/ ACCURACY JUDGE
Is what AI says about you actually true?
The Citation Probe tells you whether AI cites you. The Accuracy Judge tells you whether what AI says about you is correct. Every claim each engine makes gets fact-checked against your own page — supported, unsupported, or contradicted.
/ The question
AI engines hallucinate. They confidently misstate your pricing, wrong-attribute your features, conflate you with a competitor, or invent quotes you never gave. When a buyer asks ChatGPT 'what does this company do?' and the answer is wrong, you lose the deal silently. The Accuracy Judge runs every engine response through a structured claim-by-claim check so you catch hallucinations the moment they appear.
/ How it works
Methodology
- 01
Persist the full engine answer
Every Citation Probe run now stores the complete answer text each engine returned about your brand. We don’t throw away the 90% the existing probe doesn’t need — it’s the raw material the judge feeds on.
- 02
Extract structured claims
Claude reads each engine answer and pulls out every factual assertion as a discrete claim: "Acme charges £49/mo", "Acme was founded in 2018", "Acme integrates with Stripe". One verdict per claim, not one verdict per engine.
- 03
Verify against your page
Each claim gets verified against your audited page content. Supported (your page backs it), unsupported (silent — neither confirms nor denies), or contradicted (your page actively disagrees). Contradictions are the urgent ones.
- 04
Aggregate + cache
Report is cached per probe run so the dashboard renders instantly on repeat visits. Re-run on demand when you ship a content change you want fact-checked.
/ Why it’s different
Versus everything else in the category
Question
Other tools
AI Search Ready
What does it check?
Citation count + rough sentiment
Claim-by-claim factual accuracy of each engine response
How does it verify?
Manual eyeballing of excerpts
Structured LLM judge against your own page content
What about contradictions?
Treated as noise
Surfaced loudly — the urgent fix path
Can you re-check after a fix?
No persistence layer
Cached + re-runnable on demand
/ What you get
In your dashboard
- →
Per-engine accuracy score — supported / unsupported / contradicted ratio
- →
Per-claim verdict with the source line in the engine answer
- →
Contradiction queue ranked by severity — what to fix first
- →
Re-run after edit — see whether your content change closed the gap
- →
Engine consensus check — which engines agree, which one is the outlier
/ GET STARTED