Verified against Perplexity Pro · 2026-07-26
Audit a Perplexity answer's citations before you rely on any of them
A follow-up prompt that re-checks every numbered citation in a prior Perplexity answer against the actual source text, catching claims attached to a citation that only loosely or partially supports them.
The prompt
Ready to copy — highlighted parts are example details you can swap.
Audit the accuracy of every citation in the answer below. This is a verification pass, not a rewrite — do not change, improve, or "clean up" the original answer's wording, and do not add new claims that weren't in the original. ORIGINAL QUESTION What drove Company X's revenue growth last quarter? ANSWER TO AUDIT (with its citation markers and links, unedited) "Company X's revenue grew 22% year over year [1], driven mainly by its enterprise segment [2]." with [1] and [2] linking to a Q2 earnings article and a separate product-launch blog post. FOR EACH CITED CLAIM 1. Open the actual source behind the citation marker — do not rely on the citation's own preview text or headline as a stand-in for reading the relevant passage. 2. State whether the source's own text directly supports the specific claim it's attached to, partially supports it, or does not support it at all. 3. If a citation supports a more general claim, an earlier or later time period, or a slightly different number than the one it's attached to, say exactly what the gap is — do not mark it as supported just because it's topically related to the right subject. 4. Flag any claim in the answer that has no citation attached at all, and separately flag any citation number that appears to be reused for two claims that actually need different support. SEVERITY WEIGHTING Distinguish a cosmetic gap (a rounded number, a slightly loose paraphrase that doesn't change the meaning) from a material one (a wrong figure, a claim the source doesn't make, a date that changes the claim's currency). Weight the overall verdict on material gaps, not cosmetic ones — but list both. OUTPUT FORMAT A table: Claim | Citation # | Verdict (Supported / Partially Supported / Not Supported / Uncited) | Severity (Material / Cosmetic / N/A) | What the source actually says. End with one overall line on whether this answer is safe to cite further without independent verification, and if not, which specific claims are the blocking issue.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
AI search tools have a known drift where a citation marker is attached to the correct general topic but not to the specific number or claim in that sentence — for example a growth percentage cited to an article that discusses the same company's growth but reports a different figure or a different reporting period. Pasting the answer back in as a distinct verification task, rather than asking the same thread to double-check itself, narrows the job to something genuinely checkable: does this specific source text support this specific sentence, yes or no, rather than regenerating a fresh answer that could introduce new unverified claims while appearing to be a correction. The instruction to open the actual source text rather than trust the citation preview matters because Perplexity's own citation snippets are short and can look supportive on their own even when the full linked page contradicts or complicates the claim once read in context — a preview-only check would pass exactly the citations most worth catching. The severity-weighting step exists because a naive audit that treats every discrepancy as equally alarming trains you to ignore the output the second time, since most real answers have at least one cosmetic rounding difference; separating material gaps (a wrong figure, an unsupported claim) from cosmetic ones (a rounded number) is what keeps the audit's verdict actionable instead of noisy. And flagging reused citation numbers catches a specific and easy-to-miss failure — a single citation genuinely supporting one claim getting stretched to also back a second, adjacent claim it was never actually about, which reads as thorough sourcing but is really one piece of evidence doing double duty it can't support.
What you get back
Claim: "revenue grew 22% year over year" | Citation 1 | Partially Supported | Material | The earnings article reports 19% year-over-year growth for the quarter referenced, not 22% — the 22% figure appears to be a different metric (sequential growth) in the same article, likely conflated. Overall: not safe to cite further without checking the exact growth figure against the primary earnings release — this is the blocking issue, the enterprise-segment attribution in the second claim checks out cleanly.
Verified against
Perplexity Pro Sonar Pro · 2026-07-26
Changelog
- 2026-07-26 — 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
