Brand visibility in Perplexity web-grounded answers
Perplexity answers are web-grounded by design and return their search results with the answer, which makes them one of the clearest reads available on which sources are shaping a recommendation.
How this measurement is collected
Tracked using the Perplexity Sonar API, which is web-grounded by design and returns its search results alongside the answer.
- Surface we call
- Perplexity Sonar API
- Collection mode
- Official grounded API
- Collected from the provider’s own API with web grounding enabled. Citations are provider-attested.
- Label used in-product
- Perplexity web-grounded API
- The same label rides along with every metric collected this way, so a number can always be traced back to how it was obtained.
- One check is a single prompt × engine × market execution, sent to the Perplexity Sonar API.
- The model is a deployment setting rather than a fixed claim, and the exact model string is stored on every response we keep.
- The search results returned beside the answer are stored as the citation set for that check.
- The raw provider payload is retained with a SHA-256 digest, so a stored answer can be shown to be unmodified.
What AnswerRank reports for Perplexity
Every answer is normalised into the same structure across engines, so one prompt set can be compared across surfaces without blending collection methods.
- Whether the answer recommended, listed, merely mentioned, or omitted your brand.
- The cited URLs behind the answer, canonicalised and de-duplicated before analysis.
- Which sources repeatedly appear across your category — the basis of Observed Source Influence.
- Position where the answer contains an objectively ordered list of options.
What this measurement does not support
The boundary is the point. A number is only defensible when the claim around it is narrower than the data behind it.
Reflects the API surface. Perplexity’s consumer product applies additional personalisation and model routing.
- Answers are non-deterministic. A single execution is a sample, not a ranking fact, so every figure is reported with its sample size and a confidence level.
- A source cited often in our scans is influential in the sample we observed. That is not evidence about training data, which no third party can inspect.
Only two execution outcomes enter a metric: a successful answer, and an answer where the engine responded and your brand genuinely was not there. Provider errors, timeouts, rate limits and unconfigured engines are excluded from both the numerator and the denominator, so a provider outage can never be displayed as a collapse in your visibility.
How a check becomes a metricThe other engines we track
Each engine is collected differently and labelled differently. Comparing them is useful; blending them without labels is not.
- OpenAIOpenAI web-grounded APITracked using OpenAI’s Responses API with the hosted web search tool enabled. Citations are the sources the API itself attributes.
- ClaudeClaude web-search APITracked using the Anthropic Messages API with the server-side web search tool. Citations come from the tool’s own search results.
- GeminiGemini + Google Search groundingTracked using the Gemini API with Google Search grounding. We store the grounding metadata, including the search queries Gemini issued.
- Google AI OverviewsObserved Google AI Overview SERP dataObserved through a licensed structured SERP provider that captures the AI Overview element for a given query, location and device.
See how Perplexity answers your category.
Run a visibility check on your site and read the result with its sample size, its collection method and its limits stated on the page.