Guide · llms.txt

Does llms.txt work? What we measured on our four sites

Updated 2026-11-12 · kagentx

Short answer

There is no data that proves llms.txt works. Of our four sites, three have one (MochiStamp 98,063 bytes and 250 entries, Gamma AI Slide 55,645 bytes, Pagent 36,321 bytes) and kagentx.com itself has none (404). In the last 28 days the three with a file got 86, 46 and 0 sessions from chatgpt.com, but there is no before-and-after comparison and no record of crawlers reading it. Google's guide page says Google Search does not use it, and OpenAI's and Perplexity's crawler docs do not say they read it. You can do it, but do not expect it to bring recommendations.

Do our four sites have an llms.txt, and how big is it?

On 2026-11-12 we read /llms.txt on four sites with curl. Three have one; kagentx.com itself does not:

Site/llms.txt statusSize (bytes)Number of itemsSitemap URLsShare of sitemap covered by entries
mochistamp.com20098,0632501,02025%
gammaslide.com20055,645130 (119 distinct URLs)15676%
pagent.kagentx.com20036,321120 lines (143 distinct URLs)18876%
kagentx.com404(none)(none)(not counted here)(none)

The ratio is distinct URLs in the file divided by URLs in the sitemap, a rough comparison only. Entries are counted as lines starting with "- ["; Pagent's leaderboard uses a different format, so distinct URLs are given as well. kagentx.com is the site where we sell diagnostics, and its own /llms.txt returns 404, so the "sites with an llms.txt" in this article are the other three.

What is in these three files?

SiteOpeningBodyNotable
MochiStampH1, an intro paragraph, one line with the home pages in seven languages7 H2s, one section per language, each with 32 to 41 entries of "title plus one sentence"Sectioned by language, every entry has one sentence of description
Gamma AI SlideH1, an intro paragraph and a pricing note8 H2s: what it does, questions, Arabic and Persian FAQs, pages, entry points in each language, guides (102 entries), recorded examplesMixes Q&A with links
PagentH1, an intro paragraph and one line of usage instructions addressed to AI5 H2s: how we test, leaderboard, reports (115 entries), data, how to get an assistant testedTells the AI up front how to cite

As you can see, the three sites share no format: one is sectioned by language, one mixes in Q&A, one opens with usage instructions for AI. MochiStamp's llms.txt, Gamma AI Slide's llms.txt and Pagent's llms.txt can all be opened and compared directly.

Do the official sources say llms.txt works?

The full table of official statements is in What is llms.txt, and does it work?. Here are only the three we re-read today (2026-11-12) by opening the pages:

WhoWhat the page saysPage and date
Google Search CentralYou can ignore llms.txt: Google Search itself does not use it; creating it for other services neither harms nor helps your visibility and ranking in Google SearchGuide to optimizing for generative AI features, page marked updated 2026-07-10
OpenAI and PerplexityBoth publish an llms.txt on their own docs sites (OpenAI 6,201 bytes, Perplexity 43,170 bytes), but their crawler docs do not say their search crawlers read other sites' llms.txtCrawler docs page, viewed 2026-11-12
Bingllms.txt does not appear on the webmaster guidelines pageBing Webmaster Guidelines, viewed 2026-11-12

We did not re-check Anthropic's crawler documentation today, so we do not state anything about it. Not stating it does not mean they do not read it, or that they do: the official source says nothing.

Can we prove it brings AI traffic?

No. Below are the AI-source sessions we can see in GA4 for 2026-10-14 to 11-10, and what we do not have:

SiteSessions, whole siteSessions from chatgpt.comHas llms.txt
MochiStamp59186Yes
Gamma AI Slide31946Yes
Pagent60Yes
kagentx1080None

This table cannot show whether llms.txt works or not, for four reasons:

  • There is no before-and-after comparison: we did not trace back the day each site's llms.txt went live.
  • The four sites differ in content, traffic scale and launch time, and among the three with an llms.txt is Pagent, with very little traffic.
  • We have no server-log records of AI crawlers fetching /llms.txt, so we do not know whether any crawler read it.
  • Search Console's page data has no impressions for /llms.txt, which is normal: it is not meant for Google Search.

Also, MochiStamp has another 19 sessions from openai / organic, and none of kagentx's 108 come from chatgpt.com, but kagentx's traffic is mostly direct visits and local test addresses, so its source mix is different by nature and it cannot serve as a comparison.

So should you do it anyway?

Our conclusion: you can do it, but do not expect it to bring recommendations. The cost is one file to maintain, the benefit has no public evidence so far, and Google says plainly it does not use it while the others say nothing. If you decide to do it, follow these steps:

  1. 1. Check that /llms.txt opens directly

    Status code 200, content type text (text/plain). Of our four sites, three return 200 and kagentx.com returns 404.

    curl -s -o /dev/null -w "%{http_code} %{content_type} %{size_download}\n" https://your-site.com/llms.txt
  2. 2. Write it as a summary plus links for AI

    Open with a paragraph on what you are and who it is for; then list links by topic, one sentence each. You can draft it with the llms.txt generator.

  3. 3. Update it whenever pages change

    When prices, features or page URLs change, change the file too, or you are handing AI an outdated summary.

  4. 4. Test whether anyone reads it

    Search your server logs for AI crawlers requesting /llms.txt; see How do I check whether GPTBot and other AI crawlers have visited my site?.

    grep "llms.txt" access.log | grep -iE "GPTBot|OAI-SearchBot|ClaudeBot|PerplexityBot"
  5. 5. Watch AI sources separately in GA4

    That is how you know how many sessions AI sends and whether they are growing; see How do you see ChatGPT traffic in GA4?.

What are the limits of these conclusions?

  • We only looked at our own four sites, and the three with a file have no "without a file" comparison.
  • We have no server-log data, so we did not test whether AI crawlers read it.
  • We only read the official pages that opened today, and did not re-check Anthropic.
  • The traffic figures are GA4 source sessions, not users, and we did not tell whether chatgpt.com sessions came from clicking a link in an answer or another path.

FAQ

Will llms.txt make ChatGPT recommend my site more?

No evidence. OpenAI's crawler docs do not say its search crawlers read other sites' llms.txt, and our three sites with a file have no before-and-after comparison.

Does Google use llms.txt?

Google's guide to generative AI optimization (marked updated 2026-07-10) says Google Search itself does not use it and that it will not change visibility or ranking.

Can llms.txt hurt my SEO?

Google's guide says creating it for other services neither harms nor helps your performance in Google Search.

How big should llms.txt be?

The largest of our four sites is 98,063 bytes and 250 entries, the smallest, Pagent, is 36,321 bytes. We found no official size limit, so we give no number; we suggest listing only the most important pages.

How do I know whether AI crawlers read my llms.txt?

Look in your server logs for AI crawlers requesting /llms.txt. We do not have that log, so we did not test it.

Why does kagentx itself have no llms.txt?

On 2026-11-12 we found kagentx.com/llms.txt returning 404. This article only records the fact and makes no judgment about the reason.

Do sites with an llms.txt always get more ChatGPT sessions?

Not necessarily: MochiStamp 86, Gamma AI Slide 46, Pagent 0. The three sites differ in traffic scale to begin with, so it cannot be read as cause and effect.

Sources

kagentx

Do AI answers
mention you?

An evaluation asks ChatGPT, Perplexity and Claude your buyers' questions (with web search on) and records who they recommend, which pages they cite and whether you are among them. Basic $1, results usually in 2–3 minutes.