Availability checked November 4, 2026: ProtoScrape’s Perplexity engine is in validation and is not currently available for customer collection. This article describes a reporting method, not an available Perplexity API integration. Google Search, Google AI Mode and ChatGPT are available in early access. They provide separate observations, not Perplexity results.
A citation study asks which sources appear in answers to a defined set of questions. It can help an SEO team understand where its own pages appear, which third-party sources discuss the same topic and which observations deserve a closer review.
Perplexity describes source links and follow-up conversations in its Pro Search documentation. It also advises checking the referenced sources. A source link is a useful starting point for review; its presence alone does not show that every nearby claim is accurate or that the source endorses a product.
Start with a defined question set
Choose a set of questions that represents the audience and decision you want to study. Keep the question wording, topic group and version in your dataset. Separate unbranded discovery questions from questions that explicitly ask about your company.
Record whether the answer came from a fresh conversation or followed earlier questions. The same text can have a different meaning inside an ongoing conversation. If an observable search mode or account context changes between runs, flag that difference before interpreting the results as a time trend.
Retain the answer and its source list
For each authorized observation, store the exact question, collection time, answer text and the source URLs that were actually returned. Keep source titles when available and retain the original URL before normalizing it for analysis.
Do not invent a source because the answer mentions a familiar publication. A textual reference, a clickable citation and an inference about where information came from are different forms of evidence. Your dataset should preserve that distinction.
Separate three useful measures
| Measure | Counting rule |
|---|---|
| Brand mention rate | Complete answers naming the brand divided by complete answers in the same cohort |
| Domain citation rate | Complete answers with at least one source on the target domain divided by complete answers |
| Unique cited pages | Distinct source-page URLs after a documented normalization rule |
Count a domain once per answer for its citation rate, even when an answer links to several of its pages. Retain all the page links for a separate page-level analysis. Removing duplicate appearances from one metric should not erase evidence needed by another.
Use conservative URL normalization. Removing a fragment may help group links to the same page; removing every query parameter can merge pages that actually contain different content. Record your normalization rules and preserve the original values so you can revise the analysis later.
Review what the citation supports
For a sample of observations, open the cited page and identify the passage supporting the relevant claim. Record whether support is direct, partial, unrelated or unclear. A reviewer should also check the publication date and whether the statement describes the intended company or product.
For example, an answer may cite your documentation for a feature while mentioning a competitor in a comparison. Counting the citation as a recommendation for your product would overstate the evidence. Preserve a short review note with the observation instead of relying on a single positive-or-negative label.
Make gaps visible in the report
An illustrative study might schedule 20 observations, complete 16 and find a target domain in four complete answers. Report 80% collection coverage and a 25% domain citation rate among completed observations. Those figures describe that sample, not all Perplexity answers.
Distinguish an empty source list on a complete answer from an incomplete or failed collection. Do not convert failures into confirmed absences. Keep separate totals for each engine when comparing Perplexity research with ChatGPT or Google AI Mode research.
What you can build with ProtoScrape now
ProtoScrape’s available Google AI Mode engine provides structured answer text and available citations through the job API. That lets you develop a source-review workflow and test your reporting logic using clearly labeled AI Mode observations. It does not provide a substitute measurement of Perplexity visibility.
Check the current engine catalog before scheduling jobs. While perplexity is not_ready, its customer jobs cannot run. For the supported workflow, read the API documentation, then use the citation analysis guide and agency reporting guide to structure your evidence. ProtoScrape is independent and is not affiliated with Perplexity.