Does an llms.txt file get my company cited by AI engines in 2026?

Last updated October 14, 2026. Written by the Third Coast AI team.

No, not on any evidence available in October 2026: llms.txt is a proposed convention for a markdown file at the root of a site that summarizes the site for language models, none of the major answer engines has said it reads the file when choosing sources, Google has said in writing that no special AI text files are needed to appear in its AI features and a Google search advocate said plainly that "no AI system currently uses llms.txt," and the one independent factor analysis that scored it rated it 2.0 out of 10 as a citation factor. It is cheap to publish, it does no harm, and it can be useful for coding assistants reading documentation sites, which is a long way from the claim that it earns citations in ChatGPT or AI Overviews.

This page explains what the file is and where it came from, who honors it and who has said they do not, what the measurement says, and the position we take with clients, which is to publish one for the right reasons and to stop any vendor who sells it as an AI visibility lever. The schema page covers the adjacent question of structured data, where the evidence is better.

Is llms.txt a standard, and who proposed it?

It is a proposal rather than a standard, and an open one. The page at llmstxt.org, written by Jeremy Howard of Answer.AI and first published on September 3, 2024 with a revised version dated August 10, 2026, proposes "adding a /llms.txt markdown file to websites to provide LLM-friendly content," on the reasoning that "agents are best served by concise, expert-level information gathered in a single, accessible location" and that ordinary web pages are built for people, with navigation and scripts that make extraction harder. The proposal's own stated expectation is that the file "would mainly be useful for inference rather than training," meaning it is meant to help an agent that visits your site while answering, not to shape what a model learned.

None of that is objectionable, and the format is sensible for what it was designed to do, which is to help a coding assistant or an agent find the right documentation page quickly. The question a buyer is asking is different: whether publishing one makes ChatGPT, Google, Perplexity or Copilot more likely to cite the company. On that question the proposers make no claim, and the engines have answered.

Which engines honor llms.txt in 2026?

None of the major search-and-answer engines has said it uses the file to choose sources. Google's AI features documentation says "you don't need to create new machine readable files, AI text files, or markup to appear in these features," which is as close as a vendor comes to naming llms.txt without naming it, and as Search Engine Roundtable reported on June 18, 2025, Google's John Mueller wrote on Bluesky the day before that "FWIW no AI system currently uses llms.txt," adding that server logs make this evident because the model companies fetch pages and do not fetch the file. OpenAI's crawler documentation, Anthropic's crawler documentation and Perplexity's crawler documentation describe their agents, what each is for and how to block them, and none of the three mentions llms.txt at all.

The place the file demonstrably gets read is developer tooling. Documentation sites for software products publish llms.txt and llms-full.txt variants because coding assistants and agent frameworks fetch them when a developer asks about an API, and that is a real, if narrow, use, which is why software companies publish the file and why we do. For a manufacturer, a law firm, a swim school or a regional bank, no agent is arriving through that door.

Does the measurement say anything different?

No. Cyrus Shepard's May 2026 synthesis of 54 studies scores llms.txt at 2.0 out of 10 as a citation factor, the lowest tier in his analysis, against 9.5 for URL accessibility and 9.4 for search rank. Ahrefs measured 38% of AI Overview citations coming from the top ten in March 2026 and its 75,000-brand study found mentions on other sites correlating with AI citations three times as strongly as backlinks, and in neither study does a root text file appear as a variable because there is no mechanism by which it could. The Princeton researchers (Aggarwal and colleagues, KDD 2024) found quotations, statistics and cited sources lifting visibility in generated answers by roughly 41%, 33% and 28%, and every one of those levers lives on the page the engine retrieves, which is the page and not a summary of it.

Should you publish one anyway?

Yes, if it takes under an hour and you keep it accurate, and no, if anyone is charging you for it as an AI visibility service. We publish one on this site because it costs nothing, it describes the company in one place a machine can parse, and if an engine does start reading the file the companies with an accurate one will be ahead by a morning's work. We do not expect it to move a single citation, and we say so in every audit. The things that do move citations are the ones the engines and the studies describe: pages the crawlers can reach and read in plain HTML, an answer stated near the top with a number or a named source behind it, passages that cover the hidden sub-queries behind a buyer's question, and a third-party record that lets the engine trust what the page says.

• Write the file by hand: the H1, a two-sentence blockquote, and links to the ten pages that best explain what you do, each with a one-line note.
• Keep it consistent with your Organization schema and your profiles, since inconsistency is the only way the file can hurt.
• Do not generate it from the sitemap; a list of 400 URLs defeats the purpose the proposal describes.
• Spend the rest of the afternoon on the audit checklist instead, starting with crawler access and rendering.
• If a vendor lists llms.txt as a deliverable, ask them which engine reads it and for the page where that engine says so. The agency guide has more questions like that.

Frequently asked questions