Verified against Perplexity Pro · 2026-07-29
Fact-check one specific claim with a verdict you can't hedge around
A closed-verdict fact-checking prompt that requires quoting the exact claim, weighing real counter-evidence, separating bundled assertions, and committing to one of six named verdicts rather than a vague "it depends."
The prompt
Ready to copy — highlighted parts are example details you can swap.
Fact-check this exact claim. Quote it back to me first to confirm you're checking the actual wording, not a softened or strengthened paraphrase of it — even a small change in wording can turn a checkable claim into a different one. CLAIM (verbatim) "Remote workers are 23% more productive than in-office workers, according to a Stanford study." WHERE I SAW IT A LinkedIn post from a remote-work software vendor CONTEXT THAT MIGHT MATTER The post doesn't specify which Stanford study or what year. PROCESS 1. Confirm the precise, checkable assertion in the claim. If it bundles more than one assertion — a statistic and an attribution, or a cause and an effect — separate them explicitly and check each one on its own, since a claim can be half-right and half-wrong and a single blended verdict would hide that. 2. Find the best available evidence for the claim and the best available evidence against it. If you genuinely cannot find real counter-evidence after a real search, say that explicitly rather than inventing a weak devil's-advocate case just to appear balanced. 3. Give one verdict per assertion, using only these six labels: True, False, Mostly True, Mostly False, Mixed, or Unverifiable. Do not introduce a different label or soften a label with extra qualifiers. 4. Justify each verdict in two or three sentences citing the actual sources and their actual content — not the general reputation of whoever made the claim, and not what "seems plausible" given how the claim is phrased. 5. If the claim attributes itself to a specific study, report, or person, separately verify that the attribution itself is accurate — a real fact wrongly attributed to a specific source is a different kind of error than the fact being wrong, and both should be named if both apply. Do not soften a False verdict into "this is disputed" or "context is needed" if the weight of verifiable evidence actually points clearly one way — reserve "Mixed" for cases where the evidence itself is genuinely divided, not for cases where you'd simply rather not commit.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Quoting the exact claim back before checking it matters because the search step queries on the literal wording — a paraphrase can drift toward a softer or stronger version of the claim that was never actually asserted, and the model ends up 'verifying' a statement nobody made while the actual claim goes unchecked. Splitting bundled assertions apart is what catches the single most common shape of a misleading claim: a real number attached to a wrong or exaggerated attribution, where the number checks out and the source doesn't, or vice versa — collapsing both into one verdict would either wrongly clear the whole claim on the strength of the real number, or wrongly condemn a real number because of a bad attribution wrapped around it. Forcing a closed set of six verdict labels prevents the most common failure mode of open-ended fact-checking, where a model defaults to a non-committal 'this is complicated' that sounds careful but commits to nothing a reader can actually act on — reserving 'Mixed' specifically for genuinely divided evidence, rather than as an escape hatch, is what keeps that label meaningful instead of becoming the default answer to anything uncomfortable. Requiring genuine counter-evidence, and permitting the model to say none exists rather than manufacturing a weak opposing case, guards against a subtler failure — synthetic balance, where a model invents a flimsy 'on the other hand' argument purely to look even-handed, which actually makes a true claim look more contested than the evidence supports and does real damage to a reader trying to gauge how settled something is. The separate attribution check exists because misinformation research consistently finds that a real, correctly-measured statistic gets detached from its original source and reattached to a more prestigious-sounding one as it spreads — checking the number and the attribution as two distinct facts is the only way to catch that specific and very common drift.
What you get back
Claim confirmed: "Remote workers are 23% more productive than in-office workers, according to a Stanford study." Assertion 1 (the figure): a specific study found a 23% productivity gap favoring remote work. Verdict: Mostly False. A widely cited 2015 Stanford study (Bloom et al.) found a roughly 13% productivity gain for a specific call-center task under a structured work-from-home arrangement, not a general 23% figure. Assertion 2 (the attribution): the 23% figure is attributed to "a Stanford study." Verdict: Unverifiable. No Stanford study reporting 23% could be located; this figure appears to be a commonly repeated distortion of the original 13% number, possibly conflated with a different, unrelated statistic.
Verified against
Perplexity Pro Sonar Pro · 2026-07-29
Changelog
- 2026-07-29 — Initial publish, verified against Perplexity Pro Sonar Pro.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
