What Google actually requires
Google’s generative AI features are built on its core Search ranking and quality systems. They surface content from the same index that produces ordinary results, which is why Google states there are no additional requirements or special optimisations needed to appear in them.
The eligibility conditions are narrow and specific. A page must satisfy the Search technical requirements, be indexed, and be eligible to be shown with a snippet. The site must also be included in Search generative AI features in Search Console. Crawling must be permitted in robots.txt and not blocked by a CDN or hosting layer. Meeting all of this makes a page eligible; it guarantees nothing.
The snippet condition has a consequence most pages miss. Because appearing in AI features depends on snippet eligibility, applying nosnippet, data-nosnippet or an X-Robots-Tag preview control to a page also removes it from AI Overviews and AI Mode. Publishers who added those directives to protect content sometimes did not intend that second effect.
- Indexed and eligible for a snippet, per the Search technical requirements
- Included in Search generative AI features in Search Console
- Crawlable in robots.txt and not blocked at the CDN or host
- Structured data, where used, matches the visible text on the page
What the work actually consists of
Google’s own guidance points at content rather than markup. It asks for unique, expert-led, non-commodity content that goes beyond common knowledge, and illustrates the difference with an example: a first-hand account of waiving a home inspection and what the sewer line revealed, rather than a generic list of tips for first-time buyers. The distinction is between something only the author could write and something anyone could assemble.
The rest is unglamorous. Organise content for human readers with clear sections and headings. Keep important information in textual form rather than locked inside images or video. Make pages findable through internal links. Treat semantic HTML as a readability tool rather than an exercise in perfect code. Keep Merchant Center and Business Profile data current where they apply.
For measurement, Google added a Generative AI performance report in Search Console and warns explicitly against third-party tools claiming access to internal Google metrics. That warning is worth taking literally, because several vendors sell AI citation dashboards whose numbers cannot be reconciled with any first-party source.
Where the emphasis genuinely differs
Saying AEO is still SEO does not mean nothing changes. What changes is which qualities of a page matter most, and that is a matter of emphasis rather than technique.
A page written to earn a click can afford to withhold. It can tease a conclusion, spread an answer across several paragraphs, and rely on the surrounding page for context. A page written to be quoted cannot. An answer engine lifts passages out of context, so a passage that only makes sense after reading the two above it is unusable. Self-contained paragraphs, direct answers stated before they are elaborated, and real questions phrased the way people actually ask them all serve that purpose.
The honest trade-off is that being quotable can reduce clicks. If a page answers a question completely in its first paragraph, some readers will take the answer and leave. That is an acceptable cost for informational pages where the goal is authority, and a poor one for pages whose job is to sell. Deciding which is which per page is the actual strategic work, and it is the part no tool does for you.
Controls, if you want less exposure rather than more
Not every publisher wants maximum inclusion. The controls are separate and easy to conflate.
Within Google Search, the Search generative AI control in Search Console governs inclusion in generative AI features, and preview controls such as nosnippet remove snippet eligibility and therefore AI feature eligibility with it. Google-Extended is a different mechanism and applies to AI training and grounding in Google systems outside Search generative AI features, not to AI Overviews. Blocking one does not block the other, and confusing the two produces the opposite of the intended result.
What most sources get wrong
Each claim below is one we found published and being repeated. The correction is drawn from primary documentation.
Commonly claimed: AEO is a separate discipline with its own technical requirements. One widely-shared checklist states that AEO has "10-15 additional requirements that SEO alone doesn’t address" and that roughly 62% of SEO practice carries over.
What the sources say: Google states that from Google Search’s perspective, optimising for generative AI search is optimising for the search experience, and thus still SEO, with no additional requirements to appear in AI Overviews or AI Mode. Google further advises prioritising effective SEO over AEO or GEO hacks, noting many circulating suggestions are not effective or supported by how Search works.
Commonly claimed: Structured data is required for AI citation. One agency guide frames the shift as "search engines use schema as a signal, AI engines use schema as a source", calls schema the single highest-impact AEO tactic, and cites a figure of 3.2 times more citations for sites with full markup.
What the sources say: Google lists overfocusing on structured data among the things to ignore for generative AI search, stating structured data is not required and there is no special Schema.org markup to add for AI features. Schema remains worthwhile for rich result eligibility in ordinary Search. The 3.2 times figure carries no disclosed methodology, which is why this page does not repeat it as fact.
Commonly claimed: Publishing llms.txt or llms-full.txt improves visibility in AI answers.
What the sources say: Google states you do not need to create machine-readable AI text files because Google Search does not use them, and that maintaining llms.txt will neither harm nor help visibility or rankings in Google Search because Search ignores it. It may still serve other systems that consume it, but it is not a Google Search tactic.
Commonly claimed: AI citation is more equitable than search because engines cite on relevance rather than authority, so small brands can rank in weeks.
What the sources say: Because generative AI features draw on the same core ranking and quality systems as Search, the qualities that earn ranking also govern inclusion. There is no documented separate pathway that bypasses them, and no first-party source supports a two-to-six week timeframe. Nine99 does not promise one.
The Nine99 view
Our own judgement, not drawn from the sources above. Treat it as opinion formed in practice rather than documented fact.
The reason this field is so confidently wrong is commercial. A new discipline with its own file formats, its own audit and its own dashboard is far easier to sell than the accurate version, which is that the fundamentals did not change and most sites are still failing at those. An agency that tells a client their existing content is thin has a harder conversation than one that sells an llms.txt deployment.
What we do in practice, in this order: confirm the page is indexed and snippet-eligible, because everything else is irrelevant if it is not; check no preview control is silently suppressing it; then rewrite for self-contained passages and direct answers. Markup goes on for rich result eligibility, not as an AI tactic. The sequencing matters because the first two steps are cheap and occasionally explain the entire problem.
The failure mode we see most often is a page that reads well to a human but collapses when a single paragraph is extracted, because every sentence depends on the one before it. That is invisible in a content audit and obvious the moment you try to quote it. Reading a page paragraph by paragraph, out of order, is a cruder test than any tool and finds more.
Common questions
Is AEO different from SEO?
Not as a discipline. Google states that optimising for generative AI search is optimising for the search experience and is therefore still SEO, because its AI features run on the same core ranking and quality systems as ordinary Search. The genuine difference is emphasis: AEO prioritises passages that remain accurate and coherent when quoted out of context, where classical SEO also concerns itself with earning the click.
Do I need schema markup to be cited by AI?
No. Google explicitly lists overfocusing on structured data among the things you can ignore for generative AI search, and states there is no special markup you need to add for AI features. Schema is still worth maintaining because it makes you eligible for rich results in ordinary Search, and Google asks that any structured data matches the visible text on the page.
Should I publish an llms.txt file?
It will not affect your visibility in Google Search either way. Google states it does not use these files and that maintaining one will neither harm nor help your rankings, because Search ignores them. If a specific non-Google system you care about consumes llms.txt then publishing one is harmless, but it should not be sold to you as a Google tactic.
Can I stop my content appearing in AI Overviews?
Yes, through the Search generative AI control in Search Console, or by removing snippet eligibility with nosnippet or an X-Robots-Tag preview control. Be deliberate about the second method, because snippet eligibility is a precondition for AI features and removing it affects ordinary Search previews too. Google-Extended is a separate control covering AI training and grounding outside Search generative AI features.
How do I measure whether AEO is working?
Use the Generative AI performance report in Search Console, which is the only first-party view of performance in Google generative AI features. Google warns explicitly against third-party tools claiming access to internal Google metrics. Citation trackers for non-Google assistants exist and can be directionally useful, but treat any precise citation-rate figure with suspicion unless the methodology is published.
Sources
Factual claims on this page are attributed. Last reviewed 2026-08-26.