How accurately does AI know your brand?
Not how visible you are, or how it feels about you — whether what AI says is actually true. Entidex scores every engine’s knowledge of your entity against a verified record, and shows the exact fields each one gets wrong.
Ask ChatGPT, Claude, Gemini, Perplexity and Grok about your company and each will answer with total confidence. Some of what they say will be right. Some will be a founding year that is off by two years, a headquarters that moved, a product you sunset, a founder who left, or a competitor they have quietly merged you with. The fluency hides the error. LLM Knowledge Accuracy measures the error directly.
One number per engine, against the truth
For each of the five AI engines, Entidex records what it claims about you, then checks each claim against your verified record. The score is simple and honest:
Matches
Facts the engine states that agree with the verified record. These build the accuracy score.
Contradictions
Facts the engine states that disagree with the record — the exact fields it gets wrong, listed per engine.
Unaware
Fields the engine never mentions — a coverage gap, counted separately so it never unfairly drags the score down.
Accuracy is matches ÷ (matches + contradictions), carried with a 95% confidence interval so a thin sample never reads as certainty. The overall figure is exactly the accuracy published on your Entity Knowledge Statement — one coherent number, not a second opinion. And it costs nothing to compute: it reduces the knowledge probes Entidex already runs, rather than asking a model to grade itself.
Wrong is fixable — once you can see it
A per-engine list of wrong fields turns a vague worry (“AI gets things wrong about us”) into a work order. Publish the verified record so crawlers and retrieval systems can ingest it, strengthen the authoritative sources behind each disputed fact, and then watch the gap close. Accuracy is the diagnosis; the Knowledge Statement and correction channel are the treatment, and Information Lag tells you how long each engine takes to catch up.
Provenance under every claim
Every match and mismatch is anchored to the observation that produced it — which engine, when, and against which verified field. Nothing is asserted without evidence, so the accuracy score stands up to scrutiny from a customer, a journalist, or your own team.
Go deeper with Entidex
The full entity intelligence platform — beyond Explore Entidex
- Continuous multi-source entity observation
- Alerts when sources diverge or drift
- Cross-surface consensus + visibility over time
- Evidence-anchored intelligence reports
Frequently asked questions
What is LLM Knowledge Accuracy?
It is a per-engine score of how much of what an AI engine claims about your entity actually matches your verified record. Entidex probes each of the five AI engines — ChatGPT, Claude, Gemini, Perplexity, Grok — records what they assert about you, and checks each claim against the ground-truth record. Accuracy is matches ÷ (matches + mismatches); a field the engine never mentions is a coverage gap and does not count against it. The overall number is the same accuracy figure published on your Entity Knowledge Statement — not a competing metric.
How is it different from sentiment or visibility?
Visibility asks whether AI surfaces you; sentiment asks how it feels about you. Knowledge Accuracy asks a colder question: is what the AI says actually correct? An entity can be highly visible and positively described while the founding year, headquarters, product line or leadership the engine repeats are wrong. Accuracy isolates the factual layer.
Which fields does it check, and can I see what each engine got wrong?
Yes. The per-engine breakdown lists the exact fields each engine gets wrong (a contradiction against the verified record) and the fields it is simply unaware of. That turns "AI is wrong about us" into a specific, fixable list — the same list that feeds the correction loop.
How accurate is the accuracy score itself?
Every score carries a 95% Wilson confidence interval over the decidable checks, so a small sample never reads as false precision. The computation is deterministic and runs at zero model cost — it reduces the per-engine knowledge probes already stored, rather than asking a model to grade itself.
What do I do once I know an engine is wrong?
Publish and strengthen the verified record. Entidex pairs the accuracy score with the Entity Knowledge Statement (a machine-readable, citable record AI can ingest) and tracks how long each engine takes to catch up — see Information Lag. Accuracy tells you the gap; the Knowledge Statement and correction channel close it.