Free tool · Robots.txt for AI search

Robots.txt AI Visibility Audit

Reads a robots.txt file and shows, for every LLM, whether its training, search and user crawlers are allowed or blocked, what that does to your AI visibility and where the file is costing you. Built from Should Publishers Block AI Training? Why Block It, When to Allow It and What the Data Says · How it works

Instructions: Copy-paste the contents of your robots.txt below and hit Check, or click any of the example buttons for anonymised real robots.txt examples and what they mean. Hit Clear to paste a new file's content, or simply paste over it and hit Check again.

Nothing you paste is retained or shared, not even for learning purposes. The file is read in your browser and goes nowhere else.

Scope: This audits the AI visibility side of robots.txt only. For syntax errors, Googlebot and Bingbot access, sitemap lines or anything else about search engine crawling, use Google Search Console's robots.txt report, Bing Webmaster Tools, TechnicalSEO.com's robots.txt tester or Screaming Frog.

How it works and what it cannot tell you

How it reads your file:

  • The standard: It follows RFC 9309, the robots.txt standard, the same way OpenAI and Google document their crawlers.
  • Which group applies: A crawler obeys the group that names it and only falls back to the * group when nothing names it. If the same crawler is named in several groups, the rules are merged.
  • Which rule wins: Where more than one rule matches a page, the longest one wins. If an Allow and a Disallow are the same length, Allow wins.
  • Folders: Every folder the file names is checked one by one and grouped by how the 5 LLMs see it. The largest 3 groups appear in the verdict and the cards, the rest in the folder table.
  • Everything else: Every page the file does not name.
  • Housekeeping folders: Folders that also shut Googlebot out (logins, carts, admin areas) are left out, because search engines cannot see them either.

What the pills mean:

  • Full: The LLM can learn from these pages and link to them.
  • Cite only: The training crawler is blocked and the search crawler is open.
    • The LLM can link to these pages only after a live web search.
    • Nectiv found that happens on about 31% of ChatGPT prompts.
    • The other 69% of the time the answer comes from memory, where a page the model never trained on plays no part.
  • Blocked: The search crawler is blocked, so the LLM cannot cite these pages.
    • Search is what earns a citation, so once it is shut the pill is Blocked whatever the other letters say.
    • A green U on a Blocked card means the user fetcher is still open: A page will load when a person pastes its URL, but the LLM will never find it on its own.
    • In most files training is shut too, so that LLM can neither cite the pages nor draw on them from memory.
  • Train only: The LLM learns from these pages but can never link to them.
    • The training crawler is open and the search and user crawlers are blocked.
    • The model gets your content for free and you get nothing back. Nobody chooses this on purpose.
  • The percentage next to each LLM: Its share of visits to AI products (Similarweb, August 2026, website visits only). It is there for scale and nothing is scored on it.

Languages and regions:

  • When it appears: Only when the file names at least two language or region folders, such as /de/ or /pt-br/, or one with a region code.
  • What it cannot see: Language sites on their own domain or subdomain (de.example.com, example.de) have their own robots.txt and do not show up here.

What this tool cannot tell you:

  • Who actually turns up: Robots.txt is a set of directives and nothing enforces them. When researchers at UC San Diego and the University of Chicago tested the crawlers in 2025, the OpenAI, Anthropic, Common Crawl, Apple, Amazon and Meta crawlers respected it, while Bytespider ignored it and 20 of 23 smaller AI assistant crawlers never fetched it. TollBit measured 30% of AI scrapes ignoring it in late 2025. The tool reads what your file asks for, not who turns up.
  • Whether to block at all: That is a legal, IP and business decision, which the article's decision tree walks through by concern. The tool shows what the file does, so you can check it matches the decision.
  • What is behind an allowed folder: It reads only what you paste and does not fetch your pages, so it cannot tell whether a folder is behind a paywall, redirects somewhere else or has been taken down.
  • Other blocks: A firewall, a CDN rule or a meta tag can stop a crawler your file allows. On Cloudflare, since 15 September 2026, Block and Block on pages with ads apply to Googlebot, Bingbot and Applebot too, and Cloudflare's post says either setting affects search as well as training. Disallow AI Training is the setting that stops training and keeps search, and it writes the no-training lines into your robots.txt for you.
  • New crawlers: The 5 LLMs and their crawler names come from each company's own documentation as of September 2026. Companies add and rename crawlers, so re-check your file every quarter against knownagents.com/agents, which lists every known crawler in one place and is updated daily.
  • Traffic shares: Similarweb's August 2026 figures for website visits only. They understate products used mostly inside an app, Gemini above all.
  • Advice: It is a decision aid, not legal advice and not a substitute for an SEO, GEO or AEO consultation.
  • Latest version and feedback: The latest version lives on GitHub. For feedback or a bug, message Orit on LinkedIn or X.