# AGENTS.md and agent-instructions builder > Fill in what your site is for, how fresh its data is, where to route questions and who you are, and get an llms.txt block and an AGENTS.md, with the audits that check them scored as you type. This is the markdown version of https://llmsrank.dev/tools/agent-instructions. The tool is beta. It runs in the browser: nothing you type is sent anywhere, unless you choose to score a file against your site’s latest scan. Claims checked 2026-10-02. These files are instructions, not controls. Our audits check that the sections are there. They can’t tell you what any agent does with them. ## What it does You answer eight sections: 1 role and scope, 2 citation, 3 never-confirm rules, 4 data-freshness tiers, 5 intent routing, 6 handoff to a person, 7 identity, 8 API. The builder writes an llms.txt block (“For AI agents”, “Intent routing”, “Data freshness”, “Identity”) and an AGENTS.md, both in editable text boxes. Every line holds a value you entered; text you type cannot create links, headings or comments in the output. The audit panel scores your llms.txt text for [Role and scope](https://llmsrank.dev/docs/audits/role-and-scope.md), [Data-freshness tiers](https://llmsrank.dev/docs/audits/data-freshness-tiers.md), [Intent routing](https://llmsrank.dev/docs/audits/intent-routing.md), [FAQ blocks](https://llmsrank.dev/docs/audits/faq-blocks.md), [Entity identity](https://llmsrank.dev/docs/audits/entity-identity.md), and shows whether the AGENTS.md is at least 200 bytes (the size part of [Agent files](https://llmsrank.dev/docs/audits/agent-files.md); a scan checks that the file is served). The form has no industry field, so the industry compliance audit (P2) is not shown. The panel does not add the audits into one number. Optionally load your current llms.txt (one of your free scans) to append the block to it and score the result. Export copies or downloads each file as UTF-8 text with LF line endings, one final newline and no byte order mark. ## Presets Each preset fills the form with made-up example values (the domains end in .example). Replace every one before you publish. - Local business - SaaS docs - Online store ## Notes beside each section - **The llms.txt block helps only agents that fetch llms.txt.** The block does nothing for an agent that never fetches your llms.txt. We checked vendor documentation on 2026-09-29. No AI vendor has documented reading other sites’ llms.txt files, in an assistant, a search product or a crawler. Google says Google Search ignores them. The tools documented to read llms.txt read a file that a user points them at. Sources: [Our research: Does ChatGPT, Claude, Perplexity or Google read llms.txt? (vendor statements, checked 2026-09-29)](https://llmsrank.dev/learn/does-ai-read-llms-txt); [Google Search Central: Optimizing your website for generative AI features on Google Search (“Google Search ignores them”)](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide). - **“The engine credits this” is not “agents obey this”.** Our audits check that sections such as “For AI agents”, “Data freshness” and “Intent routing” are present and worded as the audit expects. A full score on A1, A2 or A3 means the text is there. It says nothing about whether an agent read it or acted on it. No agent is bound by your llms.txt or your AGENTS.md. Sources: [Our audit doc: Role and scope (A1)](https://llmsrank.dev/docs/audits/role-and-scope); [Our audit doc: Data-freshness tiers (A2)](https://llmsrank.dev/docs/audits/data-freshness-tiers); [Our audit doc: Intent routing (A3)](https://llmsrank.dev/docs/audits/intent-routing); [Our research: Does ChatGPT, Claude, Perplexity or Google read llms.txt? (vendor statements, checked 2026-09-29)](https://llmsrank.dev/learn/does-ai-read-llms-txt). - **Freshness and never-confirm lines are requests.** A data-freshness tier or a never-confirm rule asks an agent to be careful. It can’t stop an agent from quoting an old price or a sold-out item as current. Keep prices, stock and availability on your own systems. Check them there before you accept any order, booking or quote. Sources: [Our audit doc: Data-freshness tiers (A2)](https://llmsrank.dev/docs/audits/data-freshness-tiers). - **Identifiers help only when they are right.** An LEI, a Wikidata ID, a stock ticker or a registration number can help an agent tell you apart from a similar name. That holds only if the value is correct and matches the rest of your site. Use the same legal name and identifiers in your Organization JSON-LD (our JSON-LD agreement audit, E4) and in the Identity section (our Entity identity audit, A8). A wrong identifier points to another organization. Sources: [Our audit doc: Entity identity (A8)](https://llmsrank.dev/docs/audits/entity-identity); [Our audit doc: JSON-LD agreement (E4)](https://llmsrank.dev/docs/audits/json-ld-agreement). - **AGENTS.md is a code-repository convention.** agents.md calls AGENTS.md “a README for agents.” It says to put the file at the root of a code repository. Coding agents read it while they work in that repository, and the closest AGENTS.md to the edited file wins. It is plain Markdown with no required fields. OpenAI contributed it to the Agentic AI Foundation (AAIF) when the Linux Foundation formed that foundation on 2025-12-09. agents.md now says the AAIF stewards it. Sources: [agents.md: “a README for agents”; “Create an AGENTS.md file at the root of the repository”; “just standard Markdown”; “The closest AGENTS.md to the edited file wins”](https://agents.md/); [Linux Foundation press release: formation of the Agentic AI Foundation (AAIF), with MCP, goose and AGENTS.md (contributed by OpenAI)](https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation). - **A web-root /AGENTS.md is our rubric’s recommendation.** agents.md says nothing about websites serving `/AGENTS.md`. We found no agent documented as fetching one from a website. Our rubric’s Agent files audit (E7) checks for one at your site root, and counts it from 200 bytes. That is our rubric’s recommendation, not a published standard for websites. Sources: [agents.md: “a README for agents”; “Create an AGENTS.md file at the root of the repository”; “just standard Markdown”; “The closest AGENTS.md to the edited file wins”](https://agents.md/); [Our audit doc: Agent files (E7)](https://llmsrank.dev/docs/audits/agent-files); [Our guide: llms.txt vs AGENTS.md](https://llmsrank.dev/learn/llms-txt-vs-agents-md). ## Before you publish Before you publish: these are instructions for agents, not controls. A full audit score means the sections are present. Check prices and availability on your own systems. Make sure every identifier is correct. ## Diff, check again and score a draft - **llms.txt.** The proposed file is shown against your live one as a line diff, with Copy and Download for the proposed file. After you publish, “Check again” fetches only your live llms.txt again (through our server, as LLMSRank-User; it counts as one of your free scans) and says what changed since the last fetch and whether it now matches. - **AGENTS.md.** The proposed file is shown against your live one as a line diff, with Copy and Download for the proposed file. After you publish, “Check again” fetches only your live AGENTS.md again (through our server, as LLMSRank-User; it counts as one of your free scans) and says what changed since the last fetch and whether it now matches. “Score this draft” re-scores your site’s latest scan with the draft in place of the live file and shows the scan’s score with and without it, any cap, and every audit that would change. Nothing is published and the draft is not fetched; it is sent to our server only to be scored and is not kept. Audits that depend on pages the scan fetched for your live file’s links are kept from the scan and named. Limits: llms.txt 1 MB, AGENTS.md 256 KiB; 100 drafts a day, shared with the robots.txt fetch.