# OpenAI API + web search

OpenAI's Responses API with its native web_search tool, capped at two searches per answer. We record the search queries the model ran.

## Why measure it

Reproducible and controllable: pick the locale, turn search off for a parametric probe, and compare against the consumer app.

## Ask it

```bash
llmentions ask "best crm for startups" --brand Acme --engines chatgpt -n 5
```

Over MCP, pass `engines: ["chatgpt"]` to `ai_answers` or `visibility_run`. Over REST, `"engines": ["chatgpt"]` on `POST /v1/ai/answers`.

Every answer is labelled `source: "api"`. Set `web_search: false` for a parametric probe (`source: "parametric"`): what the model knows without searching, at $0.005 per answer.

## Price

$0.05 per answer. Failed answers are refunded, and test keys are free. See [Pricing](/pricing/).

Read how we sample, label and compare engines in [Methodology](/docs/methodology/).

---
https://llmentions.com/engines/chatgpt/
