Does llms.txt Help With ChatGPT? What the Evidence Says

This page answers one question: does putting an llms.txt file on your site change whether ChatGPT mentions or cites you? It does not explain how to write the file. The format (the H1 title, the one-line summary, the ## Section headings with `title: description` links), the common errors and llms-full.txt are covered in our llms.txt format guide. Read that one if you have already decided to publish a file.

If you are deciding whether the file deserves a place on your ChatGPT to-do list at all, read on. The short version: it ranks far below crawler access, clear pages and third-party listings. The full order of work is on the ChatGPT SEO overview.

Nothing that ties it to ChatGPT. We re-read OpenAI's crawler documentation and its web search guide on 2026-11-09. The crawler page describes three bots (OAI-SearchBot for ChatGPT search, GPTBot for training and ChatGPT-User for pages a user asks ChatGPT to open), how to allow or block each one in robots.txt, and where OpenAI publishes their IP ranges. Neither page says that any OpenAI bot fetches or reads a site's llms.txt, or that the file affects which pages ChatGPT cites.

One detail causes confusion. OpenAI's developer documentation site publishes its own llms.txt, and each docs page points to it as "the complete documentation index". That is OpenAI offering its documentation to coding tools and agents. It is not a statement that ChatGPT search reads other sites' llms.txt files, and we found no OpenAI page that makes that statement.

So the documented way to be readable by ChatGPT search is the same as before: let OAI-SearchBot in. What each OpenAI bot does explains the three tokens and which robots.txt rule to check.

The largest public study we know of is SE Ranking's, published 2026-10-15. It looked at nearly 300,000 domains, found an llms.txt on about one in ten, and found no correlation between having the file and how often AI engines cited the domain. When the researchers removed the llms.txt variable from their prediction model, the model became more accurate, not less. The study covers AI engines generally, not ChatGPT alone.

Google is the only large search company that has said something direct: its AI optimization guide states that Google Search does not use llms.txt and ignores it. That is about Google, not OpenAI, but it shows how the large engines treat the file today: as optional, and not as a ranking input.

On 2026-10-25 Leadee's tracker measured the 7 websites of Netbase JSC, the company behind Leadee, on three engines: Perplexity sonar, Google AI Mode and OpenAI gpt-5 with web search through the official API. Each per-brand analysis lists the ten domains cited most. Together those lists hold 59 domains; 7 of them are Netbase-owned (part of Netbase JSC), which leaves 52 third-party domains. On 2026-11-09 we requested /llms.txt from each of them once. A file counted as present only if it returned HTTP 200 and the body was Markdown, not an HTML error page.

One caution before the numbers. The citation counts are pooled across all three engines: our analyses do not split cited domains by engine, so these are not ChatGPT-only counts. The OpenAI part of that round was small (25 runs before the account ran out of credit).

Of the 52 domains, 25 had an llms.txt, 20 did not and 7 could not be read from our server (HTTP 403 or 500). Domains with a file collected 940 citations, those without 530. That looks like a difference until you look inside it: apps.shopify.com (164) and designnbuy.com (112) carry a large part of the first group. Leave those two out and the average is 28.9 citations per domain with a file against 26.5 without, close enough that the type of site matters far more than the file.

Several domains were cited often with no llms.txt at all: woocommerce.com (103 citations), fancyproductdesigner.com (36), infigo.net (34), youtube.com (29), linkedin.com (27) and dnschecker.org (25). Among those with a file, only 10 of the 25 passed Leadee's validator with zero format problems, so a well-formed file is not what those domains have in common either.

All 8 Netbase JSC sites (netbasejsc.com, cmsmart.net, printcart.com, softwsp.com, web2print.solutions, adminkit.net, outsourcingvn.com and leadee.ai, all Netbase-owned, part of Netbase JSC) serve an llms.txt. They were added during the October GEO work, which is why we left them out of the third-party count above.

Having the file is not the same as having a good one. cmsmart.net's file (Netbase JSC) showed 48 format problems in the validator, mostly "Key facts" and "Contact" lines written as plain text instead of `title` links. We have not re-measured whether any of the 8 sites is cited more often since the files went up, so we make no claim either way. In the 25 OpenAI runs of the October round, a Netbase brand was named in 3 (an indicator only; see what ChatGPT citations are).

Add one if it is cheap for you and you want any tool or agent that looks for the file to find your key pages. Do not add one expecting ChatGPT to mention you more, and do not spend time on it before the things that are documented to matter.

In order, those are: OAI-SearchBot allowed in robots.txt and not blocked by a firewall; pages whose passages answer buyer questions plainly (how to write pages ChatGPT can quote); accurate company facts and markup (schema markup that helps AI understand your company); and presence on the third-party pages that get cited for your topic. How ChatGPT's approach differs from classic search work is covered in ChatGPT SEO vs Google SEO.

If you publish a file, keep it honest and current: the same titles and descriptions as the pages themselves, no keyword lists, and no pages you would not want a buyer to land on. The ChatGPT llms.txt Generator drafts one that way and re-validates it; to check a file you already have, use the Validate tab of the general llms.txt generator.

Only by measuring the same buyer questions before and after, over several weeks, with enough runs to see past the noise. One ChatGPT answer is one sample; the same question can name different companies each time. How to check if ChatGPT mentions your brand shows how to set that up without a paid tool.

Server logs give a second, weaker signal: whether a request for /llms.txt came from an OpenAI IP range at all. A request with an OpenAI user agent is not proof, because anyone can send that header. Finding ChatGPT bots in your server logs explains how to verify the IP. Even a verified fetch would only show the file was read, not that it changed an answer.