How we determine a fact score
Every Infact verdict comes with a number from 0 to 100. Here's how that number is built — the evidence behind it, what it means, and where it stops.
A fact score is our estimate of how factually accurate a claim is, from 0 (false) to 100 (true) — not a decree. We weigh the evidence we find and what's well-established knowledge, then show our verdict, the sources behind it, and a separate confidence level so you can judge for yourself. We'd rather be honest about uncertainty than falsely certain.
From a claim to a score
Extract
We pull the specific, checkable claims out of your text, link, or image.
Gather evidence
We search independent web sources and professional fact-check databases.
Weigh signals
Multiple signals are combined — not a single opinion.
Score & calibrate
A 0–100 score is computed and tuned against real-world feedback.
Verdict
The score maps to a clear label, from True to False.
The four signals behind a score
No verdict rests on one source or a single model's guess. Each claim's score blends several independent signals, so one weak input can't swing the result on its own:
★Professional fact-checks
Ratings from established, independent fact-checking organizations when they've already examined a matching claim.
◎Web consensus
How strongly independent, reputable sources on the open web agree or disagree with the claim.
⚑Source quality
How credible and authoritative the supporting sources are — not just how many there are.
✦AI reasoning
The model's own assessment of whether the claim is true — weighing context, nuance, and well-established knowledge — checked against the gathered evidence rather than overruled by a thin search result.
What the score means
Alongside the truth score, each check carries a confidence level — a separate measure of how certain the analysis is, based on how much clear, relevant evidence was available. High confidence and a high score is a strong verdict; low confidence signals "treat this as a lead, not a last word."
Built-in humility
Cautious when evidence is thin
For well-established facts we give a clear verdict even if a live search comes up short. But when a claim is obscure or contested and we can't corroborate it, we hold back to "Mixed" rather than guess — and the confidence level says so.
Opinions aren't facts
Predictions and matters of opinion aren't scored as if they were verifiable facts.
Recency matters
Older fact-checks carry less weight over time, so a verdict reflects what's known now — not only what was true years ago.
We learn from feedback
When people consistently tell us a kind of verdict feels off, the system gently recalibrates toward reality.
Never 100% certain
Even a strongly-supported claim is never asserted at absolute certainty. Calibrated humility beats false confidence.
Deterministic, not a black box guess
The final score is computed by fixed rules from the evidence signals — the same inputs always produce the same score.
What a fact score is not
A high score means the claim is, to the best of our analysis, factually accurate — not that it's eternally, unconditionally true. A low score means it's inaccurate as stated, which isn't the same as "the opposite is true."
Infact is a fast, evidence-based first pass, not a replacement for expert judgment on high-stakes decisions. When a claim is genuinely contested or evidence is scarce, we tell you so rather than manufacturing certainty. To keep the system honest and improving, the exact weightings, thresholds, and models behind the score are refined continuously — but the principles on this page don't change.
See it in action
Check any claim and you'll get the score, the verdict, and the evidence behind it.
