# LLMSRank: is your site readable by AI? > Grade your llms.txt and AI-readiness in 20 seconds. Free, no signup, and we identify ourselves on every request. This is the markdown version of https://llmsrank.dev/. The site's index for agents is https://llmsrank.dev/llms.txt. ## Run a scan - In a browser: open https://llmsrank.dev/#scan, enter a domain (for example example.com) and press Scan. - Over HTTP: POST https://llmsrank.dev/api/scan with the form field `domain=example.com` (application/x-www-form-urlencoded). The answer is a 303 redirect to /scan/. - Poll GET https://llmsrank.dev/api/scan/ (JSON) until `done` is true, then read the report as markdown at https://llmsrank.dev/scan/.md. - Free scans are limited to 5 per day per client. Scans fetch as LLMSRank-User; see /bot.md. ## What is llms.txt? llms.txt is a plain markdown file that gives AI assistants and agents a short, curated guide to your most useful pages. It can sit at the root of your site (yoursite.com/llms.txt) or under a path like /docs/llms.txt; the August 2026 revision defined that a file covers the pages below its path. It was proposed at llmstxt.org by Jeremy Howard in September 2024. The format is deliberately simple: a title, a one-line summary, and lists of links, each with a sentence about what the page answers. Only the H1 title is required. Pages can also have markdown versions, at page.html.md or page.md, and can point to them and to the llms.txt that covers them with rel="alternate" and rel="describedby" links. A section headed ## Optional is still allowed, but it is now just a heading with no special meaning. A complete, valid llms.txt for a small practice: (Source: llms-txt-coverage.md) ```markdown # Example Family Dental > Family dental practice in Springfield, open six days a week. New patients welcome. ## Services - [Cleanings and check-ups](https://example.com/cleanings): What a visit includes, how long it takes, and which plans cover it. - [Emergency care](https://example.com/emergency): Same-day appointments and what to do before you arrive. ## Visit - [Book an appointment](https://example.com/book): Online booking, opening hours, and parking. ## About - [Meet the team](https://example.com/team): Our dentists and hygienists, and how long each has been with the practice. ``` ## Why publish one - **A curated map for AI.** Instead of piecing your site together from menus and scripts, an agent that reads the file gets your best pages, in the order you choose, each with a sentence on what it answers. - **Tell agents how to act.** Point agents to your API, OpenAPI spec or MCP server, and give them plain instructions for booking, quoting or reaching support. 41 of the 187 files we parsed already include instructions written for AI agents. (Source: corpus/CORRECTIONS.md) - **Describe yourself in your own words.** Your summary, your terminology, your guidance — including which details change often and where the current figures live — so assistants start from what you actually say. - **One small file.** It is one text file, and it does not need to be big: one of the top 20 files in our study is 4 KB. (Source: llmsrankings.md) - **Works alongside what you have.** robots.txt says who may crawl, sitemap.xml lists everything, and llms.txt picks out what matters most. It replaces neither, and our scan reads your robots.txt alongside it. - **Still early outside tech.** 64 of the top 100 AI companies publish one. No other group we checked gets above 15%. (Source: llmsrankings.md) ## What 723 domains taught us We checked 723 domains — the top 100 US websites, 400 more of the Fortune 500, the top 100 global brands and the top 100 AI companies — in July 2026, and re-checked every one of them in October. July found 158 files; October found 182. We archive the raw bytes of each one, so every finding here can be re-checked against the file we actually read. (Source: llmsrankings.md) ### Most AI companies already publish one Adoption is growing, and unevenly. AI companies publish llms.txt at roughly four times the rate of other leading sites. Everywhere else it sits between 15% and 18%, up from 11% to 15% in July — except among global brands, where it did not move at all. These rates count every domain we checked, including the 186 that blocked us or did not answer, so each one understates adoption; among the domains that did answer, the same cohorts run 23% to 75%. | Cohort | Checked | With llms.txt | Adoption | |---|---:|---:|---:| | Top 100 AI companies | 99 | 70 | 71% | | Top 100 US websites | 100 | 18 | 18% | | Fortune 500 (next 400) | 397 | 61 | 15% | | Top 100 global brands | 100 | 15 | 15% | Source: llmsrankings.md ### By the numbers - **+24** more sites publish one since July. Across the same 723 domains: 158 files in July 2026, 182 in October. 39 files are new; only three sites we could still read had removed theirs. Adoption rose in three of the four cohorts and held flat among global brands. (Source: llmsrankings.md) - **58** files unchanged since July. Of the 147 files present in both scans, 58 are byte-identical eleven weeks later and 89 changed. Publishing the file is not the same as maintaining it, and with only 20 of 187 files carrying a date stamp, a reader cannot tell which is which. (Source: llmsrankings.md) - **57** files mention MCP. 57 of 187 files mention the Model Context Protocol, just ahead of the 54 that use ## Optional. (Source: llmsrankings.md) - **54** files use ## Optional. At the time, the spec’s way to mark links an assistant could skip. The August 2026 revision keeps the heading but drops the special meaning. (Source: llmsrankings.md) - **41** files include agent instructions. About one in five files speaks directly to AI agents about how to use the site. (Source: corpus/CORRECTIONS.md) - **3 of 7** AI-visibility vendors publish a working llms.txt. We pointed our own link and markdown checks at seven companies selling AI-visibility tools, ourselves included (2026-10-05). Three publish a well-formed file, three publish none, and one publishes a file in which 4 of its 103 links still resolve. We do not name them: each is a half-hour fix, and some may be fixed already. A file nobody validates is a file that quietly stops working. (Source: corpus/CORRECTIONS.md) - **10 KB** median file size. Size is not quality: the highest-scoring file in the October scan is 2.8 KB, and the largest is 2.3 MB. (Source: llmsrankings.md) ## What the top files do Engine scores — rubric v2.1, files captured 2026-10-10. Every file here is scored by the same engine that grades a free scan, from the archived bytes. All twelve open with a summary; nine speak directly to agents; seven document an API and seven name an MCP server. What no longer holds is that a top file needs a long link list — two of the top three have no links at all, and earn their place on guidance instead. | Site | Score | Grade | Profile | What earns it | |---|---:|---|---|---| | [netlify.com](https://www.netlify.com) | 81.4 | B | generic | The highest engine score in the corpus, from 2.8 KB: 27 links, every one described, and plain instructions addressed to agents | | [aflac.com](https://www.aflac.com) | 79.2 | C | regulated | No link list at all — seven sections of prose written for agents, with guidance on what to offer and what never to confirm | | [taylormorrison.com](https://www.taylormorrison.com) | 78.3 | C | regulated | Thirteen sections naming which content goes stale and how fast, plus explicit agent rules | | [expedia.com](https://www.expedia.com) | 75.8 | C | commerce | States who it is and gives agents prose guidance on when to act, with all ten links described | | [tavily.com](https://tavily.com) | 75.5 | C | generic | 109 links, every one described, an API with auth and rate limits documented, and an MCP endpoint | | [www.khov.com](https://www.khov.com) | 75 | C | regulated | The most complete file in the corpus at 60.8 KB: freshness tiers, intent routing, never-confirm rules, identity, and multi-step journeys with handoffs | | [you.com](https://you.com) | 74.9 | C | generic | Instructions for agents, 98% of 100 links described, and an API documenting errors, limits, auth and writes | | [cloudflare.com](https://www.cloudflare.com) | 73.5 | C | generic | 90 links, every one described, an API and an MCP server | | [developers.deepl.com](https://developers.deepl.com) | 72.8 | C | technical | 140 links at 96% described, freshness tiers, and the only file in the corpus documenting all four API reliability parts | | [attio.com](https://attio.com) | 72.3 | C | generic | All 21 links described, an MCP endpoint, and guidance on when to offer an action | | [retool.com](https://retool.com) | 72.3 | C | generic | 43 links at 95% described, agent instructions and an MCP server | | [zapier.com](https://zapier.com) | 71.9 | C | generic | Agent instructions, all 24 links described, an API, an MCP server and WebMCP | Edition 1 listed standouts beyond tech separately, because none could place on merit. That category is now redundant: an insurer, two homebuilders and a travel site hold four of the top six places, and the four non-tech names Edition 1 singled out would all grade D or F today. Scores are the engine's, reproducible from the archived files: 187 files captured 2026-10-10, rubric v2.1 against llms.txt v2 (2026-08-10). The corpus holds one B and sixteen Cs, so an honest list is topped by a C. Scores are computed within a profile — generic, commerce, regulated or technical weight the categories differently — so each row names its own, and a row is not strictly comparable with one from another profile. Sites are named only as good examples, never as failing. Edition 1 published hand-assigned scores for these slots; those are retired rather than shown alongside, and the corrections log explains what the gap turned out to be. (Source: llmsrankings.md) ## Easy wins we see most The most common problems in the October 2026 scan, counted, each with a short fix. ### Let AI agents fetch it - Finding: 10 of the 182 sites publishing an llms.txt name an AI crawler in robots.txt and disallow it from fetching the file. (Source: llmsrankings.md) - Fix: Allow the AI agents and tools you want (GPTBot, ClaudeBot, PerplexityBot and others; see /docs/audits/named-ai-crawlers) to fetch /llms.txt. No AI vendor documents reading llms.txt, but a blocked file can’t be read at all. ### Start with a title - Finding: 5 of 187 files have no H1 — the one element the spec requires. (Source: llmsrankings.md) - Fix: Make the first heading # Your Site Name. ### Serve the file, not a web page - Finding: 53 domains return an HTML page at /llms.txt instead of a markdown file. (Source: llmsrankings.md) - Fix: Serve plain markdown, not a web page, and a real 404 until your file is ready. The proposal doesn’t name a content type; our rubric (R3) looks for text/plain or text/markdown. ### Put a date on it - Finding: Only 20 of 187 files carry a version or last-updated stamp. (Source: llmsrankings.md) - Fix: Add one line, such as Last updated: 2026-09-01, and change it when the file changes. ### Name the AI crawlers you welcome - Finding: Only 111 of the 556 robots.txt files we could read name any AI crawler at all. (Source: llmsrankings.md) - Fix: Add explicit User-agent rules for the AI crawlers you want, so your intent is unambiguous. ### Publish on your main domain - Finding: 19 of the 70 AI-company files sit on a docs or developer subdomain rather than the main domain. (Source: llmsrankings.md) - Fix: A subdomain is a separate site, not a path, so its file does not cover your main domain. Publish one at yoursite.com/llms.txt too; it can link to the docs file. ## How LLMSRank helps Five stages, from first scan to measured results. Stage one, the free scan, is live now. Stages two to five are being built; this is what each will do. 1. **Diagnose.** The free scan fetches your robots.txt, llms.txt, llms-full.txt and AGENTS.md, scores the file against the audits that apply to your site, and leads with anything broken before the number. 2. **Author.** Will turn findings into a draft file: a diff against your current one, built from your own pages, asking only about the elements (of ~80) that fit your site. 3. **Deploy.** Will help get the file live at the right path, with robots.txt and companion files around it, then check that it is live and readable. 4. **Maintain.** Sites change. Will check daily that your file is still there and unchanged, and propose updates when pages move, links break or details go stale. 5. **Prove.** Will measure how often AI engines mention and cite you, over time. So far we have run it by hand: our first baseline ran 78 prompts across four engines. ### How we score - **Six categories.** Retrievability & Access, Spec Conformance, Curation Quality, Agent Readiness, Compliance & Trust, and Ecosystem Integration — graded A (90+) to F. - **Scored only on what applies to you.** A 12-page site with no API is never marked down for lacking an API section. A simple site can score 100. - **Unverified is never counted against you.** If our scan can’t confirm something, it is flagged for you to check and left out of the score. - **Corrections are published.** When we get something wrong, the fix and its cause go in a public corrections log — never a silent patch. [What we got wrong](/learn/what-we-got-wrong) tells three of those stories: a null result we published instead of the headline, a metric that turned out to be measuring the wrong thing, and claims in this field that aged badly within months. Sources: plan.md §4, scoring.md, llms-txt-coverage.md, examples/khov-visibility-baseline.md, corpus/CORRECTIONS.md ## See where your site stands Give the scan a domain and it returns a graded report: what works, what stops AI crawlers fetching the file, and the first fix to make. No file yet? It says so, and does not grade you for it. Start a scan: https://llmsrank.dev/#scan ## Frequently asked questions ### What does LLMSRank check? A scan fetches your robots.txt, llms.txt, llms-full.txt, AGENTS.md and sitemap.xml, then scores the llms.txt against the audits that apply to your site. The report leads with anything that stops agents reading the file, then the grade and the fixes worth the most points. ### Is it free? Yes. The scan is free, up to five scans a day from each network connection (IP address). No account is needed. ### How does the score work? Audits are grouped into six categories: Retrievability & Access, Spec Conformance, Curation Quality, Agent Readiness, Compliance & Trust, and Ecosystem Integration. Only audits that apply to your site count, anything the scan couldn’t verify is left out rather than scored as zero, and the total maps to a grade: A (90+), B (80–89), C (70–79), D (60–69) or F. The full method is at [/docs/method](https://llmsrank.dev/docs/method). ### What is LLMSRank-User, and does it obey robots.txt? LLMSRank-User is the user agent for scans a person starts on llmsrank.dev. It acts on a person’s request rather than crawling, so a scan doesn’t consult robots.txt; it fetches the same handful of files once per scan. The llms.txt draft and markdown versions tools do honour robots.txt for LLMSRank-User: pages it disallows are skipped and never fetched. Vendors differ here: OpenAI says robots.txt rules may not apply to ChatGPT-User and Perplexity says Perplexity-User generally ignores them, while Anthropic says its bots, Claude-User included, honour robots.txt. Our research crawler, LLMSRankBot, does obey robots.txt; both are described at [/bot](https://llmsrank.dev/bot). ### What happens if my site has no llms.txt? You get no grade, because there is nothing to grade: the report says no file was found, gives you a three-line starter file with your domain filled in, and lists the most common easy fixes. Publish the file and scan again. ### How is this different from Lighthouse’s llms.txt check? Chrome’s Lighthouse checks that the file can be fetched: it fails on a server error and marks a missing file as not applicable. LLMSRank reads the file and grades what is in it, along with the robots.txt, sitemap and agent files around it. ### Will a high score get my site cited by AI assistants? No score can guarantee that. A good llms.txt makes your site easier for agents to read and use correctly, but whether an assistant mentions or cites you depends on the assistant, and we don’t claim otherwise. ### What is llms.txt? A proposed markdown file, from llmstxt.org, that gives AI agents a short, curated guide to your site: a title, a summary and lists of links with a line on what each page answers. [About llms.txt](https://llmsrank.dev/learn/llms-txt) explains it in full. ## Sources - llmsrankings.md: Corpus sweeps of the same 723 domains, July and October 2026: cohort adoption, file-size and corpus findings. Leaderboard scores are engine scores (rubric v2.1) from the October capture, not the retired Edition 1 judgments. - corpus/CORRECTIONS.md: Every post-publication correction, the parser-verified counts, and the comparison between expert-judgement and engine scores. - scoring.md: The LLMSRank score: six categories, grade bands, applicability and unverified handling. - llms-txt-coverage.md: The ~80-element llms.txt capability matrix and the llmstxt.org spec skeleton. - examples/khov-visibility-baseline.md: Our first machine-measured AI visibility baseline (78 prompts across four engines). - plan.md §4: The five-stage journey: Diagnose, Author, Deploy, Maintain, Prove. - https://llmstxt.org: The llms.txt proposal (Jeremy Howard, Answer.AI, September 2024; v2 revised August 2026).