# How GetIntel collects data

> How GetIntel asks ChatGPT, Perplexity, Gemini and Claude your buyer questions, how answers are classified, and why a score can differ from what you see.

Source: https://getintel.ai/docs/academy/start-here/how-getintel-collects-data/

GetIntel asks each AI engine the buyer questions you track, keeps the full answer, and classifies how your brand appears in it. Wherever possible it reads the real product your buyers use, not a developer API, because API answers are retrieved differently and miss what a buyer actually sees.

## How each engine is read

| Engine | How GetIntel reads it | How often |
| --- | --- | --- |
| **ChatGPT** | The real chatgpt.com interface | Daily |
| **Gemini** | The real gemini.google.com interface | Daily |
| **[Perplexity](/perplexity-visibility-tracker/)** | The real perplexity.com interface on the daily scan. Your very first scan uses Perplexity's API so results arrive in minutes. | Daily |
| **Google AI Overviews** | Real Google results pages, reading the AI Overview when Google shows one | Daily |
| **Claude** | Anthropic's API with web search switched on | Weekly, on Growth |

## Asked from your market

Questions are asked from your [market](/docs/features/markets/): a country and a language, picked during setup. AI answers vary by country, so a brand selling in the UK should be measured from the UK. Pro and Growth can track several markets, each with its own questions and results.

## What happens to each answer

1. **The full answer is stored**, with the sources it cited.
2. **Every brand named is found**, in the order it appears.
3. **The answer gets a state for your brand**: recommended, mentioned, absent, open or no answer. See [How measurement works](/docs/features/how-measurement-works/#answer-states).
4. **Scores are rebuilt** for your topics, competitors and pages, then your task list is refreshed.

## When an engine returns nothing

Sometimes an engine fails to respond, or Google shows no AI Overview for a search. GetIntel records that as **No answer** and leaves it out of your scores. A missing answer is never counted as the engine ignoring you.

## AI crawler data

AI crawler numbers come from your own [Cloudflare](/docs/integrations/cloudflare/) analytics. Each bot visit is checked against the IP ranges the AI company publishes, so scrapers pretending to be GPTBot are not counted.

## Why an answer can differ from what you see

If you ask ChatGPT the same question yourself, you may get a different answer. That is expected:

- **Your account shapes answers.** Chat history, memory and personalization change what you are told. GetIntel asks every question in a fresh, clean session, so no history or personalization shapes its answers.
- **Location matters.** Answers differ by country.
- **Answers vary run to run.** The same question can name different brands on different days, which is why GetIntel scores over many answers and a 7-day window instead of a single response.
- **Models change.** Engines update their models and search behaviour without notice.

<Aside type="tip">
To compare fairly, open the answer GetIntel stored for that day in **Prompts** rather than asking the engine again later.
</Aside>