how do i measure visibility in perplexity searches

Define whose visibility you are measuring, which product names count, which market matters, and which Perplexity surface you will sample. A consumer search session and an API response are different data sources. Record them in separate series even if you ask identical questions.

Write the scope in one sentence, for example: “This report covers our English language scheduling product in 15 clinic software comparison questions, checked manually in new Perplexity conversations once a week.” This is an example scope, not a result from this website. Add the collection dates, conversation conditions, and a link to your evidence folder.

The contract protects you from changing the definition after seeing the answers. If a product alias counts this week, apply the same alias rule to earlier answers before claiming a trend. If the model or access method changes, document a new series.

Count each answer at most once for each rate. Ten mentions inside one answer do not become ten visible answers. Define recommendation criteria explicitly: an answer can name a product as an unsuitable option or a warning, which should not count as a positive recommendation.

You can also record competitor presence and the domains supplying evidence. Keep the competitor set fixed for comparisons. If you calculate share of voice, publish your formula and comparison set. There is no reason to assume that another vendor’s similarly named score uses the same denominator.

Suppose you run 12 questions twice and receive 23 completed answers plus one timeout. In this hypothetical example, nine answers mention your brand, five cite your website, and four recommend the product. Mention rate is 9/23, or 39.1%; citation rate is 5/23, or 21.7%; recommendation rate is 4/23, or 17.4%.

Report coverage as 23 of 24 planned runs as well. Excluding failed requests from the visibility denominator keeps a technical failure from being counted as a negative answer, but the coverage figure makes missing evidence visible. Never silently discard an answer because it did not support the outcome you wanted.

With a sample this small, avoid treating small percentage changes as precise estimates of market demand. Show the counts beside the percentages. The question sample is chosen by you, not randomly drawn from all customer searches, so a confidence interval alone would not solve the sampling problem.

1. Assign a prompt ID. Keep the full text and intent category, such as discovery, comparison, integration, or buying constraints.
2. Capture conditions. Save date, language, product surface, visible model or mode, and whether a fresh conversation was used.
3. Save outputs. Keep the unedited answer, cited URLs, and a screenshot or export that a reviewer can inspect.
4. Apply fixed rules. Mark the brand match, owned-domain link, recommendation, and any ambiguity. Add a brief reason where judgment is involved.
5. Review disagreements. Have a second person inspect ambiguous rows when the report informs a significant decision. Record the final rule and apply it consistently.

Our monitoring walkthrough explains question selection and repeat collection. Start with the evidence table before investing in a large dashboard.

Freeze the original question cohort for the reporting period. When you add new questions, show their results separately until you have a comparable baseline. Otherwise a change in question mix can create an apparent gain or loss with no change in any original answer.

Use repeated checks to understand variation. Keep separate results by intent category and by product surface. A brand can appear often in branded support questions while being absent from unbranded buying questions. Combining both into one large percentage can hide the business problem.

When you update a page, record the URL, publication date, and substantive change. Allow time for discovery and collect comparable observations afterward. Describe an association between an edit and changed answers unless your study supports a stronger causal claim.

Read the pages that Perplexity cites. Its crawler documentation describes search discovery controls, but allowing a crawler does not guarantee a citation. Technical access, a relevant page, and a favorable answer are separate checks.

Perplexity’s current API migration guidance also matters when maintaining an integration. Keep endpoint and model changes in your log. Do not label a Perplexity API answer as a measured ChatGPT result.

Finally, join the visibility review to landing-page sessions and key events in GA4. A citation without a click remains visibility evidence; a tracked visit is traffic evidence. Neither alone proves a completed sale.