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.
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.
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 guide | What it requires |
|---|---|
| Technical requirements— Same as Search | Indexability, snippet eligibility, spam compliance |
| Structured data | Good idea generally, not required for AI features |
| Special markup files (llms.txt)— Industry myth | Not supported, not required |
| Chunking content for retrieval— Industry myth | Can be ignored |
| Rewriting for AI systems— Industry myth | Can be ignored |
| Inauthentic mentions across the web— Industry myth | Can be ignored |
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 tactic | Google's stance (May 2026) | What vendors often promise |
|---|---|---|
| Add llms.txt to root | Not supported, ignore | "AI agents need it" |
| Chunk content into 200-word blocks | Ignore | "RAG-friendly content" |
| Rewrite copy in question-answer form for LLMs | Ignore | "AI-first writing" |
| Buy mentions on AI-relevant sites | Ignore (inauthentic mentions) | "Brand seeding for AI" |
| Make pages indexable and useful | Required, primary signal | Underrated |
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 for | Status | Notes |
|---|---|---|
| Indexable, snippet-eligible pages | Required | Primary signal |
| Compliance with spam policies | Required | Same as Search |
| Unique, useful, original content | Required | Helpful Content carryover |
| Firsthand reviews / expert insight | Recommended | Differentiator vs commodity |
| Structured data | Good idea, not required for AI | Rich result eligibility |
| JavaScript SEO basics | Where applicable | RAG needs crawlable HTML |
Chapter 04
Three industry myths Google's own document refutes
Chapter 05
What to do this week
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.
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
Read-only access. We never write to your site.