llms.txt Is the Meta Keywords of 2026

If you have read an SEO newsletter in the past year, someone has told you that you need an llms.txt file or ChatGPT will not understand your business.

The short answer: llms.txt does nothing for your visibility in Google Search, AI Overviews, or AI Mode. Google has said so explicitly and repeatedly, and correlation studies find no relationship between having the file and being cited. It does have one legitimate use case, which is developer documentation consumed by coding agents. Anyone selling it as an AI visibility lever is selling the exact thing that makes GEO skeptics sound right.

This is the piece in this series where I side with the critics.

What Google actually said, and when

The record here is unusually clear, which is rare in this field.

Gary Illyes said at Search Central Live in July 2025 that Google does not support llms.txt and has no plans to. John Mueller was blunter, comparing the idea to the old meta keywords tag and noting that bots were not even requesting the file. Meta keywords is the self-declared signal search engines stopped trusting more than a decade ago because site owners could write whatever they wanted.

Then, on June 15, 2026, Google added a note to its AI optimization guide, logged in its changelog as “Clarifying guidance on llms.txt files”. The note states that the file is not needed to appear in Google Search and that keeping one will neither harm nor help visibility or rankings, because Google Search ignores it. The same guidance states plainly that you do not need to create new machine-readable files, AI text files, markup, or Markdown to appear in Google's AI features.

There was a brief moment of confusion in December 2025 when an llms.txt appeared on Google's own developer documentation. It was removed from the Search docs the same day, and Mueller later confirmed it was not an endorsement; Google's documentation platform had added the file automatically.

The meta keywords comparison is the right one, and it is worth sitting with. Meta keywords failed for a structural reason, not a technical one: any signal a site declares about itself, with no verification and no cost to exaggerate, gets gamed until it becomes worthless as evidence. llms.txt has the same shape. It is a self-description with no verification mechanism.

The evidence beyond Google

Google's self-interest is a fair objection to raise, since I have spent an entire article in this series arguing Google talks its book about GEO. So look at the independent data.

SE Ranking's correlation work found that having an llms.txt file did not predict AI citations. More tellingly, adoption was nearly identical across low, mid, and high-traffic sites. That second finding matters more than the first. If llms.txt were a lever that worked, you would expect high performers to adopt it at higher rates and others to copy them. Flat adoption across performance tiers is what a null result looks like in the wild.

As of Q1 2026, none of OpenAI, Google, Anthropic, Meta, or Perplexity had publicly confirmed that their production answer systems read or act on llms.txt when generating responses. Adoption sits somewhere around 10% of domains, which tells you about marketing advice, not about efficacy.

Where I break from the pure skeptics

Now the part the "llms.txt is useless" crowd gets wrong, because a balanced answer is more useful than a satisfying one.

The file is not inert everywhere. Profound's tracking data, as cited by Mintlify, showed crawlers from Microsoft and OpenAI fetching both llms.txt and llms-full.txt. Chrome's Lighthouse includes an agentic-browsing audit that checks the file. And much of the documented demand comes from coding tools and agents reading developer documentation.

That last point is the real use case. AI coding assistants like Cursor, GitHub Copilot, and Claude retrieve documentation in real time while a developer works. A clean, curated map of your docs genuinely helps an agent find the right page. Anthropic, Stripe, Cursor, and Cloudflare all publish one, and they are all documentation-heavy developer products. That is not a coincidence.

So the accurate framing in 2026 is narrower than either camp wants. llms.txt is not a search visibility lever. It is an agent-readiness signal for documentation, with real but limited consumers.

The decision, in one line each

Ship one if you run developer documentation or an API product and your users' coding agents fetch your docs. It is cheap, the consumers are real, and it may improve how accurately agents cite your docs.

Skip it if your goal is Google visibility, including AI Overviews and AI Mode. Google's documentation is explicit and the time is better spent on what Google does count.

Do not remove one you already have. Google's position is neutral, not hostile. It will not hurt you.

And do not pay for it. This is the part that matters. Adding llms.txt is a few minutes of work. If an agency proposal lists it as a deliverable, or a retainer leans on it as evidence of AI visibility work, you are being charged for a checkbox that Google publicly ignores. Ask instead what the engagement does for content depth, per-engine citation tracking, and the source relationships that actually drive inclusion.

Why I am writing this one

There is a version of this article that does not serve my interests. I build a GEO platform (disclosure: GrackerAI, my company, works with B2B SaaS), and the easy move for anyone in my position is to nod along with every technical ritual that makes the category feel sophisticated.

But the fastest way to prove the skeptics right is to let unfalsifiable tactics circulate unchallenged. Every time a vendor sells llms.txt as an AI visibility lever and the client sees no movement, the entire discipline loses a little credibility that the people doing real measurement work have to earn back.

The category has enough genuine substance to withstand losing its bad tactics. Passage-level structure, per-engine citation tracking, source graph analysis, entity corroboration: these survive scrutiny. A self-declared text file that no major answer engine confirms reading does not, and defending it costs more than dropping it.

Here is my question: what is on your current AI visibility checklist that you have never actually tested? For most teams there are two or three items on that list that are there because someone read a newsletter in 2025, and llms.txt is usually one of them.

Frequently Asked Questions

Does llms.txt help with Google Search or AI Overviews?

No. On June 15, 2026, Google logged a change called "Clarifying guidance on llms.txt files" and added a note to its AI optimization guide. The note states the file is not needed to appear in Google Search and that keeping one will neither harm nor help visibility or rankings, because Google Search ignores it. Gary Illyes said in July 2025 that Google does not support it, and John Mueller compared it to the meta keywords tag.

Is there any evidence llms.txt improves AI citations?

Independent correlation work from SE Ranking found that having an llms.txt file did not predict AI citations, and adoption was nearly identical across low, mid, and high-traffic sites, which is what a null result looks like. As of Q1 2026, none of OpenAI, Google, Anthropic, Meta, or Perplexity had publicly confirmed their production answer systems read the file when generating responses.

When is llms.txt actually worth creating?

When you run developer documentation or an API product. Much of the documented demand comes from coding tools and agents reading docs in real time, such as Cursor, GitHub Copilot, and Claude. Profound's data, cited by Mintlify, showed Microsoft and OpenAI crawlers fetching llms.txt and llms-full.txt, and Chrome's Lighthouse includes an agentic-browsing audit that checks it. Anthropic, Stripe, Cursor, and Cloudflare all publish one.

Should I remove an llms.txt file I already have?

No. Google's position is neutral rather than hostile, stating the file will neither harm nor help. There is no penalty for keeping it. The practical caution is commercial rather than technical: adding the file takes minutes, so it should never anchor an agency retainer or appear as significant evidence of AI visibility work.