How Citepoint measures AI visibility
What we ask, where the answers come from, how we read them and what the numbers mean. We also say where the method stops.
Citepoint asks ChatGPT, Gemini, Claude, Perplexity, Google AI Overviews and Google AI Mode the questions your buyers ask, in your country and language. Four engines are checked daily, and Claude and Perplexity weekly. A language model lists the products each answer recommends, and we match them against your brand and your competitors. The result is a trend, not a promise about what any one person sees.
Engines we track
| Engine | Where answers come from | How often |
|---|---|---|
| ChatGPT | The consumer app at chatgpt.com | Daily |
| Gemini | The Gemini app | Daily |
| Google AI Overviews | Live Google results | Daily |
| Google AI Mode | Live Google results | Daily |
| Claude | Anthropic's API with web search on, using a fixed model | Weekly |
| Perplexity | Perplexity's API, which always searches the web | Weekly |
ChatGPT and Gemini are asked in the consumer apps, where your buyers ask, in your country and language. If an app can't be read, we fall back to that engine's own API with live web search, and the answer says so.
Google AI Overviews and AI Mode come from live Google results for your country and language.
Claude is checked through Anthropic's API with web search on, using a fixed model.
Perplexity is checked through its own API, searching from your country. The Perplexity app searches the web for every answer too, so the two answer much alike.
We don't track Grok, Copilot or open-source models. Grok still shows up in your AI traffic stats, as a referrer and as a crawler.
Shared checks
When several customers track the same question in the same country and language, we fetch one answer per engine per day (per week for Claude and Perplexity) and reuse it. Everyone tracking that question sees the same answer the engine gave that day. Each customer's own brand is then read out of that shared answer.
How we read an answer
A language model reads each answer and lists every product it recommends or presents as an option, in order, with its domain. We match that list against your brand, its aliases and domain, and your competitors.
If an engine gave nothing, for example when Google showed no AI Overview, that answer doesn't count for or against you.
What the numbers mean
- Visibility (7-day). The share of answers in the last 7 days that recommend you, compared with the 7 days before.
- Share of voice. Your mentions as a share of all brand mentions across the answers.
- Per-question results. For each tracked question, how many of the last 28 days of answers recommended you, per engine.
- Cited as a source. How often engines link to your own domain.
Why we don't lead with rank
AI answers rarely list products in the same order twice. Average position mostly measures that noise, so we lead with how often you're recommended instead.
How long we keep answers
We keep the full text of each answer for 180 days. The verdicts we read from each answer are kept after the text is deleted.
Limits of the method
- AI answers vary. The same question can get a different answer on the next run, and for a different person.
- We don't model logged-in personalization, memory or follow-up questions. A buyer deep in a conversation may see something else.
- One sample per engine per day is a trend, not a guarantee of what any single person sees.
Read the numbers as a direction over weeks, not a verdict on a single day.
Questions about the method? Email hello@citepoint.ai. For changes to how we track, see the changelog.