Web Development • HIPAA-Aware Healthcare Web • SEO • AI Search Optimization (407) 409-8383   |   [email protected]
llms.txt & AI Discovery

llms.txt, shipped honestly: cheap hygiene, not a promise.

llms.txt is an emerging proposal for a curated Markdown map of your key pages, plus a deliberate AI-crawler policy in your robots.txt. Adoption is thin and unconfirmed, and Google has said plainly it does not use the file. We still ship one, because it costs almost nothing, tidies your site, and gets read by the agents people point at you. We will not tell you it earns citations. It does not, yet.

EMERGING PROPOSAL ROBOTS.TXT POLICY NO GUARANTEES
The overview

A discovery file, told straight instead of oversold.

llms.txt is a single Markdown file you place at the root of your domain. Its job is to hand an AI tool a short, curated map of your most important pages, so a model working inside a limited context window does not have to reconstruct your whole site from raw HTML full of navigation, ads, and scripts. The reasoning behind the proposal is sound, which is part of why it spread: the site author knows best which pages matter, so the author curates the short list. That much holds up well.

What does not hold up is the way the file usually gets sold. You will read that llms.txt is the AI-era robots.txt, a settled requirement you cannot afford to skip. It is not. It is a proposal, made in September 2024, that the people who run the major crawlers have so far declined to adopt as a production signal. Google has been the most direct about it, and the measured evidence for citation gains is thin to absent. We would rather lead with that than sell you a comparison that does not survive contact with the data.

So this is the small, honest page of the hub. We write you a good llms.txt by hand from your service taxonomy, and we set an AI-crawler policy in your robots.txt that reflects a decision you actually made. Both are cheap, both are tidy, and neither is a promise. If you want the work that genuinely changes how AI tools read and cite you, that is the structural and schema work on the pages themselves, and it lives on the neighboring pages here.

What this covers

Two small, honest pieces of discovery hygiene, done by hand and explained plainly.

  • A hand-authored llms.txt from your real taxonomy
  • An optional llms-full.txt for content-heavy sites
  • A deliberate AI-crawler policy in robots.txt
  • Per-bot allow and disallow, decided with you
  • Curation that surfaces stale and duplicate pages
  • A straight read on what it does and does not do
The honest picture

What the crawlers do, and what the data shows.

We think you deserve the unflattering version before the pitch, so here it is: the major AI providers have not committed to the file, and the studies do not show a citation lift. Read this first, then decide.

// a proposal the crawler operators have, so far, declined to adopt

THE CRAWLERS

Not committed

No major model provider honors llms.txt as a production signal. OpenAI's GPTBot respects robots.txt but does not officially use llms.txt. Google manages AI crawling through robots.txt and the Google-Extended token, with no llms.txt support, and said in June 2026 the file has no effect on Search or AI Overviews. Anthropic publishes its own llms.txt for developer docs but points to robots.txt in its crawler guidance.

Google's John Mueller compared the file to the discredited keywords meta tag and pointed out you can check your own server logs and watch the crawlers never request it. That is about as plain a signal as an emerging convention gets.

THE DATA

Thin and unproven

Ahrefs looked at roughly 137,000 domains, found about 28 percent published a valid llms.txt, and then found around 97 percent of those files got zero requests in the month they measured. AI systems did not go looking when the file was absent, so it is fetched only when something explicitly points at it.

SE Ranking studied roughly 300,000 domains, put adoption near ten percent, and found no measurable link between having the file and citation frequency. Their model performed slightly better with the signal removed. Ahrefs' own answer to whether you should implement it, as of 2026, was no, not yet.

Capabilities

What is included when we ship one.

A good llms.txt follows the real spec, not the loose descriptions that circulate. The only required element is a single H1 with your site name. Everything else is optional structure: a one-line summary, some detail, and H2 sections of curated Markdown links with short factual notes. A section literally named Optional marks links a model can skip for a shorter context.

We write it by hand from your service taxonomy so the pages that matter are named and ordered, and we avoid the two common failure modes: dumping your whole sitemap into it, and writing the notes in marketing voice. For content-heavy sites we can add a longer llms-full.txt carrying the actual content of priority pages in Markdown.

What's included

  • A hand-authored llms.txt at /llms.txt
  • Curation from your real service taxonomy
  • Spec-correct H1, summary, and H2 link sections
  • An Optional section for skippable links
  • Short, factual notes, not marketing voice
  • Optional llms-full.txt for content-heavy sites
  • An AI-aware robots.txt with explicit crawler rules
  • Per-bot allow and disallow decisions, made with you
  • Linked-page reachability and Markdown validation
  • Documentation your team can keep current
Crawler policy

The AI-crawler policy is the real decision here.

It is worth separating the two files, because they do different things and carry different weight. llms.txt is an unproven proposal about content discovery. Your robots.txt is a widely honored file about access, and the AI-crawler tokens in it are a genuine, deliberate policy choice, not a ranking trick.

The tokens split into three kinds. Training crawlers, like GPTBot, ClaudeBot, and CCBot, fetch content to train models, so disallowing them is how you opt out of future training. Retrieval bots, like OAI-SearchBot, Claude-SearchBot, and PerplexityBot, power search and answers and are the ones that can actually cite you. Training-control tokens, like Google-Extended and Applebot-Extended, make no requests of their own and only govern existing crawlers.

A common stance in 2026 is to allow the retrieval bots that can cite you while opting out of training. That is a reasonable default, not a rule. We will walk you through the tradeoff for your business and set robots.txt to match the position you actually hold, rather than paste in someone else's template.

Two files, two jobs

llms.txtrobots.txt (AI tokens)
StatusEmerging proposalWidely honored standard
JobCurated content mapAccess and crawl policy
Honored by crawlersNot yet, in practiceYes, broadly
The real valueHygiene and a hedgeA deliberate policy choice

The approach. We treat the crawler policy as the decision that matters and the llms.txt as low-cost hygiene alongside it. You get a robots.txt that reflects a position you chose, an llms.txt written by hand, and a plain explanation of why each bot is allowed or blocked, so your team can adjust it as your stance changes.

How we work

Curate honestly, serve it, keep it cheap.

This is a small engagement on purpose, usually a single pass rather than an ongoing retainer. We curate the file by hand, set a crawler policy you actually decided, validate that the links resolve, and hand it over documented so your team can keep it current without us.

// small, honest, and cheap to maintain, exactly as it should be

  1. Curate the pages that matterWe work from your real service taxonomy, not your XML sitemap, and pick the thirty or so pages a knowledgeable person would point an AI tool at. The act of curating usually surfaces stale or duplicate pages worth cleaning up regardless of who reads the file.
  2. Write it to the real specA single H1 with your site name, an optional one-line summary, and H2 sections of Markdown links with short, factual notes. We add an Optional section for skippable links, and for content-heavy sites we can produce a longer llms-full.txt carrying the actual page content in Markdown.
  3. Decide the crawler policy with youWe lay out the training, retrieval, and training-control tokens and what allowing or blocking each one means for your business, then set robots.txt to match the position you hold. This is the part that is a real decision, so we treat it as one.
  4. Validate and serveWe check the Markdown is well-formed and every linked page actually resolves, then serve the file at /llms.txt. No broken links, no pages that redirect or 404, nothing that would make the file useless to the agents that do fetch it.
  5. Document and hand overYou get the files, a plain-language note on why each crawler is allowed or blocked, and instructions for keeping the curation current as your site grows. Because it is cheap to maintain, it stays yours to run rather than a standing invoice.
Read the long version

We wrote the honest, sourced case in full.

If you want the whole argument, with the studies cited and the failure modes named, we published it. It covers what llms.txt actually is, why the idea exists, the adoption picture, what a good one contains, and the clear-eyed case for shipping one anyway. It is the reading we would hand a skeptical client, because we are skeptical too.

Questions, answered plainly

Frequently asked questions

It is a single Markdown file placed at the root of your domain that hands an AI tool a short, curated map of your most important pages, so a model does not have to work your whole site out from the raw HTML. It was proposed in September 2024 by Jeremy Howard at Answer.AI, and the spec lives at llmstxt.org, where it is described plainly as a proposal to standardise on. The word proposal is the important part. It is a community convention site owners can choose to follow, not an adopted standard the way robots.txt and sitemap.xml are.

Mostly no, and we would rather say that plainly than sell you a comparison to robots.txt that does not hold. No major model provider has committed to honoring llms.txt as a production signal. Google has been the most direct: Gary Illyes said Google does not support it and is not planning to, and John Mueller compared it to the discredited old keywords meta tag and noted you can look at your own server logs and see the crawlers do not even check for it. In June 2026 Google updated its guidance to state plainly that llms.txt has no effect, positive or negative, on Search or AI Overviews. Where it genuinely does get read today is developer and agent tooling: IDE assistants and coding agents fetch the file when someone points them at your site.

It does not support the popular claim that shipping the file lifts your AI citations. Ahrefs looked at roughly 137,000 domains, found about 28 percent published a valid llms.txt, and then found around 97 percent of those files received zero requests in the month they measured. AI systems did not go looking for the file when it was absent, so it gets fetched only when something explicitly points at it. Separately, SE Ranking studied roughly 300,000 domains, put adoption near ten percent, and found no measurable link between having the file and how often a domain gets cited. Their model actually performed slightly better with the llms.txt signal removed. Ahrefs' own recommendation, as of 2026, was no, not yet.

For the right reasons, with clear eyes. The file is cheap to produce and harmless to serve. The act of curating your key pages and describing each in one plain sentence is good hygiene no matter who reads it, and it tends to surface stale or duplicate pages you had forgotten about. There is a concrete use today that has nothing to do with search: coding agents and in-product AI assistants will fetch it when pointed at your site, which already earns its keep for documentation-heavy and product sites. And if adoption does grow, you have a low-cost hedge already in place. That is the honest size of it, and it is still worth doing on those terms.

Follow the real spec, not the loose descriptions floating around. The only required element is a single H1 with the site or project name. From there you can add a one-line summary as a blockquote, some free-form detail, and H2 sections holding Markdown links with an optional short note after each. One H2 has a defined meaning: a section literally named Optional marks links a model can skip when it needs a shorter context. Two failure modes to avoid: dumping every URL from your XML sitemap into it, which defeats the whole point of curation, and writing the notes in marketing voice. Point at the thirty or so pages a knowledgeable person would, and describe them factually.

Different file, different job. llms.txt is an unproven proposal about content discovery. Your robots.txt is a widely honored file about access, and the AI-crawler tokens that live in it are a real, deliberate policy choice. You can decide, per bot, whether to allow training crawlers like GPTBot, ClaudeBot, and CCBot, retrieval bots like OAI-SearchBot, Claude-SearchBot, and PerplexityBot, and training-control tokens like Google-Extended and Applebot-Extended, which make no requests of their own and only govern existing crawlers. A common stance in 2026 is to allow the retrieval bots that can cite you while opting out of training. We will lay out the tradeoff and set it up to match your position, rather than paste in a template.

No, and anyone who tells you an llms.txt file will lift your citations in ChatGPT, Perplexity, or Google's AI answers is ahead of the evidence. A before-and-after citation test today is unlikely to show any llms.txt-driven effect you could trust. We treat the file as cheap, sensible hygiene and a small bet on an emerging convention, not a ranking lever. If you want the work that actually moves how AI tools read and cite you, that is the structural and schema work on the pages themselves, and it lives on the neighboring pages of this hub.

Want an llms.txt and a crawler policy you can trust?

Send us your URL. We will curate a spec-correct llms.txt by hand, lay out the AI-crawler tradeoff so you can set a policy you actually chose, and give you a straight read on what it will and will not do. No promises about citations, because there are none to make.