Google Says llms.txt and Chunking Don't Help AI Search Visibility

Google has updated its official guide on optimising for generative AI search, stating plainly that llms.txt files, content "chunking" and bespoke AI schema do nothing for visibility in AI Overviews or AI Mode. The refreshed guidance says AEO and GEO aren't separate disciplines from SEO, just new labels for the same work. If your business is paying for these tactics, it's a clear signal to redirect that budget.

What changed?

Google's developer documentation page, "Optimising your website for generative AI features on Google Search," was first published on 15 May 2026 and has now been refreshed with more detail, according to Search Engine Journal. The guidance is explicit: for Google Search, you can ignore tactics like "chunking" content, creating unnecessary AI text files such as llms.txt, or pursuing inauthentic mentions.

The page sits inside a "Mythbusting generative AI search" section that names specific tactics directly, something Google's guide has rarely done before. Google states that, from its perspective, optimising for generative AI search is optimising for the search experience, and is therefore still SEO. This echoes positions Google employees have taken at conferences, but now it's in published documentation as an official reference to cite, as Semrush notes.

On llms.txt specifically, Google says Search doesn't use machine-readable files, AI text files, markup or Markdown to determine visibility, and adding one will neither harm nor help a site's rankings, since Google Search simply ignores them. On chunking, the guide is equally direct: there's no need to break content into tiny pieces, because Google's systems can already understand nuance across multiple topics on a page and surface the relevant part to users.

Independent data broadly backs this up. A 90-day log study by OtterlyAI, cited by CXL, found that only 0.1% of AI crawler requests ever touched /llms.txt, with most crawlers ignoring the file entirely. Separately, Ahrefs tracked 1,885 pages that added JSON-LD schema against a matched control group and found no citation lift across AI Overviews, AI Mode or ChatGPT, per the same CXL analysis.

Not everything has landed cleanly. Google's warning against "inauthentic" mentions has drawn pushback from well-known SEO voices, with Lily Ray calling it "classic Googlespeak" on X, arguing the language is vague enough to mean whatever Google needs it to mean later, as flagged by LinkedIn coverage of the guide.

Why does this matter for your business?

An entire cottage industry has grown around "AI search optimisation", selling llms.txt generation, chunking services and bespoke AI schema as must-have infrastructure. Google's guidance suggests much of that spend, for the Google channel specifically, has been wasted.

That matters because Google still dominates how people find things through AI. Nearly 40% of Google's AI Overviews rank in the top 10 organic search results, and nearly 70% rank in the top 100, according to CXL. That means the foundation for showing up in Google's AI features is still classic ranking performance, not a separate AI-specific playbook.

The guide also reinforces why grounding matters. Google's generative AI features rely on retrieval-augmented generation, a technique that improves accuracy and freshness by pulling relevant, up-to-date pages from Google's Search index and generating a response with clickable links, as described in Google's own guide. If your page isn't retrievable through normal indexing, no amount of AI-specific markup will fix that.

However, the guidance is scoped narrowly to Google. It doesn't necessarily apply to how ChatGPT, Perplexity or other assistants source information, and that distinction matters given how much traffic those platforms now send. ChatGPT led the AI chatbot market as of July 2026 with 53.9% of worldwide web visits across the seven largest generative AI chatbots, ahead of Google Gemini at 27.9% and Anthropic's Claude at 9.2%, according to Momentic. A tactic Google says is pointless might still carry weight elsewhere, even if the evidence for that is thin so far.

What should you do now?

Stop paying for llms.txt generation, chunking services or bespoke AI schema as if they were guaranteed levers for Google visibility. The data and Google's own documentation agree these are largely wasted effort for that channel.

Instead, focus on fundamentals Google explicitly says still matter: solid technical SEO, clean indexing and genuinely useful content. Google draws a sharp line between generic "commodity content" and original, first-hand material, contrasting a generic listicle with a non-commodity alternative such as "Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line", where the distinction is whether the content offers unique insight beyond common knowledge, per Google's guide.

Don't treat Google's word as the whole picture, though. As one analysis from Jasper put it, core SEO fundamentals apply everywhere, but the off-site visibility layer that Google's guide de-emphasises is likely more important for AI platforms that aren't anchored to Google's own index. That means checking your presence on the review sites, forums and comparison pages that ChatGPT and Perplexity actually cite is still worthwhile, even if it isn't what Google's guide is talking about.

Given how differently each platform sources and cites information, it's worth actually checking where your brand does and doesn't show up across these systems rather than assuming. Running a free audit, such as the one at Sited, is a quick way to see your current AI visibility before deciding where to spend time and budget.

Finally, keep watching for divergence between what Google's documentation says and what its own product teams do in practice, since even Google's teams haven't been fully aligned on every point of this guidance.

Frequently asked questions

Do I still need an llms.txt file?

Not for Google Search. Google Search doesn't use these files to determine visibility, and creating one will neither help nor harm your rankings there. It may still be worth maintaining one if you specifically want to communicate with other AI platforms that read it, though evidence on actual usage remains limited.

Is content chunking still worth doing for AI search?

According to Google, no. There's no requirement to break content into tiny pieces, since Google's systems can understand nuance across multiple topics on a page and surface the relevant part to users. Independent schema testing found similar results, with no measurable citation benefit from the tactic.

Are AEO and GEO different from SEO?

Not according to Google. From Google Search's perspective, optimising for generative AI search is optimising for the search experience, and is therefore still SEO. The terms describe a focus area rather than a separate technical discipline, at least as far as Google's own systems are concerned.

Does this guidance apply to ChatGPT and Perplexity too?

No, it's scoped specifically to Google Search. Other platforms have different retrieval and citation behaviours, and tactics Google dismisses may still carry some weight on platforms that don't share Google's index or ranking systems.

What should I actually prioritise instead?

Solid technical SEO, including crawlability, indexing and page experience, combined with original, first-hand content that goes beyond generic summaries. Google's own framing is to create content people find genuinely useful, since that is what its systems are ultimately trying to surface.