Provider model APIs vs AnswerLine
Calling OpenAI's API and asking the ChatGPT app are different products. One returns a completion from weights; the other searches the web, cites pages and shows product cards first.
What each returns
| Provider model API | AnswerLine | |
|---|---|---|
| Answer text | A completion — the model's output for your prompt and parameters. | The answer a logged-in user actually received. |
| Sources | Only if you build the retrieval yourself. | `sources[]` — the real citations with display position. |
| Search fan-out | Not exposed. | `searchQueries[]` — the live searches the engine ran. |
| Commerce blocks | None. | Shopping cards, inline products, places — the units users see. |
| Geo variation | None — one model, one answer. | Per-country and per-US-state answers, as users there get them. |
When each is right
- **Provider APIs** when you're generating — building a feature on the model itself, where the answer's shape doesn't matter.
- **AnswerLine** when you're measuring — visibility, citations, recommendations, competitive presence. The question is what users *see*, and only the consumer surfaces answer that.
The two compose: many teams use a provider API to build the product and this API to measure how the product performs inside the consumer answers.