Google has published official guidance on its generative AI features, which puts AI Overviews in an unusual position: it is the only AI search surface where the operator has told you how it works.
The guidance is more deflationary than most coverage suggests. Its central claim:
“From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.”
That is not marketing language. It follows from the mechanism.
How AI Overviews are assembled
Two mechanisms do the work, and neither involves a separate index.
Retrieval-augmented generation, which Google calls grounding, uses the existing Search ranking systems to retrieve relevant pages. The model reads those pages and composes an answer with links back.
Query fan-out generates related queries alongside the original. Google's own example: a search for how to fix a weedy lawn also fires queries about herbicides and weed prevention, pulling results for each.
The consequence is direct. If AI Overviews are built from pages retrieved by normal ranking, then ranking normally is the requirement. There is no separate entrance and no separate optimisation.
Query fan-out has one practical implication: your page can be surfaced for a query it doesn't target, if it answers a related question the fan-out generated. This argues for covering a topic properly rather than for producing a separate page per phrasing — and Google warns explicitly against the latter, noting that creating separate content for every possible query variation is not the useful response.
The setting that removes you entirely
Before any optimisation work, check two things. Either can make you absent while everything looks normal.
Google-Extended in robots.txt. This governs whether Google's AI products may use your content. It is separate from Googlebot and has no effect on rankings. Block it and every ranking stays exactly where it is — you simply stop appearing in AI Overviews and Gemini, with no notification.
Most robots.txt files have never named it. That's a default, not a decision.
The Search Console preference for generative AI. There is a control determining whether your site appears in and grounds generative AI responses. Opting out removes both the appearances and the impressions.
Neither produces an error. Both produce silence.
The four tactics Google says to ignore
The guide contains a section called “Mythbusting generative AI search: what you don't need to do.” Four items are currently sold as services.
llms.txt. “You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them.” Maintaining one “will neither harm nor help” your Google visibility.
Chunking content into small pieces. “There's no requirement to break your content into tiny pieces for AI to better understand it.” And: “There's no ideal page length.”
Rewriting content for AI. “You don't need to write in a specific way just for generative AI search.” Google's systems handle synonyms and intent, so chasing phrasings is redundant.
Chasing mentions. AI features do surface what's said about you elsewhere, but “seeking inauthentic 'mentions' across the web isn't as helpful as it might seem.”
A proposal built mainly on these four is proposing work Google says has no effect on Google.
What Google says does matter
The guidance is equally direct here:
“Creating content that people find unique, compelling, and useful will likely influence your website's presence in generative AI search in the long run more than any of the other suggestions in this guide.”
It draws the distinction with its own examples:
| Google's example | Type |
|---|---|
| “7 Tips for First-Time Homebuyers” | Commodity |
| “Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line” | Non-commodity |
The test Google offers: whether the content “could easily be produced by a generative AI model.” If it could, it is unlikely to be what one chooses to cite.
This is the part that resists checklists. There is no technical implementation of having something to say.
The technical baseline that still applies
Since AI Overviews draw on normal Search, everything that makes a page rank continues to matter — and a few things have added weight because passages get extracted rather than whole pages read.
Answer early. A direct answer near the top of the page is more likely to be extracted accurately.
Structured data. Product, FAQ, Organization, Article and Breadcrumb markup help systems understand what a page contains. Google's Rich Results Test validates it.
Dates. datePublished and dateModified. Recency is a factor and unknown recency is not treated as recent.
Clean headings. Descriptive H2s that name what's below them, because they guide extraction.
Speed and mobile. Unchanged from ordinary SEO, and unchanged in importance.
Measuring it
Google added generative AI performance data to Search Console in 2026. AI Overview visibility is now observable rather than inferred, which is a meaningful advantage over every other AI surface — nothing comparable exists for ChatGPT, Claude or Perplexity.
Worth knowing what the data can and can't show: appearances and clicks are visible; the composition of the answer and why a particular source was chosen are not.
The wider pattern reported across the industry is that AI Overviews reduce clicks for informational queries — the answer is on the results page, so fewer people click through. This is real and is not something optimisation reverses. The rational response is to shift emphasis toward queries where someone needs to reach you to act, which are the queries that were worth more anyway.
Where this guidance stops
One qualification worth holding onto: Google is describing Google.
ChatGPT, Claude and Perplexity are separate systems that do not use Google's index, do not run Google's ranking, and have published nothing equivalent. Two differences are load-bearing: each uses its own crawler requiring its own permission, and most of them do not execute JavaScript at all, where Google does on a delayed pass.
A site can therefore be fully indexed by Google, appearing in AI Overviews, and functionally invisible to ChatGPT — for reasons that have nothing to do with content.
Questions
Does appearing in AI Overviews help or hurt traffic? Both, depending on query type. Informational queries lose clicks. Commercial queries, where the user still needs to reach a provider, are affected less.
Can I opt out of AI Overviews but keep rankings? Yes — that's precisely what Google-Extended does. Rankings are unaffected. Whether it's a good idea depends on whether your content's value is in being read on your site or in being found at all.
How long after publishing before a page can appear? It depends on normal indexing and ranking, since AI Overviews draw on the same systems. A page has to rank for the underlying queries before it can be retrieved for an answer.
Should I create FAQ pages specifically for this? FAQ content genuinely answering questions people ask is useful. FAQ pages manufactured to capture AI Overview appearances are the commodity content Google's guidance warns about.
Does schema markup guarantee inclusion? No. It helps systems understand a page. Nothing guarantees inclusion, and any service claiming otherwise is claiming control it doesn't have.
Sources: Google Search Central, “Optimizing your website for generative AI features on Google Search.”