llms.txt in 2026: the file Google ignores

This article separates what Google states explicitly from what LinkedIn threads repeat. We tested llms.txt on several client sites, and the result matches the official position exactly: nothing changes on Google. Here are the facts, the figures, and the decisions we make.

No. Google's AI guide, published on May 15, 2026, states this in its mythbusting section: Google Search does not use llms.txt, either as a positive signal or as grounds for a penalty. The file does not help a site appear in AI Overviews and offers no protection. On Google, it is neutral.

The confusion comes from an appealing analogy with robots.txt. The idea is a text file at the root that lists your content for language models. The format exists, websites have deployed it, and some third-party crawlers read it. But Google has settled the question: its crawler does not take it into account at all. The same guide also rejects three other myths: there is no need to break content into chunks (content chunking), no requirement to write specifically for AI, and structured data is not required for generative search — useful for conventional rich results, yes, but not an AI prerequisite.

Google's message has been consistent: optimizing for generative AI means optimizing for search — in other words, SEO. Search Engine Journal sums it up this way: AEO and GEO are still SEO. No separate discipline, no magic file. What matters has not changed.

In practice, the factors that matter are the ones we have always worked on: useful, original content, a readable heading hierarchy, coherent internal linking, fast mobile pages, and real E-E-A-T signals (an identified author, practical experience, sources). Google's E-E-A-T guide remains the reference. In our audits, a page that answers a question precisely, cites its sources, and has a sound structure gains both conventional visibility AND AI citations — without llms.txt. For a clear distinction, see our comparison of SEO, AEO, and GEO.

Because the format is not useless everywhere — it is useless on Google. Some third-party crawlers and tools read llms.txt. But no public evidence shows an effect on Google Search or AI Overviews. The risk is not the file: it is believing that it replaces the underlying work.

An llms.txt file costs almost nothing: a few minutes to list your key URLs. If your audience uses tools that consume it, why not. What we reject is the reversal of priorities: spending a day polishing a file Google ignores while content remains thin and Core Web Vitals are in the red. Our rule at Pharelia: the file comes last, never first.

There is nothing specific to llms.txt here, but two major updates are a reminder of where the real risks lie. The May 2026 core update ended on June 2 after a rollout lasting around twelve days. The June 2026 spam update launched on June 24, worldwide and across all languages.

The Quality Rater Guidelines (September 2025 version, around 182 pages, used by nearly 16,000 raters) define three spam categories: Expired Domain Abuse, Site Reputation Abuse, and Scaled Content Abuse, plus a section on content with little effort and little originality. Google's position, expressed by Danny Sullivan, is clear: quality is judged, not the production method. There is no blanket penalty against AI — but mass-produced, low-value content receives the lowest rating. That is where an llms.txt file will never save you.

Google announced the launch of AI Overviews and AI Mode in France on July 22, 2026. AI Overviews appears when Google considers a generated answer useful for the query; it does not replace all conventional results.

Search Console provides a generative AI performance report for AI Overviews and AI Mode impressions. Google reports a global rollout as of August 31, 2026, with access limitations depending on the property and impression volume. This data is also included in the Web report: do not add it to that report's total. In GA4, check the AI Assistant channel and source/medium data for identifiable sessions. Traffic with no identifiable source may appear as Direct — a custom channel group does not reconstruct a lost source. Do not therefore reattribute all Direct traffic to AI. An llms.txt file measures neither impressions nor conversions. Our dedicated overview explains the monitoring method.

The decline is real for informational queries, but AI traffic converts better. Ahrefs measures a 15–30% drop in CTR where AI Overviews has been deployed, with around 69% of queries generating no clicks. Meanwhile, traffic from AI Search converts at 14.2%, compared with 2.8% for organic traffic (Seer Interactive).

The other reality is the fragmentation of citations. AI engines do not point to the same pages as Google. According to Higoodie (2026), ChatGPT citations overlap with only around 12% of Google's SERP; nearly 80% of LLM citations do not appear in Google's top 100 (Ahrefs, 15,000 queries); and only 11% of domains are cited by both ChatGPT and Perplexity. Stability is deteriorating further: the share of AI Overviews citations drawn from Google's top 10 fell from around 76% in mid-2025 to 38% in early 2026. In other words, ranking first on Google no longer guarantees an AI citation. ChatGPT alone accounts for around 87% of AI referral traffic (Conductor 2026), with Perplexity in third place (~7.7%, Statcounter May 2026).

Forget the file; work on the substance and distribution. Here is the order we follow in production, from the most worthwhile work to the least essential. None of these factors depends on llms.txt — they depend on quality, structure, and a presence where LLMs look for information.

One distribution channel deserves a mention: Reddit is disproportionately cited by LLMs because crawlers can retrieve its full text. A useful, non-promotional presence in the right subreddits is a more effective GEO tactic than any file at the root. To prioritize all of this, our SEO audit checklist gives the exact order — and our services explain how we put it into practice.