This pack turns llms.txt validator, llms validator, llms.txt checker, llms.txt not found, invalid llms.txt, and fix llms.txt errors into one proof-linked path for humans and AI agents.
The current Search Console query map shows 85 validator/checker impressions, 1 click, and 46 zero-click impressions across the validator cluster.
The llms_validator_proof_classifier separates paste-draft validation, live root-file fetch, Markdown structure, public URL quality, private-path risk, publish readiness, AI retrieval discovery, and real traffic proof before a human or AI assistant cites the result.
Proves: Google is testing the llms.txt validator/checker cluster and the validator page has one observed click.
Does not prove: It does not prove stable CTR, ranking gains, AI citations, or conversion quality.
Current observation: 85 cluster impressions, 1 click, and 46 zero-click impressions.
Next action: Protect the winning validator path, keep checker variants bridged, and wait for Search Console refresh before judging CTR.
Proves: A pasted draft can be checked for H1, summary, Markdown links, public URLs, and private-path warnings.
Does not prove: It does not prove the file is deployed at the public root path or fetchable by agents.
Next action: After fixing draft warnings, publish the file and run a live root-file check.
Proves: The public /llms.txt endpoint can be fetched from the canonical domain.
Does not prove: It does not prove the file is curated, privacy-safe, useful, cited by AI systems, or ranking anything.
Next action: Pair live fetch proof with structure, private-path, and publish-readiness checks.
Proves: The file has readable Markdown basics such as a single H1, summary, sections, and valid links.
Does not prove: Markdown validity alone does not prove link quality, canonical correctness, privacy safety, or usefulness.
Next action: Use the publish evidence matrix to combine Markdown checks with public URL and usefulness checks.
Proves: Listed URLs are intended to be public and aligned with canonical pages.
Does not prove: It does not prove every URL is the best AI context target or that external agents will follow it.
Next action: Remove localhost, staging, token, checkout, account, redirect-heavy, and duplicate URLs.
Proves: The validator found or cleared obvious public-private boundary risks.
Does not prove: It does not prove the site has no private data exposure outside the checked file.
Next action: Use the private path risk pack before publishing client, Shopify, or account-heavy sites.
Proves: The file passed the selected draft/live/public URL checks needed for a safer launch handoff.
Does not prove: It does not guarantee rankings, clicks, AI citations, or business value.
Next action: Keep the launch note and revalidate after deployment or CMS/theme changes.
Proves: AI agents can find answer packs, proof lookup, citation bundles, and machine-readable evidence faster.
Does not prove: It does not force AI products to cite the site or guarantee inclusion in answer engines.
Next action: Cite answer packs and proof links, then measure referrals, clicks, and tool activations separately.
Proves: Real traffic requires Search Console clicks, qualified referrals, sessions, tool starts, copied reports, or conversions.
Does not prove: Validator runs, crawler hits, generated reports, and impressions alone are not traffic proof.
Next action: Measure validator starts, report copies, live checks, Search Console clicks, and qualified referrals without self-clicking.
A draft can look valid locally while the live root file still returns 404, redirects to the wrong host, blocks fetches, or is not reachable at /llms.txt.
Action: Run the live validator against the canonical domain and confirm the response is the public root /llms.txt file.
The most common useful fixes are one clear H1, a short summary, readable sections, and valid Markdown links rather than a noisy sitemap dump.
Action: Paste the draft, fix missing H1 or summary, check links, then rerun the validator before publishing.
A file can be syntactically readable but still weak if it lacks a concise site summary or points assistants to low-value pages.
Action: Use one site-level H1, one short summary, and curated public Core links that explain the site clearly.
Localhost, preview, duplicate, redirect-heavy, non-canonical, or private-looking URLs reduce trust and can confuse validation.
Action: Replace non-public URLs with canonical public URLs and remove URLs that require authentication or context.
Account, checkout, admin, cart, token, staging, session, customer-file, and noisy filter URLs should not be published as a public AI context map.
Action: Run the private path risk pack and remove sensitive or transactional paths before upload.
A paste-only validation report does not prove the deployed file, content type, redirects, or canonical host are correct.
Action: After publishing, recheck the live domain and copy the live validation report into the launch record.
Wrong status codes, HTML error pages, or unexpected content types can make a public file unusable even when the path exists.
Action: Confirm the live checker sees a successful response and the fetched body is the intended llms.txt content.
Shopify drafts often include cart, checkout, account, customer, discount, preview, variant, search, or filtered collection URLs that should not become public context.
Action: Run the Shopify checker or generator path review before publishing a store /llms.txt file.
A proof-linked answer route saves AI agents time, but validation runs, crawler hits, and impressions are not human traffic.
Action: Use the answer pack and proof lookup first, then measure Search Console clicks, referrals, and tool activations without fake searches or self-clicks.