What Should You Do When Perplexity Cites the Wrong Source?

The most common case is a mismatch, where the linked page exists but never actually states the claim attributed to it. This happens when the model paraphrases loosely from a page discussing the topic generally without confirming the specific number, date, or fact in the answer. It reads as a citation but offers no real support for the statement.

Other cases include stale citations, where the source once supported the claim but has since changed or been deprecated without an update elsewhere, and competitor credit, where a rival page states your original fact without linking back. A fourth case is a missing primary source, where no page clearly owns the claim, so Perplexity picks whatever ranks or reads as most authoritative.

Log every incident with the exact prompt text, the full generated answer, every cited URL in order, and the specific sentence or number the citation is supposed to support. Note the date and time, since answers shift across sessions, and note any account or location context if you tested while logged in, since personalization can affect results.

Screenshot the full answer with citations visible, and separately archive the cited page's content at that moment, since pages change and you may need to show what it said when cited. Then ask one question per citation: does this page, as written now, actually state the claim? A no confirms a real error worth acting on.

Perplexity documents PerplexityBot as designed to surface and link websites in search results, and Perplexity-User as supporting user actions that may visit a page to help provide an accurate answer with a link, per Perplexity's crawler documentation. Both depend on reaching and parsing your page at all, so a blocked, script-heavy, or thin page simply is not available to be chosen.

Specificity matters too. A page stating the exact claim plainly, with a date and clear entity names, is easier to cite than one implying the same fact across vague narrative paragraphs. If a competitor's page states the number more directly, that page becomes the more retrievable option, regardless of who originated the fact.

Write the page so it states the exact answer in the first few sentences, includes the date the fact was true, names the specific entities involved, and avoids burying the claim in marketing language. Add a comparison or proof section for competitive claims, since vague pages are the ones most likely skipped in favor of a sharper source.

Confirm the page is reachable, meaning it is not blocked by robots rules, does not require login, and renders core text without depending entirely on client-side scripts. Then link it internally from FAQ, glossary, or comparison pages, establishing which page is the canonical answer, similar to the logic behind a structured source map.

First, judge whether the third-party page is factually wrong or just less specific than what you could publish. If wrong, a direct correction request is worth trying, though a fast edit should not be expected. If it is simply thinner, publish a sharper, more specific version of the fact yourself and let retrieval quality work over time.

Second, identify the page type Perplexity keeps reusing for that query, whether a comparison roundup, directory listing, or review aggregator, and pursue inclusion on that type rather than trying to replace it. This is slower than fixing your own page but realistic when you do not control the cited source.

Rerun the exact prompt after your changes, ideally after giving crawlers time to revisit the page, and compare new citations against your original log entry. Label the outcome clearly: resolved if your source now appears and supports the claim, partially resolved if the citation changed but is still imperfect, or unresolved if nothing changed.

Keep testing on a schedule rather than once, since answers can shift as pages update elsewhere on the web. Manually rerunning dozens of prompts across categories is tedious without a system, which is the gap a repeatable visibility audit is meant to close.