Skip to content

Official guidance · Primary source · 2026

Google told you what works. Most of the industry is ignoring it.

In May 2026 Google published a dedicated guide to optimizing for its generative AI features, following its December 2025 "AI features and your website" page. It explicitly tells site owners to ignore four tactics the industry is selling. Here's what the document actually says — and why the boring advice is the only advice that holds up.

PublishedMay 16, 202613 min read

The short version

On May 15, 2026 Google published a dedicated guide to optimizing for generative AI features, and it names four things site owners can ignore: chunking content into small pieces, rewriting content specifically for AI systems, creating AI-specific files such as llms.txt, and pursuing inauthentic mentions across the web. What it says does work is what Google has asked for since the Helpful Content guidance: be indexable, meet snippet eligibility, and publish unique, useful content with original perspective. Structured data "remains a good idea as part of your overall SEO strategy" but "isn't required specifically for AI features." The guide has roughly tripled in length since launch, absorbing guidance on AI agents and on local, shopping, image and video content, without softening any of that.

Official
May 2026
Google publishes its dedicated guide to optimizing for generative AI features
Google Search Central · primary source
4
tactics Google explicitly says you can ignore: chunking, AI rewrites, AI-specific files, inauthentic mentions
Same document
0
new technical requirements introduced for AI Overviews or AI Mode
Same document
1
real mechanism: be indexable, satisfy real users, write with original perspective
Same document

Chapter 01

What the document actually is

On May 15, 2026 Google Search Central published "Optimizing your website for generative AI features on Google Search" under a new Generative AI fundamentals section — a follow-up to "AI features and your website," which carries a December 2025 date. It's not a blog post. It's not a Twitter thread from a search advocate. It's official documentation, written in the same register as the rest of Search Central, and it's the closest thing to a primary source on what Google AI Overviews and AI Mode reward.

The document launched short and has roughly tripled since — it now runs to several thousand words, last updated July 10, 2026, having absorbed guidance on AI agents, on local, shopping, image and video content, and an explicit note that AEO and GEO are new names for SEO. What has not changed through the expansion is the register: it promises no ranking advantage, and introduces no new file format, no new schema, no sitemap variant. It restates, in unusually direct language, that the same indexability and quality requirements that govern Google Search govern its AI features too.

The reason this matters is that an entire "AI SEO" cottage industry has built up around the assumption that AI engines need to be optimized for differently. Vendors sell llms.txt audits, chunking services, and "LLM-readable" content rewrites. Google's official guide, in plain words, tells site owners to ignore all of these.

The document doesn't say these things are forbidden or penalized. It says they are not required, not useful, and not part of how Google generates AI Overviews. The recommendation to "ignore" them is direct, attributed, and quotable.

Section of the guideWhat it requires
Technical requirementsSame as SearchIndexability, snippet eligibility, spam compliance
Structured dataGood idea generally, not required for AI features
Special markup files (llms.txt)Industry mythNot supported, not required
Chunking content for retrievalIndustry mythCan be ignored
Rewriting for AI systemsIndustry mythCan be ignored
Inauthentic mentions across the webIndustry mythCan be ignored
Same
the technical bar for AI Overviews is identical to the standard Search snippet bar — Google states this explicitly
Google Search Central · May 2026
There is no AI-specific indexing pathway, no AI-specific crawler tier, no AI-specific markup. If you're already indexable and snippet-eligible, you're already eligible for AI features.

Chapter 02

What "ignore" means in Google's mouth

Documentation writers at Google are usually careful with action verbs. They prefer phrases like "we recommend," "consider," or "not required." The May 2026 guide is unusual in that it uses the word "ignore" and a blunt "Mythbusting" section. That word is doing real work.

The guide says you can ignore content chunking, ignore rewriting content just for AI systems, ignore the creation of AI-specific files, and ignore pursuing inauthentic mentions across the web. The phrasing matters because each of these is currently being sold as essential by some part of the SEO industry. Google's verb is corrective, not optional.

Read alongside other Google communications, the pattern is consistent. John Mueller has said no AI service has claimed to use llms.txt, pointing out that server logs show the crawlers do not even request the file, and Gary Illyes confirmed in July 2025 that Google does not support it and does not plan to. The AI Overviews team has stated repeatedly there is no special markup. The May 2026 guide is the first time these scattered signals appear in formal documentation.

Industry tacticGoogle's stance (May 2026)What vendors often promise
Add llms.txt to rootNot supported, ignore"AI agents need it"
Chunk content into 200-word blocksIgnore"RAG-friendly content"
Rewrite copy in question-answer form for LLMsIgnore"AI-first writing"
Buy mentions on AI-relevant sitesIgnore (inauthentic mentions)"Brand seeding for AI"
Make pages indexable and usefulRequired, primary signalUnderrated
"ignore"
the exact verb Google uses for chunking, AI rewrites, AI-specific files, and inauthentic mentions — direct and corrective
Google Search Central · May 2026
0
new files, new markup, new schemas introduced for AI features in the official guide
Same document
llms.txt
The clearest case study
llms.txt is a proposal by Jeremy Howard at Answer.AI from September 2024. It's an interesting idea. It is not a standard, it is not supported by Google, and there is no public confirmation that any major LLM provider uses it as a ranking signal. Despite this, it has become a default item on "AI SEO" checklists. Google's May 2026 guide is the official corrective — site owners can ignore it unless a specific partner integration requests it.
Google Search Central · May 2026

Chapter 03

What the guide says actually works

Strip away the negative space and the positive guidance is conventional, almost boring. Google asks for unique, compelling, useful content with fresh perspective. It asks for firsthand reviews and expert insights instead of summaries. It asks for editorial standards and a people-first approach. These are the same things Helpful Content guidance has asked for since 2022.

The technical requirements are the same: be indexable, be eligible for snippets, comply with spam policies, support modern page experience. JavaScript SEO matters where applicable. Duplicate content should be reduced. For local and ecommerce, Google Business Profiles and Merchant Center remain relevant.

Structured data is positioned carefully. The guide says structured data "isn't required for generative AI search, and there's no special schema.org markup you need to add," but that "it's a good idea to continue using it as part of your overall SEO strategy, as it helps with being eligible for rich results." Read against the Ahrefs causal study, this is consistent: schema can help eligibility for rich results in classic Search; it does not appear to lift AI citations.

The implicit claim is that AI Overviews and AI Mode are retrieval layers on top of Google's index, not separate systems with separate ranking criteria. If you make a page that satisfies real users, you've done the work. If you try to make a page that satisfies AI specifically, you're optimizing for a confounder, not a cause.

What Google asks forStatusNotes
Indexable, snippet-eligible pagesRequiredPrimary signal
Compliance with spam policiesRequiredSame as Search
Unique, useful, original contentRequiredHelpful Content carryover
Firsthand reviews / expert insightRecommendedDifferentiator vs commodity
Structured dataGood idea, not required for AIRich result eligibility
JavaScript SEO basicsWhere applicableRAG needs crawlable HTML
Same
the positive guidance is the same Google has asked for since the Helpful Content guidance in 2022
Cross-reference with Search Central history
Boring
the reason the industry resists this is that boring advice doesn't sell consulting hours
Our editorial take

Chapter 04

Three industry myths Google's own document refutes

Myth — AI engines need llms.txt to read your site
Google explicitly states no special markup file is needed. The crawlers powering AI features are the same crawlers indexing your site for Search. If a partner integration explicitly requests llms.txt, fine. Otherwise it's optional decoration.
Myth — You need to chunk content for RAG
Google does the chunking and retrieval. Your job is to publish coherent, high-quality content that a retrieval system can extract. Pre-chunked content is harder to read, harder to rank, and Google tells you to ignore the tactic.
Myth — There's a special way to write for AI
Google asks for writing that satisfies real users. The most effective causal levers measured in peer-reviewed research (Princeton GEO paper) are inline citations and statistics — both useful for human readers too. Writing "for AI" usually produces worse content for both audiences.

Chapter 05

What to do this week

01
Audit your site's indexability first
Per Google's guide, AI features inherit Search's indexability requirements. Fix robots.txt mistakes, broken canonicals, soft 404s, and noindex leaks before doing anything AI-specific. This alone fixes most AI invisibility.
Official guide
02
Drop llms.txt unless a partner asks
No major LLM provider has confirmed using it as a ranking signal. Google does not support it. Keep it only if a specific integration (custom GPT, partner agent) explicitly requests it.
Google + community
03
Stop the chunking and AI-rewrite experiments
Google says ignore both. The budget is better spent on original analysis and firsthand evidence — the strategies the Princeton GEO paper found lift visibility up to 40%.
GEO paper · KDD 2024
04
Audit content quality with Helpful Content lens
Since the AI guide doesn't change the quality bar, the existing Helpful Content guidance is the most direct checklist. Use it on every page you want cited.
Search Central history
05
Add inline citations and statistics
These are measured to lift visibility in AI answers (Princeton GEO study, up to 40%). They also satisfy the "firsthand insight" guidance Google asks for. Strong for both humans and engines.
GEO paper · KDD 2024
06
Stop buying "AI-relevant mentions"
Google explicitly lists "inauthentic mentions across the web" as a tactic to ignore. Coordinated mentions are the spam pattern the guidance is pointing at — and they're likely a net negative once detection catches up.
Official guide

Chapter 06

How That SEO Agent aligns with the official guide

This chapter used to be written in the conditional — "our analyzer should reflect that" — which is a promise, not a description. It is now in the indicative, because the code caught up.

Three checks decide whether Google can index the page, and they run first
Three checks that decide whether Google can index the page at all: Googlebot not blocked, an HTTP 200, and no noindex directive. Two are Google's own minimum technical requirements; the third is documented separately, and is just as decisive. Until recently the status was worth 3 points inside a category of thirty-odd checks, so a page Google cannot index came back graded like any other. It now runs before the score and leads the report, because until that is fixed the rest of the report is about a page nobody can find.
Tactics the guide tells you to ignore earn nothing, and say so
llms.txt is detected and reported as what it is: harmless to keep, worth no points, not read by Google Search. Nothing recommends creating one. The same now applies to the advice we used to give around chunking and AI-first writing — the guide names both as unnecessary, and we had been selling them with multipliers no AI platform has ever published.
Every check says whose rule it is
A finding backed by a Google sentence is shown plainly, because Google's rules are the baseline. A finding that is our own judgement is labelled as ours, in the tool output and in the shared report. Google's own guidance on third-party SEO tools warns that vendors do not have access to its ranking data — a report that cannot separate the two is making exactly the claim it warns about.
The one thing in the guide worth building: can an agent operate the page?
The guide says "browser agents may access your website to gather the data they need to complete these tasks, such as analyzing visual renderings (like screenshots), inspecting the DOM structure, and interpreting the accessibility tree". That is the only sentence in it naming a mechanism, and the accessibility tree is derived from the DOM by published rules — so it is checkable without a browser. We report controls with no accessible name, form fields with no label (a placeholder is not one), and a missing main landmark. It is the rare case where the accessibility fix and the AI-era fix are the same fix.
Primary sources
Official platform docs
Google Search Central — Optimizing your website for generative AI features (May 15, 2026)developers.google.com
Google Search Central — AI features and your website (Dec 10, 2025)developers.google.com
John Mueller — no AI system currently uses llms.txt (June 2025)seroundtable.com
Google Search Central — FAQPage structured data (May 2026 deprecation)developers.google.com
Jeremy Howard / Answer.AI — llms.txt proposal (Sep 3, 2024)answer.ai
Industry analysis
Ahrefs — Does schema markup help with AI citations? (causal study, 2026)ahrefs.com
Joost de Valk — The FAQ schema cycle (May 2026)joost.blog
Studies
Aggarwal et al. — GEO: Generative Engine Optimization (Princeton / IIT Delhi, KDD 2024)arxiv.org/abs/2311.09735

Run a Google-aligned audit

Score your site against the official guide, not the hype cycle.

That SEO Agent audits your site against the indexability and quality requirements Google actually documents — and flags the tactics Google explicitly tells you to ignore. Stop optimizing for files Google doesn't read.

  • Never writes to your site
  • Diagnostic, not automation
  • Connects via MCP to Claude, ChatGPT & Cursor
See pricing

Read-only access. We never write to your site.