Settings

Tune Fact It for how you work

Sensitivity

Higher sensitivity means more claims extracted per response and more evidence per claim.

History sync

Off by default: your checks stay on your device. Turn this on to sync them to your account so they show up in Historyacross surfaces. Turning it back off deletes what we’ve stored.

Sync my check history to my account

Verifier model

The default is tuned and eval-gated for everyone. You can run the verifier on another model with your own API key (Account → Browser keys) — non-default models are experimental and not eval-gated. The extractor always stays Claude Haiku.

Evidence providers

Fact It classifies each claim by subject and routes it to that domain’s specialist source plus a general web tier — automatically. This is one tuned pipeline for everyone (not a per-user setting), so quality stays measured and reproducible.

General tier

Tavily
AI-native general web search.
keyed
Brave Search
Independent general web index. Metered: $5 per 1k requests.
keyed
Perplexity Sonar
Citation-finding general search.
keyed
Wikipedia
Evergreen & biographical reference. Always available.
always on

Specialized — routed by claim domain

Europe PMCmedical
Biomedical & life-sciences literature.
always on
openFDAmedical
FDA-approved drug labels: indications, contraindications, warnings.
always on
SEC EDGARfinancial
US public-company filings.
always on
eCFR / Federal Registerlegal
US federal regulations and agency rulemaking, primary source.
always on
CourtListenerlegal
US case law & court opinions.
keyed

Whitelist

Which AI hosts Fact It should attach to. Default-off on sensitive domains.

ChatGPT (web)
on
Claude.ai
on
Gemini
off
Perplexity
off

Extractor

Claim extraction is a fixed, cheap, structured-output step — the same for everyone (the verifier is selectable above).

Claude Haiku 4.5

Fixed.