Does GEO replace SEO? What Google's own documentation requires, and a 30 page eligibility test

No. GEO does not replace SEO, and the reason is in Google's own documentation rather than in anybody's opinion. The figure to look for is the count of new technical requirements that Google states are needed beyond ordinary Search eligibility in order to appear as a supporting link in its AI features, and that count is zero. We are deliberately not reprinting Google's sentence as a dated fact of ours, because a help page is edited without notice and a restated help page goes stale into a false claim. Open Google Search Central's documentation on AI features and your website, find the eligibility section, write down the URL and the date you read it, and keep a screenshot. If it still says there are no additional technical requirements, your count is zero and your evidence is that URL plus that date. What GEO adds is not a second set of technical requirements. It is a different unit of winning, and the one measurement we hold on that is from ChatGPT on 18 August 2026, where across eighteen blind commercial questions in legal technology one company was the top source on 1 of 18.

This page is written the way it is because of something we watched happen. On 17 September 2026 Claude read two comparison pages that stated figures correctly and gave no link, no date and no source. It checked the figures independently, found them correct, and still treated the pages as advocacy and used official sources instead. Correct and unsourced lost. So a page telling you what Google requires, with no instruction to go and read Google, would be committing the error it is warning you about.

How it was measured

There are two separate things to measure here, and almost every argument about GEO and SEO is two people measuring different ones. Keep them apart.

The documentation question. How many technical requirements exist that are specific to AI features and do not already exist for ordinary Search. This is not measured by testing. It is read off the operator's own documentation, on a date, and it is only true until the page is edited. The method is four steps.

• Name the document, not the claim. For Google, the relevant material sits in Search Central documentation under the pages about AI features and your website, and the general guidance on Search Essentials. Go to the source and read the current text.
• Record four things together: the full URL, the date you read it, the exact section heading, and a screenshot. Any one of those missing and you have an assertion rather than a citation.
• Write the count you found, as a number, and write what it was a count of. Additional technical requirements beyond Search eligibility is a different count from recommended practices, and a page that mixes the two will mislead you six months later.
• Recheck on a schedule and keep the old readings in a change log. When the documentation changes, you want to know which of your decisions were based on the previous text.

The outcome question. Whether the pages you already have get used as supporting links, and which ones. This is measured by testing, and it is the 30 page eligibility test below.

The 30 page eligibility test, in full. The point of the test is to separate three failures that look identical from the outside: the page was never crawled, the page was crawled and not chosen, and the page was chosen and gave the answer nothing worth quoting.

• Pick 30 pages you already have, spread across the kinds you publish. Ten pages is too few to see a pattern by page type. Thirty gives you roughly six per type across five types.
• For each page, record whether it is indexed and on what date, from URL Inspection in Search Console. This is the first gate and it is the one people skip. A page that was never indexed did not lose on quality.
• Check what a crawler actually receives. Fetch the page without running scripts and read the text you get back. This gate catches a failure we have measured: on 6 August 2026, for clawlaw.in, the pricing page rendered its prices only after scripts ran, so what a crawler received contained no prices at all, and the prices the assistants did quote had come from an app store listing instead of from the company's own site.
• Check the page carries at least one quotable, checkable fact. A count, a dated price, a named coverage list, a numbered process. A page with nothing specific on it gives an answer nothing to say about you.
• Write one blind query per page, the question that page is the answer to, with your brand in none of them. Thirty pages, thirty queries, frozen in a file with a version and a date.
• Run each query on four engines, three times each. That is 30 times 4 times 3, which is 360 responses. Arithmetic, printed so the denominator on your report is not in doubt.
• Score each response with the five states: brand named, own domain linked, top source, recommended and cited together, and used without attribution.
• Report by page type, not only in total. The total tells you how you are doing. The split by page type tells you what to write next, which is the only part that changes your week.
• Record every claim an engine makes about having searched, and mark it unverified. In three separate batches of our September 2026 audit, Perplexity withdrew its own earlier statement, saying it could not honestly substantiate the claim that it had run a live search for each question.

What the numbers were

On the documentation count. The number to carry is zero additional technical requirements, and its evidence is the Search Central page you read plus the date you read it. We publish no quoted passage and no date of our own for it, on purpose, under the rule that we do not restate a vendor's documentation. If you read that page and it now says something different, the count is whatever it now says, and this page is wrong and should be corrected. Tell us and we will correct it.

On the outcome side, here is everything we hold, with sectors named. None of it is an AI Overviews measurement, because we have not run an AI Overviews capture, and none of it should be read as one.

• 1 of 18 responses had the company as the top source. Legal technology, clawlaw.in, 18 August 2026, ChatGPT, eighteen blind commercial questions. The company had published pages. Publishing is the start of the work, not the end of it.
• 21 of 78 responses named the company, and 0 of 78 involved a live search that the engine would confirm. Legal technology, clawlaw.in, 27 July 2026, ChatGPT. The same reading put ProVakil on 28 of 78 and Legistify on 18 of 78. A memory based reading like that one is a ranking of what the model absorbed, which is why ordinary indexing and ordinary SEO still decide whether it ever absorbs you.
• One page, cited fourteen days after publication. On 6 August 2026 ChatGPT cited clawlaw.in/blog/how-to-check-a-companys-court-cases-in-india for a vendor due diligence question, where the same question set had named the company nowhere at baseline. That URL is public, so this is the one figure here a reader can verify today. It is also one page on one engine. One first appearance is not a median, and we publish no median.
• 0 of 6 questions in which any third party review or directory source reached an answer. Legal technology, clawlaw.in, 17 September 2026, Claude. On the how to questions in that same run, official court portals took every position above any commercial product, and the assistant said plainly that no commercial product should rank above the official portal for a question about using that portal.

That last one is the clearest available answer to whether GEO replaces SEO. The sources that decide these answers are heavily official pages and vendors' own pages. Counted across the whole September 2026 aiknowsus.com capture, the domains cited most often were the assistants' own documentation and the vendors' own websites, with a single well known review site far down the list. Your own site being crawlable, indexed and specific is the input to both jobs. It is not a legacy activity you graduate out of.

What has not been run, as of 29 September 2026. We have not executed the 30 page, 4 engine, 3 run test on any domain, so this page publishes no eligibility rate over 360 responses, no split by page type, and no comparison between indexed and not indexed pages. We have also not run any AI Overviews capture, so nothing here describes AI Overviews behaviour.

What this cannot tell you

• It cannot tell you Google's rule for when an AI feature appears, or which pages it will choose. That is not published and no test from outside recovers it.
• A documentation count is true on a date only. Help pages are edited. Yours should be rechecked and logged, and so should ours.
• Nothing here is an AI Overviews measurement. Every figure above is from ChatGPT, Claude or Perplexity. Carrying them across to Overviews would be the same error as carrying a legal technology number onto a marketing agency page.
• Zero additional technical requirements does not mean no work. It means the technical gate is the ordinary one. The competitive gate, which is whether your page is the most specific answer available, is where the work is, and it is not a technical setting.
• A 30 page test measures your domain. Domain age, existing traffic and internal linking travel with the result, so your number is about you rather than about the web.
• No method here promises a position in an AI answer, and we do not sell one.

Sources and change log

Documents to read yourself, rather than trust us on. Google Search Central documentation, the pages covering AI features and your website and Search Essentials. Read the current version, record the URL and the date, and keep the screenshot. We name the document instead of quoting it so that this page cannot become a false claim by standing still while the document moves.

Our own figures. The clawlaw.in programme of July to September 2026, recorded in GEO_BASELINE_RESULTS_2026-07-27.md, GEO_GAP_ANALYSIS_2026-08-18.md and the assistant audit files of 17 September 2026, and the aiknowsus.com audit of September 2026 across 24 batches and 72 conversations. The 6 August 2026 citation is externally checkable because the URL is public. The counts over our capture files are not checkable from outside until those files are published, and we say so rather than implying otherwise.

Version. 29 September 2026, first publication. It changes when the 30 page test completes, when Google's documentation changes in a way that contradicts anything above, and whenever a reader shows us an error.

Common questions

So do I stop doing SEO?

No, because the first two gates in the test above are ordinary SEO gates. Is the page indexed, and does a crawler receive the content. A page that fails either of those cannot be chosen by anything, and we have a dated case of the second failure: on 6 August 2026 a pricing page's prices existed only after scripts ran, so crawlers received no prices and the assistants quoted an app store listing instead.

Is there a separate robots or schema setting that makes me eligible for AI answers?

Go and read the documentation for the operator you care about, on the day, and record what it says. We will not restate it here. What we can tell you is the shape of the answer to look for: a count of requirements that are additional to ordinary eligibility, and the number to expect for Google is zero.

What is actually different about GEO then?

The unit of winning. A results page can show ten links and let the reader choose. An answer usually names three and cites a handful of sources. There is no second page. That changes what you write, not what you configure, and our own numbers show how thin the top of that funnel is: top source on 1 of 18 for a company that had published pages, on ChatGPT on 18 August 2026.

Do I need a new page for every buyer question?

One page per question is the pattern that produced the only verifiable citation we hold. The page cited by ChatGPT on 6 August 2026 answered one question, and it was cited fourteen days after publication for a question the same set had not named the company on at baseline. That is one observation, not a rule, and it is the observation we have.

Can I test this without Search Console access?

Partly. You can run the blind queries and score the responses without it. You cannot separate never crawled from crawled and not chosen, which is the difference between a technical problem and a content problem. Get the access before you draw conclusions about your content.

How long should I wait before judging a new page?

Set the horizon before you start and record the pages that never appear, rather than quietly dropping them. We hold one first appearance at fourteen days, on one page, on ChatGPT, on 6 August 2026. Because we have no failure set published beside it, that number cannot be turned into an expectation, and we do not present it as one.

What to do first

Do the documentation step today, because it takes ten minutes and it settles the argument. Open Google Search Central's material on AI features and your website, read the eligibility section, and paste the URL, the date and a screenshot into a file called eligibility evidence. Then run a script free fetch of your three most commercial pages and read what comes back. If your prices, your coverage or your key numbers are missing from that text, you have found the work to do this week, and it is ordinary SEO work.