CiteLint

A CLI that scores your docs site against the structured data actually correlated with AI-answer citations, and drafts the JSON-LD — instead of another llms.txt file nobody fetches.

Problem

A lot of teams shipped an llms.txt file in 2026 because it was pitched as the new sitemap.xml. Ahrefs checked 137,000 sites on June 15, 2026 and found that among the ones that published the file, 97% got zero requests to it — and most of the traffic that did show up was SEO crawlers and validators, not the AI systems it was supposedly written for. SE Ranking went further with a roughly 300,000-domain study in May 2026: once you control for site authority, schema density, and content recency, llms.txt shows no measurable citation lift at all. Teams maintain a file that's doing nothing, and nothing tells them that.

Target user

A solo developer or small team maintaining a docs site or product blog, who added llms.txt after reading one of this year's "how to be visible to AI" guides and has no way to check whether it's working. Job to be done: before spending another afternoon on AI-visibility busywork, find out which pages are actually missing the markup shown to matter, and generate it.

MVP scope

• Crawl a sitemap.xml or a local build directory (Next.js, Astro, Docusaurus, Hugo output) and parse each page's existing JSON-LD.
• Score each page against the four schema types SE Ranking's study tied to citation lift — FAQPage, ClaimReview, Speakable, Organization sameAs — and flag which are present, valid, or missing.
• Draft FAQPage JSON-LD by matching heading-and-following-paragraph pairs, and flag the paragraph it would mark speakable, for a human to accept or edit. Nothing auto-commits.
• --llms-txt-audit flag that checks an existing llms.txt against the real sitemap and reports drift, instead of treating the file's presence as done.
• CI mode: exits non-zero when a new page ships without required schema; SARIF output for GitHub code scanning.
• Ships as a single Go binary plus a GitHub Action wrapper — no Node or Python runtime required to run it in CI.

Monetization

Freemium. The CLI is MIT-licensed and free. A paid Cloud tier ($19/month) re-crawls the site weekly, tracks the citation-readiness score over time, and diffs it against the prior scan so a regression surfaces before a quarter goes by. A team tier ($99/month) covers agencies running the scan across client sites.

Why now

The correction is happening right now, not hypothetically. SE Ranking's May 2026 study found FAQPage schema correlated with a 34% citation lift on Perplexity, ClaimReview with 41% on Google AI Mode, Organization sameAs with 22%, and Speakable with 18% on AI Mode — while llms.txt showed nothing once other factors were controlled for. Ahrefs' June 15, 2026 check confirming the 97%-unread number landed three weeks after that. Teams that followed the llms.txt advice are now the exact audience asking what to do instead, and no tool answers that question with a scan and a generated fix.

Risks & open questions

• The citation-lift numbers come from an observational study, not a controlled experiment — if AI platforms change ranking signals, the premise ages out fast.
• Auto-drafting FAQPage schema from arbitrary HTML is pattern matching on heading structure. A wrong Q/A pairing is worse than no schema, which is why nothing auto-commits — but that also means the tool can't promise a one-command fix, only a one-command draft.
• Ahrefs, Semrush, and Screaming Frog already audit structured data as one feature inside much bigger products. A free standalone CLI needs the git-native, CI-gating angle to not just be a worse Screaming Frog.
• Demand rides on continued anxiety about AI answer engines. If that anxiety cools, so does the reason to install this.
• The four-schema ruleset needs to track the underlying study's assumptions as they shift, which is ongoing maintenance — the same staleness problem this tool exists to call out in llms.txt.

Next step

Crawl 10 real docs sites — five with an existing llms.txt, five without — publish their citation-readiness scores, and see whether any of the ten site owners engage enough to wire the CI gate into their repo.

Sources

• https://inite.ai/en/blog/is-llms-txt-dead-2026 — SE Ranking's ~300,000-domain May 2026 study: llms.txt adoption/citation-lift findings and the per-schema lift figures (FAQPage, ClaimReview, sameAs, Speakable).
• https://emarketed.com/aeo/llmstxt-files-go-unread-2026/ — Ahrefs' June 15, 2026 check of 137,000 sites: 97% of published llms.txt files receive zero requests, and most of the traffic that does arrive isn't from AI systems.