Do you need an llms.txt file to get cited by AI? What 300,000 domains say

Here is the pitch you have probably heard: ship an llms.txt file, and the AI engines will finally read your site properly and start citing you. It sounds right. We already give crawlers a robots.txt and a sitemap.xml, so a file that hands models a clean map of your best pages feels like the obvious next step. A wave of vendors and checklists in 2026 treat it as table stakes.

The problem is that the data does not back the promise. And since the whole point of GEO is to earn citations, a tactic that does not move citations is worth being honest about before you spend a sprint on it.

What llms.txt actually is

llms.txt is a proposed standard, first floated in 2024, for a plain-text file at your site root (yoursite.com/llms.txt) that lists your key pages in simple markdown so a language model can find and read the content that matters. The intent is good. The gap is that, unlike robots.txt, it has no enforcement, no standardization body behind it, and crucially no adoption from the companies that run the engines.

What the 2026 data shows

Several large studies landed this year, and they point the same direction:

• No correlation across 300,000 domains. An analysis by SE Ranking found no measurable relationship between having an llms.txt file and how often a domain is cited in AI answers. Tellingly, their citation-prediction model got better when the llms.txt variable was removed, meaning the file behaved like noise, not signal.
• Almost nothing reads it. A large Ahrefs study reported that the overwhelming majority of llms.txt files received no meaningful crawler traffic. In one separate 90-day server test, roughly 0.1% of all AI-bot requests touched the file at all.
• Statistically indistinguishable from noise. One analysis of hundreds of thousands of citations put the effect at a p-value around 0.85, which is about as close to "no effect" as these things get.

Figures vary by study and method, so read them as directional rather than exact. But no serious 2026 dataset shows llms.txt moving citations, and that consistency is the point.

The sources AI engines actually cite are pages, profiles and third-party mentions, not a file at your domain root. Illustrative data.

Why it does not move the needle

It helps to compare it to the files it is modeled on. robots.txt works because every major crawler agreed to obey it and does. llms.txt has neither the agreement nor the obedience. Google has been explicit: Gary Illyes said in 2025 that Google does not support it and has no plans to, John Mueller likened it to the long-discredited keywords meta tag, and Google's own guidance updated in mid-2026 states that Search ignores llms.txt and that you do not need to create new files of this kind. None of OpenAI, Anthropic, Meta or Mistral has committed to honoring it for their answer surfaces either.

That leaves the mechanism. AI engines decide who to cite based on the quality and clarity of the actual pages they retrieve, the third-party mentions and links that corroborate you, and your entity authority on the topic. A map file at your root changes none of those. It is a signpost to a house the engine was never going to visit on the strength of the sign.

llms.txt is a signpost. Citations are earned inside the house, by the pages and the mentions, not by the sign at the gate.

When it is still worth shipping

Here is the nuance the "it is dead" takes miss. llms.txt is not useless, it is just aimed at the wrong target for GEO. The tools that do fetch it are agent and developer tools: coding assistants like Cursor, Claude Code and Copilot, MCP servers, and a growing set of in-product AI assistants that navigate documentation. If you serve developers, sell an API, or want your docs to be agent-readable, an llms.txt file is a reasonable, cheap thing to publish. It typically costs about half a day and it does no harm.

So the honest verdict is not "never do it." It is: do it for agent-readability if that matters to you, and do not expect a single extra AI search citation from it. Do not let shipping a file feel like you have done your GEO work, because the work that earns citations is still waiting.

The real lesson: test tactics, do not trust them

llms.txt is a useful case study in a bigger trap. GEO is young enough that a lot of advice is confident and untested, and the tactics that sound most intuitive are not always the ones that move outcomes. Schema markup, freshness passes, a new comparison page, a round of digital PR, an llms.txt file: each one is a hypothesis about what will earn you more citations, not a guarantee.

The only way to tell the winners from the busywork is to measure the outcome you actually care about, which is whether AI engines cite you more after the change than before. Do that, and llms.txt stops being a debate you argue about and becomes a thirty-minute experiment you can settle with your own data.

How llemmy helps

llemmy tracks the answers AI engines actually give about your brand, across ChatGPT, Claude, Gemini, Perplexity and Google AI, for the questions your buyers ask, over a rolling window. So when you ship a change, an llms.txt file, a new page, a schema pass, a PR push, you can watch whether your citation share and visibility actually move, with a sample size and a 95% confidence interval on every read so you are reading signal and not a lucky week. It turns every GEO tactic into something you can test instead of take on faith. Run a free GEO audit to see where you stand today, or start tracking free and settle the next "should we do X" with evidence.

FAQ

What is an llms.txt file?

It is a proposed plain-text file at your site root (yoursite.com/llms.txt) that lists your key pages in a simple markdown map so an AI model can find and read your most important content. It was proposed in 2024 by analogy to robots.txt and sitemap.xml. Unlike those two, it has no adoption from the major AI companies and no engine is known to require it.

Does llms.txt help you get cited by AI?

The 2026 evidence says no, not measurably. A 300,000-domain analysis by SE Ranking found no correlation between having the file and how often a domain is cited, and its citation model improved when the llms.txt signal was removed. Ahrefs found most files get effectively no crawler traffic, and Google has said Search ignores llms.txt. Citations are driven by page quality, third-party mentions and entity authority, not a root file.

Should I still create an llms.txt file?

It can be worth it for a different reason. Developer and agent tools such as coding assistants, MCP servers and some in-product AI assistants do fetch llms.txt to navigate docs. If you serve developers or want agent-readable content, shipping one is cheap and harmless. Just do not expect it to move your AI search citations.

How do I know if a GEO tactic actually works?

Measure your citations before and after the change. Track the answers AI engines give about you across engines over a rolling window, ship the change, and watch whether your citation share moves, with a sample size and a confidence interval so you are reading signal and not noise. That turns every tactic, llms.txt included, into a testable hypothesis.

By the llemmy team, July 2026. Grounded in 2026 reporting and studies on llms.txt, including SE Ranking's 300,000-domain citation analysis, Ahrefs' llms.txt traffic study, and public statements from Google's Gary Illyes and John Mueller, whose figures vary by source and method and should be read as directional. Related reading: What AI engines actually cite, What makes a page AI-readable, and How to measure content effectiveness in AI search.