Brand visibility in OpenAI web-grounded answers
AnswerRank puts your buyers’ questions through OpenAI’s web-grounded API and records what came back — whether you were recommended, who was recommended instead, and which sources the answer was built on.
How this measurement is collected
Tracked using OpenAI’s Responses API with the hosted web search tool enabled. Citations are the sources the API itself attributes.
- Surface we call
- OpenAI Responses API + web search tool
- Collection mode
- Official grounded API
- Collected from the provider’s own API with web grounding enabled. Citations are provider-attested.
- Label used in-product
- OpenAI 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 Responses API with the hosted web search tool enabled.
- The model is a deployment setting rather than a fixed claim, and the exact model string is stored on every response we keep.
- Citations are the sources the API itself attributes to the answer. We do not infer sources by pattern-matching the answer text.
- 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 OpenAI
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.
- Which competitors appeared in the same answer, and how they were characterised.
- The cited domains and pages behind the answer, classified as owned, competitor-owned, review platform, community, publication or third party.
- Position — but only where the answer contained an objectively ordered list. Prose answers carry no rank, so none is invented.
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 web-grounded API surface. It is not a capture of a signed-in ChatGPT session, which is personalised by account history and settings we cannot observe.
- 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.
- Retrieval and index freshness belong to the provider. A source visible in one week’s sample may be absent from the next, through no change on your site.
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.
- 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.
- PerplexityPerplexity web-grounded APITracked using the Perplexity Sonar API, which is web-grounded by design and returns its search results alongside the answer.
- 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 OpenAI 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.