Which AI search engine gets AI pricing right?
Same pricing questions, every major AI-answer engine, one neutral WASM referee. We pull each engine's answer via SerpApi, extract its claims, and score them against official vendor pages. No LLM judge that can be talked into anything — a deterministic verifier with no prompt.
Scoreboard populates as probes complete
The probe harness (scripts/probe-crossengine.mjs) fires the same question at
Bing Copilot, DuckDuckGo, and Brave AI mode through SerpApi, archives the full payload,
and runs every extracted claim through FactJudge against official vendor pages.
Google's column is already live on /ai-audit.
Budget note: each engine-question pair costs exactly one guarded search, then is cached forever.
Scoring method
Identical to our audit pipeline: claims → verbatim check → WASM NLI (c026port,
sha-pinned) against fetched official vendor pages → SUPPORTED / MIXED / UNRESOLVED_JUDGE /
CONTRADICTED per claim. Engine answers are archived unmodified in
data/probes/crossengine/; nothing is paraphrased before grading.
Engines currently probed: Bing Copilot · DuckDuckGo · Brave AI Mode — alongside Google AI Overview from the main audit pipeline.