Every AI engine tells a slightly different story about your brand, and none of them tells exactly the verified one. The Truth-Gap Score measures that distance — claim by claim — between what the engines assert and what the record actually supports.
The verified, citable record for the entity — the anchor every other signal is measured against.
What ChatGPT, Claude, Gemini, Perplexity and Grok actually say when asked — read directly, not inferred from a single prompt.
The live conversation, which often moves a day or more before the engines and the press catch up.
Each claim about an entity is scored across all three axes and given a coded verdict. Confirmed means the axes agree. Disputed means they diverge — usually an engine repeating a fact the verified record contradicts, which is the clearest thing to correct. Died means a claim that was once supported and now has no backing anywhere — the quiet decay that flat accuracy numbers miss.
No model is asked to decide what is true. The verdict is coded from where the axes line up, so the score itself carries no AI opinion — it is a measurement of agreement, not a judgement. That keeps the Truth-Gap gaming-resistant and auditable: every verdict points back to the sources that produced it.
Or read the full methodology · watch it live on the live demo.
The three-way gap (Authority ↔ LLMs ↔ Social), per-engine accuracy, and tracked claims for a live entity:
It is a measure of the distance between the verified record for an entity and the account an AI engine gives of it. A small gap means the engine is describing the entity accurately; a large gap means it is working from stale, weak, or fabricated associations. The score is computed per claim and rolled up per entity, so you can see exactly which facts have drifted rather than just a single accuracy headline.
Entidex reads three independent axes for every claim: the authority (the verified, citable record), the AI engines (ChatGPT, Claude, Gemini, Perplexity and Grok), and the live social conversation. Each claim is then labelled — confirmed (the axes agree), disputed (they diverge), or died (a once-supported claim that no source still backs). No model is asked to judge truth; the verdict is coded from where the axes line up, so the measurement itself carries no AI opinion.
Because the interesting failures live in the disagreements. A claim that authority confirms but engines have not yet adopted is a lag to close. A claim the engines repeat that authority contradicts is a correction to push. A claim trending on social before any authority confirms it is an early-warning signal. A two-way check collapses all three into one number; the three-way gap keeps them separable.
Related, but broader. A hallucination is one failure mode — an engine asserting something with no support. The Truth-Gap Score also captures staleness (a fact that was true and no longer is), divergence between engines, and claims that have quietly died. It is the full distance to the verified record, not just the count of invented facts.
Run a free scan. Entidex resolves your brand to its canonical entity, reads the verified record, and surfaces the claim-level gaps the AI engines currently hold — what they get right, what is stale, and what is disputed.