Skip to content
search-engine-optimizedby Netsolution

The engine has published its generative AI optimisation guide

Current recordDocument dated , published by Google Search Central, read on

The engine publishes an optimisation guide for its generative AI features. Its answer fits in one sentence: the usual good practices still apply, because those features rest on the same ranking systems. The document then names four useless practices, sold everywhere.

What the document says

The engine has published a documentation page devoted to optimising for its generative AI features, announced on its blog in May 2026 and updated since. The question opening it is the one everybody is asking: do the usual search practices still serve?

The generative AI features of search rest on the ranking and quality systems of search itself, and SEO good practices therefore still apply to them.

officialSource : Google Search Central, “Google's Guide to Optimizing for Generative AI Features on Google Search”

The answer fits in one sentence of the document:

In short, yes! The best practices for SEO continue to be relevant because our generative AI features on Google Search are rooted in our core Search ranking and quality systems.

In other words, the generative features are not a separate channel with their own access rules: they rest on the same ranking and quality systems as ordinary results.

That sentence settles a question that had been hanging for two years, and it opens another: if nothing new is required, what exactly was being sold under the name GEO?

What the document says not to do

This is the most useful part, and the one no summary in the sector will willingly reproduce. The document names four practices and sets them aside one by one.

Files aimed at AI

The document writes that no machine-readable file, AI text file, markup or Markdown is needed to appear in search, generative features included, because search itself does not use them.

officialSource : Google Search Central, “Google's Guide to Optimizing for Generative AI Features on Google Search”

The text is explicit:

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.

This site publishes a file of that kind, and it will keep publishing it. Consistency demands saying why rather than quietly removing it: it costs nothing to produce, it documents the site’s structure for a human reader, and no benefit is attributed to it here. Publishing it in the hope of an effect on this engine would now be an untenable position; publishing it knowing it produces none is a defensible choice. The difference is total, and it lies in what you say about it.

Chunking the content

The document writes that there is no requirement to break content into tiny pieces for an AI to understand it better, nor to write in a particular way aimed at generative search.

officialSource : Google Search Central, “Google's Guide to Optimizing for Generative AI Features on Google Search”

The document writes:

There’s no requirement to break your content into tiny pieces for AI to better understand it.

And it adds, about the writing itself:

You don’t need to write in a specific way just for generative AI search. AI systems can understand synonyms and general meanings.

Those two sentences remove a whole layer of services: rewriting “for AI”, splitting into two-sentence blocks, phrasings supposedly easier to extract. Clear text remains clear text, and that was already the advice before generated answers arrived.

Structured data

The document writes that structured data is not required for generative search, and that no particular schema.org markup needs adding for it.

officialSource : Google Search Central, “Google's Guide to Optimizing for Generative AI Features on Google Search”

Structured data isn’t required for generative AI search, and there’s no special schema.org markup you need to add.

The point deserves reading precisely, because it lends itself to a symmetrical misreading: the document does not say structured data is useless, it says it is not required for that particular feature. Its documented role remains eligibility for certain rich displays in the classic results, and that role does not change.

What the document does not say

Three silences count as much as the four dismissals.

It does not say how an answer is composed. Knowing that the generative features rest on the ranking systems does not say which pages are kept to build an answer, in what order, or why some are cited and others are not. The question of “how to be cited” stays entirely open.

It says nothing about other systems.

A document published by an engine describes what that engine does, and nothing else: it does not say how a third-party conversational assistant selects its sources, and so cannot settle the question for every answering system.

hypothesis

That is the limit the market will hurry to ignore in both directions: some will conclude that GEO is dead everywhere, others that this document concerns one engine and that business can continue as before elsewhere. Both go beyond what is written. A document published by a vendor describes what that vendor does — that is the rule applied to every page on this site, and it holds here too.

It promises no result. Following this guide in full commits the engine to nothing: the generative features display what they display, and no compliance guarantees appearing there. A document that reframes practices is not a contract.

What this changes about what you were doing

For a properly kept site, nothing, and that is the most interesting result of this record. The document’s recommendations are, in order: produce content that brings something not found elsewhere, keep a clean technical structure — crawlable pages, semantic HTML, accessible rendering, duplication under control — and keep your information up to date.

For a site that had invested in AI-specific services, this document is an invoice. It does not say those services did harm; it says the engine does not use them. The distinction matters for judging a supplier acting in good faith: selling in 2024 a practice nobody knew anything about was not a fault, continuing to sell it after this document would be.

What does not work

Add an llms.txt file and specific markup so that AI understands your content and cites you in its answers.

The engine writes in plain words that it does not use those files, and that no particular markup is required for its generative features. The promise is therefore unverifiable on this engine and undocumented elsewhere, since no other system has published an equivalent position. The real cost is not the file, which is trivial: it is the budget and attention diverted from the one factor the document places first, the content's own worth.

What would have to be measured to settle the rest

The document closes the question of files and markup. It leaves open the question of source selection, and that is where the only thing a publisher cares about is decided: being picked up.

An honest protocol would require tracking, across a corpus of comparable pages, which ones are cited in generated answers and which are not, at identical search intent — and keeping that record long enough to tell a trend from chance. This site has run no measurement of that kind, and will publish none until it has a replayable protocol to give with it.

The performance report for generative features announced in the engine’s console in June 2026 is the first official instrument on this ground. It has not been recorded here: the post announcing it could not be read in full at the time of this record, and its exact content is therefore not reported. That will be the subject of a separate entry, or of none.

Sources

  1. Google's Guide to Optimizing for Generative AI Features on Google Search — Google Search Centralaccessed on
  2. A new resource for optimizing for generative AI in Google Search — Google Search Central Blogaccessed on
  3. Latest Google Search Documentation Updates — Google Search Centralaccessed on

Block revalidated on

Check it yourself

Observe that the generative features' robots reach your pages like any crawler, and that nothing else is required to be eligible.

With Lighthouse.

  1. Open your robots.txt file and check that no rule disallows crawling the pages you want cited.
  2. Audit one of those pages with the browser's tool and read the crawling checks panel: the page must be announced as crawlable and indexable.
  3. Display the rendered source and check that the main text is there with no extra script running.
  4. Remove, if one exists, any file aimed specifically at AI, then rebuild: no engine check should change state.

What you should observe: None of the four points depends on an AI-specific file or markup. If your page is crawlable and readable, it is eligible; if it is not, no extra file will change that.

Frequently asked questions

Should you publish an llms.txt file?
The document answers no for this particular search engine: no machine-readable file is needed, since search does not use them. It says nothing about other answering systems, which have published no equivalent position. Publishing one costs almost nothing and proves nothing.
Does structured data become useless?
No, but its role is shifted. The document says it is not required for generative search. It remains the means of obtaining certain rich displays in the classic results, which is a sufficient reason to keep it correct, and a different reason.
What is left to do, concretely?
Exactly what this site recommended before: content that brings something original, pages that are crawlable and readable without technical acrobatics, and up-to-date information. The document ranks those three first, before any generative consideration.
Does this document apply to ChatGPT or Perplexity?
No, and that is the limit to keep in mind. It describes what one engine does with its own generative features. Third-party conversational assistants select their sources differently and have published no equivalent document, which leaves the question open for them.
Why does this guide arrive now?
The engine announced it in May 2026 and updated it in July. The gap between what the market had been selling for two years and what the engine had never confirmed had grown considerable, and a framing document was the only way to reduce it.