Verified against Perplexity Comet · 2026-08-09
Fact-check a specific claim by making Comet show its browsing trail, not just its verdict
A verification prompt that requires Comet to name what it actually checked, where the primary source lives, and what a good-faith opposing reading would look like — instead of a flat true/false with no visible work.
The prompt
Ready to copy — highlighted parts are example details you can swap.
Fact-check the claim below the way a careful editor would before it goes into something I'm putting my name on — not by summarizing what search results say about it, but by actually tracing it back to where it originates. CLAIM TO CHECK "A recent study found that remote-first companies have 40% lower voluntary turnover than office-mandated ones." WHERE I ENCOUNTERED IT A LinkedIn post summarizing an HR newsletter, which itself summarized a longer report. HOW STRICT TO BE High bar — this is going into a board deck recommending a policy change, not a casual Slack message. STEPS 1. Identify what kind of claim this is — a statistic, a quote, an event, a causal assertion — because each needs a different check. A statistic needs its original dataset or methodology; a quote needs the original recording or transcript, not a paraphrase in a later article; an event needs at least one primary account, not just coverage of coverage. 2. Trace it to the earliest identifiable source you can actually reach, and note how many steps of paraphrasing sit between that source and the version I encountered (A LinkedIn post summarizing an HR newsletter, which itself summarized a longer report.). More than one or two steps of paraphrasing is itself a risk signal worth naming. 3. Check whether the claim has been corrected, retracted, or disputed anywhere since it was first published — a claim that was true when written but has since been walked back should be flagged as such, not reported as still-current. 4. State the strongest good-faith version of a different reading of the same underlying facts, if one exists — not a strawman, but a real alternative interpretation someone informed could hold. VERDICT Give one of: Confirmed / Confirmed with caveats / Unconfirmed — can't verify / False / Partially true, misleadingly framed. Do not use "true" or "false" alone if the honest answer is closer to "technically accurate but missing context that changes its meaning" — that distinction is the actual point of this check. SHOW YOUR TRAIL List the specific pages you checked, in the order you checked them, and what each one told you — including any that turned out to be a dead end or a repeat of an earlier source.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Asking for a flat true/false invites the single most common fact-checking failure: a claim that is technically accurate but stripped of the context that made it meaningful reads as 'true' under a binary system, when the honest and useful answer is that it's misleading. The five-option verdict scale exists specifically to make that distinction sayable instead of forcing it into a box it doesn't fit. The paraphrase-distance step matters because claims degrade predictably as they pass through summarization layers — a hedge in the original study becomes a flat statement two summaries later, and a model asked only 'is this true' has no structural reason to notice how many hops separate what it's checking from where the number actually came from; asking it to count the hops as a named step forces that degradation into view. Requiring Comet to check for retractions or corrections addresses a specific weakness of confident single-pass answers: a claim can have been accurate at publication and false now, and a check that only verifies the original claim's initial accuracy without checking what happened since will happily confirm something that's since been walked back. The 'show your trail' requirement is not cosmetic — a browser agent that can actually navigate to primary sources should be held to a different standard than a search-snippet summary, and forcing it to list the specific pages checked, in order, makes it possible to catch the difference between genuine verification and a plausible-sounding restatement of what search results already implied.
What you get back
Claim type: statistic. Traced to: a 2025 workplace-analytics firm report (not the HR newsletter or LinkedIn post, both of which paraphrased it). Paraphrase distance: 2 steps, and the LinkedIn post dropped the original's caveat that the sample was self-selected companies already committed to remote work. Since publication: no retraction found, but a 2026 follow-up from a different analytics firm found a smaller 22% gap using a broader sample. Verdict: Partially true, misleadingly framed — the 40% figure is real but comes from a non-representative sample and a more recent, broader study found a substantially smaller effect.
Verified against
Perplexity Comet 2026 browser agent · 2026-08-09
Changelog
- 2026-08-09 — Initial publish, verified against Perplexity Comet 2026 browser agent.
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
