New The Modern Operations Path™: five steps, enter where you are
All guides Guide · feeds The Scan

Declare your AI crawl policy. Here are the exact lines.

Right now your site has an AI crawl policy whether you wrote one or not — silence just means "whatever each bot decides." This guide is the ten-minute fix: the exact robots.txt lines for the AI crawlers that matter, the allow and the block variants, how to check they took, and the minimal llms.txt that goes with them. If you're still deciding whether to open or close the door, start with the trade behind blocking AI crawlers — this page is for when you've decided and want it written down.

A harbor master at dusk painting clear pictogram rules onto a signboard at the dock gate while a queue of small boats waits on the water

Where the file lives, and how the rules work

robots.txt is one plain-text file at the root of your host: yoursite.com/robots.txt. Not in a folder, not one per page — one file, at the root, per host (a www and non-www version each read their own).

The format is blocks. Each block names an agent, then lists what it may or may not read. A crawler looks for the block that names it; if none does, it falls back to the User-agent: * block; if there's no file at all, everything is allowed by default. That default is why most sites' policy is implied — the bots are reading you because you never said anything.

Implied and stated are different things. A stated policy survives the next developer, the next replatform, the next security plugin. An implied one lasts until someone pastes a template over it.

The five agents worth naming

  • GPTBot — OpenAI. Feeds model training and the reading ChatGPT does.
  • ClaudeBot — Anthropic's crawler, the one that fetches pages.
  • anthropic-ai — Anthropic's training token. Name both to cover Anthropic fully.
  • PerplexityBot — Perplexity, the answer engine that cites sources.
  • Google-Extended — not a crawler but a switch: it controls whether Google may use your content for AI training and AI features. Blocking it does not touch regular Google search — that's Googlebot, a different name, and you almost certainly want it left alone.

The open door, written down

For most expert-led services businesses — where the site is marketing for judgment you sell in engagements — the right policy is open, on purpose. Yes, no rule already means allowed. Writing the allows anyway is the point: the file now says what you decided, and a future template can't silently reverse it without someone noticing the difference.

# AI crawlers: allowed, on purpose.
User-agent: GPTBot
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: anthropic-ai
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Google-Extended
Allow: /

User-agent: *
Allow: /

Sitemap: https://yoursite.com/sitemap.xml

Want the door open but one room private? Rules within a block stack — disallow the directory, allow the rest:

User-agent: GPTBot
Disallow: /members/
Allow: /

The closed door, written down

If your content is the product — you sell the words, the data, the course material — the block looks like this. Note what's missing: no User-agent: * disallow, and no Googlebot block. You're turning away AI readers, not leaving search.

# AI crawlers: blocked, on purpose.
User-agent: GPTBot
Disallow: /

User-agent: ClaudeBot
Disallow: /

User-agent: anthropic-ai
Disallow: /

User-agent: PerplexityBot
Disallow: /

User-agent: Google-Extended
Disallow: /

One honest caveat, stated once: robots.txt is a convention. Reputable crawlers honor it; hostile scrapers ignore it. That doesn't weaken the file — the crawlers that decide your visibility are the reputable ones — it just means robots.txt is policy, not security.

Check it took

Open yoursite.com/robots.txt in a browser and read what's actually there. Not what you uploaded — what the server serves. On site builders (Wix, Squarespace, Webflow, Shopify) the file is generated from a settings screen, so edit it there, then check the URL again.

Three things that quietly break the file: it isn't at the root; the agent name is misspelled (they're case-insensitive, but GPT-Bot matches nothing); or a stray Disallow: / sits under User-agent: * from staging and outranks everything you just wrote. Crawlers refetch the file on their own schedule — figure a day before behavior changes.

The llms.txt that goes with it

robots.txt says who may read. llms.txt says what to read first — a briefing you write so machines stop assembling one from scraps. The minimal version is four parts, plain text, at yoursite.com/llms.txt:

# Your Business Name

> One sentence: who you help, and with what.

## Offers
- [The Assessment](https://yoursite.com/assessment) — what it is, $X.
- [The Program](https://yoursite.com/program) — what it is, $Y/month.

## Contact
- Email: [email protected]
- Site: https://yoursite.com

Facts, prices, links. No marketing language — machines quote a briefing better than a brochure. An hour of writing, no developer, and it compounds with everything else that makes your site readable to AI: the same telling-instead-of-guessing that the free search signals do page by page.

Ten minutes, then verify it

The free Site Readiness Scan checks this exact thing — whether your robots.txt names the AI bots, whether llms.txt exists, whether a leftover disallow is overruling the policy you just wrote — along with the structure and content signals machines read.

Run the free Scan

If the doors are open and stated but the answers about your business still come back wrong, the problem is deeper than the front door — the Read ($450) finds where it lives and ranks the fixes.