When ChatGPT gets your company wrong, you can't email it. But you can publish the record it reads next time. Every Entidex entity carries a public, machine-readable Knowledge Statement — verified facts, corrections to common AI claims, cryptographically signed, discoverable by crawlers and agents alike. It's not optimization. It's the source of truth, published where the machines look.
A canonical record assembled from authoritative sources — each fact carrying where it came from and when it was observed, anchored to a stable entity identity.
Where the AI engines currently say something wrong, the statement publishes the verified value alongside the divergence — the correction delta a model can read and reconcile.
Served as Markdown, schema.org ClaimReview, or JSON, cryptographically signed, and discoverable from the site-root llms.txt and a ground-truth endpoint for agents and CI.
Most tools tell you what AI says. Entidex also gives the AI layer a record to get you right — read it, and submit corrections back through a review-gated, append-only channel.
For developers, the statement ships as the Grounding Packet — one small, deterministic, signed payload (facts + corrections + provenance) fetched at the top of an agent or RAG workflow, on the free API tier. Grounding, not enrichment.
Entities monitored on Watch and above get an embed endpoint — a public, always-fresh verified-fact block you publish on your own domain: schema.org JSON-LD for your page markup, and a Markdown block for your own llms.txt. It rebuilds from the verified record on every fetch, so the facts on your site never go stale the way a hand-maintained llms.txt does.
The Knowledge Statement is the signed receipt; the Truth Gap page is the diagnostic that names every gap the receipt is closing — per-engine accuracy, tracked claims with three-way verdicts, and narrative divergence. Live examples:
You can’t email a model — but you can publish the record it reads next time. The Knowledge Statement is a public, machine-readable, cryptographically signed record of verified facts plus corrections to the claims AI currently gets wrong, discoverable from the site-root llms.txt and a ground-truth endpoint.
It is served as Markdown, schema.org ClaimReview, or JSON, and advertised from llms.txt, so crawlers and agents can fetch it directly. Each correction publishes the verified value alongside the divergence — the delta a model can reconcile against its own answer.
No. It is not optimization — it is the source of truth, published where the machines look, with provenance on every fact and a review-gated, append-only correction channel back. Entidex observes and publishes the record; it never manipulates a value to game a result.
Read the developer docs for the grounding endpoints, or see the live llms.txt discovery index.
All signals are observational, not definitive. Entidex observes — never manipulates.