it up. In practice that means:

• The direct answer first: who it's for, what it costs, how it works.
• A comparison table where buyers compare things, and an FAQ where they ask follow-ups.
• Facts a reader can check, each one true on the day it's published.
• Clean titles, canonical URLs and schema, so the page is easy to read and attribute.

No page format forces an AI to cite you. It makes a good page easier to use, which is all anyone can honestly claim.

How Dinkum does it: we write or rework the page that should answer each missing question, then judge it at 14 and 28 days against the rest of the site.

5. Get onto the lists and directories AI cites

Engines often repeat what third-party lists say. If every "best X" article and directory the AI draws on leaves you out, a better page on your own site only goes so far. The work is finding those lists, working out which are worth being on, and approaching them honestly. Paid links that pass ranking signals are not part of it; they break Google's spam policies.

How Dinkum does it: each month we list the directories and listicles that engines cite for your market, and draft outreach pitches for you to send.

How the pieces fit together

How Dinkum works, in one paragraph

Dinkum ships one change a day to your site as a pull request. A second AI reviews it, pages are rendered on desktop and phone first, and you approve it, ask for changes or close it. Every change is judged at 14 and 28 days against the rest of the site. We track Google, Bing, ChatGPT, Perplexity and Google AI answers. It costs $2,000 a month for one site, month to month, with no setup fee, and everything lives in your repository so you keep it if you stop. We need your site's code (on GitHub; if it isn't, we'll migrate it), Search Console access, a Slack channel and about ten minutes a day to approve a change. There are no published client results yet, so we won't quote any. For the general buyer's checklist, read how to choose an AI SEO agency.

"GEO agency" and "generative engine optimization agency"

GEO and generative engine optimization are another name for the same job. Some people use GEO for getting cited in AI-written answers and AEO for getting picked as the single answer, but the five parts above cover both, and no agency can show you a clean line between them. If a provider says GEO is something entirely separate, ask what they would do that isn't in this list.

"AI search optimization agency"

This one is the widest label. It can mean AEO, or it can mean ordinary SEO done with AI tools. Ask which. Dinkum does both: the normal search work (titles, redirects, internal links, speed, new pages) and the AI answer work, from one daily change queue. Treat "AI search optimization" as a reason to ask for a worked example of one change, not as a service in itself.

Questions to ask an AEO agency

1. Which questions do you track for us, on which engines, and how often?
2. Show me the raw answers, not a summary score.
3. What have you changed on a site because of an AI answer, and what happened afterwards?
4. How do you check what AI says about our prices and features?
5. Who edits my site, and do I approve each change?
6. What do you refuse to do? (Link buying and guarantees should be on the list.)

Questions

What does an AEO agency do?

It measures what AI tools say about your market with a fixed set of buyer questions, finds who is named instead of you and which pages are cited, corrects wrong facts, writes pages AI can quote, and gets you onto the lists and directories AI draws from.

What is the difference between AEO, GEO and AI SEO?

In practice, very little. AEO and GEO both mean getting named and cited in AI answers. "AI SEO" sometimes also means using AI tools for ordinary SEO. Ask the provider which they mean.

Can an agency guarantee I'll be cited by ChatGPT or Perplexity?

No. Nobody controls an AI model's answer. A provider can show you what is said today, fix what's wrong, and make your pages easy to quote. Treat a guarantee as a red flag.

How much does AEO cost with Dinkum?

$2,000 a month for one site, month to month, with no setup fee. It covers AEO, GEO and ordinary SEO together.

How would I find out what AI says about my business today?

Send us your site and your first five Dinks are free. We measure it, check what AI tools say about it, then make the five changes that will help most, each for you to approve. To check by hand, our step-by-step guide has the prompts to paste into ChatGPT, Perplexity and Google AI.

By Mitch Flindell

Short version: pick the one that shows you the exact change before it goes live, asks for the access it needs to make it, and judges results against your own Search Console data. Walk away from anyone who guarantees rankings, buys links, or won't say who touches your site. Last updated 6 October 2026.

"AI SEO" gets used for two different things. Some agencies mean using AI tools to do ordinary SEO faster. Others mean getting your business named in ChatGPT, Perplexity and Google's AI answers (also called AEO or GEO). Most good ones now do both, so the first question is which one you are buying. The rest of this page is a checklist you can take into any sales call, ours included.

1. What access will they need, and why?

Real SEO work needs real access. Expect to be asked for:

• Google Search Console (and Bing Webmaster Tools if you use it). This is where your clicks, impressions and rankings actually live. Read access is enough for measuring; ask what else they want and why.
• Your site's code or CMS. Fixing titles, redirects, internal links, schema and page speed means changing the site. An agency with no way to change it can only send you a list.
• Analytics or sales data, so results can be judged on customers and revenue, not only traffic.

Ask where each credential is stored, who can see it, and how you revoke it. Give the least access that does the job, and use accounts you can remove.

Where Dinkum stands: we need your site's code. Dinkum works by opening pull requests against your repository, so a git-hosted site (GitHub) is a requirement. If yours isn't on GitHub, we'll migrate it for you first, keeping every URL, redirect and page title. We also need Search Console access, a Slack channel, and about ten minutes a day to approve a change.

2. What does each deliverable look like?

"Content", "technical fixes" and "link building" are labels, not deliverables. Ask to see a real example of each one, and what you'd receive on day 1, week 1 and month 1.

• A fix: the specific page, what was wrong, the before and after, and the evidence it mattered (for example, a page at position 11 with a title that isn't earning clicks).
• A new page: the search it targets, why you can win it, and what it says that the pages already ranking don't.
• An AI-answer change: the question buyers ask an AI tool, what the tool says today, and the page you'll change so it can say something accurate.
• A report: one that says what changed on your site, not only what moved in a third-party rank tracker.

A 40-page audit is not a deliverable. A change on your site is.

3. How are changes approved?

Somebody is going to edit pages that customers read. You should know exactly how that works.

• Is every change shown to you before it goes live, or only listed afterwards?
• Can you say no, and what happens next?
• Which parts of the site are off limits (checkout, logins, anything with customer data)?
• Can you undo a change? With version control, every change is a record you can revert.
• Who checks that claims on your pages (prices, features, limits) are true?

Where Dinkum stands: each change arrives as a pull request with its reasoning attached. A second AI reviews it, and pages are rendered on desktop and phone before you see it. You approve it, ask for changes, or close it. Nothing reaches your live site until you say so.

4. How will results be judged?

SEO is slow, so a vague answer here is how retainers drift on for a year. Ask:

• What is the baseline, and who measures it? It should be your Search Console data, not a figure only they can see.
• When is each change judged, and against what? A change should be compared with the rest of the site over the same period, so a seasonal bump isn't claimed as a win.
• What happens to a change that doesn't work?
• Do they track AI answers (ChatGPT, Perplexity, Google's AI answers) with a fixed set of questions, or just show you one screenshot?

Where Dinkum stands: every change is judged at 14 and 28 days against the rest of the site, and we track Google, Bing, ChatGPT, Perplexity and Google AI answers. Search moves slowly: a single change usually shows up in two to four weeks, and traffic compounds over months.

5. Red flags

• Guaranteed rankings, traffic or AI citations. Nobody controls Google or an AI model, and Google's own guidance warns against anyone who guarantees a #1 ranking.
• Link buying or "private blog networks". Paid links that pass ranking signals break Google's spam policies and can get a site demoted.
• No access to the site, but promises of fixes. Either someone on your team will do the work, or it won't be done. Ask which.
• No named person or process for who changes what.
• Reports built from vanity numbers, such as keywords ranked or "domain authority", with no tie to clicks or customers.
• Invented proof. Case studies with no client name or numbers you can't check. Ask for something you can verify.
• Bulk pages with the keyword swapped. Templated pages are what Google's spam updates target.
• Long lock-in with ownership of content or tooling staying with them. Ask what you keep if you stop.

Three ways to buy: agency, consultant, managed service

The right model depends on who on your side will do the work. These are typical patterns, not a statement about any one company; check each provider's own terms.

Dinkum is the managed service in this table, so weigh our view accordingly. The questions above apply to us as much as to anyone. Weighing a software platform instead of a service? See BrightEdge alternatives.

A short list to take to the call

1. What access do you need, and what's the minimum?
2. Show me a real change from start to finish.
3. Where do I approve changes, and can I refuse one?
4. How and when do you judge whether a change worked?
5. What do I keep if I leave?
6. Do you track what AI tools say about us, and how?
7. What won't you do? (A good answer includes link buying and guarantees.)

Questions

Do I need an AI SEO agency, or is regular SEO enough?

The basics are the same: a site Google can read, pages that answer real searches, and claims that are true. AI answers add a second job: making sure tools like ChatGPT and Perplexity find accurate information about you on pages they trust. Most providers now cover both. If you'd like to try it yourself first, here is how to rank in ChatGPT and get your business recommended, step by step.

How long before I see results?

Expect work shipped in the first week. Search rankings usually respond in two to four weeks per change, and the gains build over months. Treat any promise of faster as a red flag.

Can an agency work on my site without access to the code?

It can advise, and it can edit through a CMS login if your site has one. It can't fix redirects, schema, speed or templates it can't reach. Dinkum needs the code because it works through pull requests.