Verified against Perplexity Comet · 2026-08-11
Find where your sources actually disagree before you write something that pretends they don't
A cross-source audit prompt for Comet that hunts specifically for contradictions between the pages you've gathered — on numbers, dates, causes, and attribution — rather than producing a synthesis that smooths disagreement into false consensus.
The prompt
Ready to copy — highlighted parts are example details you can swap.
Compare the sources below and find every place where they actually disagree with each other — not where they simply cover different aspects of the same topic, but where two or more say something that cannot both be true as stated. SOURCES TO COMPARE four news articles and one regulatory filing about the same product recall, all currently open in tabs TOPIC THEY'RE ALL ABOUT the scope and cause of a manufacturer's recent product recall WHY THIS MATTERS RIGHT NOW deciding by Friday whether our company needs to issue our own customer notice WHAT COUNTS AS A CONTRADICTION Look specifically for: different numbers reported for the same metric (even if close — note both and the gap); different dates given for the same event; different causal explanations offered for the same outcome; different people or organizations credited or blamed for the same thing; and one source stating something as settled that another source frames as disputed or unconfirmed. Do not report two sources as contradicting each other if they're actually just covering different scopes or time periods that happen to look like disagreement at a glance — check the scope before flagging it as a real conflict. FOR EACH CONTRADICTION FOUND State the two (or more) conflicting claims side by side, name which source said which, and give your best assessment of why they might disagree — different methodology, different time period, one source being outdated, one source having a stake in a particular framing, or genuinely unresolved uncertainty in the underlying facts. Rank each contradiction by how much it would matter to someone relying on the scope and cause of a manufacturer's recent product recall for deciding by Friday whether our company needs to issue our own customer notice — a contradiction on the core number matters more than a contradiction on a minor supporting detail. WHAT NOT TO DO Do not resolve a contradiction by picking a side and reporting only the version you find more credible unless you also show the rejected claim and say explicitly why you're discounting it. Do not average two conflicting numbers together and present the average as if it were a finding — an average of two disagreeing sources is not itself evidence of anything. CLOSE WITH A one-line statement of whether the sources are in strong overall agreement with a few edge disagreements, or whether the disagreement is central enough that the scope and cause of a manufacturer's recent product recall should be treated as genuinely unsettled right now.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
A model asked to simply synthesize multiple sources will default to smoothing over disagreement into a single coherent-sounding narrative, because fluent synthesis is the path of least resistance and a paragraph that acknowledges no tension reads more confidently than one that surfaces it — this prompt exists specifically to counteract that pull by making contradiction-hunting the primary task instead of a byproduct of summarizing. The explicit list of contradiction types (numbers, dates, causal attribution, credit/blame, settled-versus-disputed framing) matters because an unguided search for 'disagreement' tends to catch only the most obvious cases and miss subtler ones, like one source treating something as established fact that another source explicitly flags as still under investigation — naming these categories forces a systematic pass rather than an impressionistic one. The instruction to check scope before flagging a false contradiction addresses a real failure mode: two sources covering different time windows or different sub-populations of the same broader topic will often look contradictory at a surface read when they're actually both accurate about different things, and treating every surface-level mismatch as a real conflict would flood the output with false positives that bury the genuine ones. Banning the average-the-numbers move matters because averaging two disagreeing figures manufactures a false sense of precision — the midpoint of two contested numbers has no evidentiary status of its own, and presenting it as a finding would be a worse outcome than honestly reporting the disagreement and letting the reader decide how to weigh it.
What you get back
Contradiction 1 (high stakes): Article A reports the recall affects 12,000 units; Article B and the regulatory filing report 45,000 units. Article A appears to cite only the initial press release before the scope was expanded — likely outdated rather than a genuine disagreement. Contradiction 2 (lower stakes): Two articles differ on whether the defect was a manufacturing error or a design flaw; this appears to be a genuine unresolved disagreement, since the regulatory filing itself states the cause is still under investigation. Overall: the units-affected number should be treated as unsettled until the regulatory filing's figure is confirmed as final, and any customer notice should use a range rather than a single figure.
Verified against
Perplexity Comet 2026 browser agent · 2026-08-11
Changelog
- 2026-08-11 — 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
