---
title: "AI Answers API"
description: "What ChatGPT, Gemini, Google AI Overviews, Google AI Mode and Naver's AI answer to a prompt, with every citation, source, product card and brand mention, in one answer type."
---

> Every apinoa docs page as markdown: https://apinoa.com/llms.txt (all in one file: https://apinoa.com/llms-full.txt; this API family in one file: https://apinoa.com/docs/answers/llms.txt)

The AI Answers API asks an AI engine a question and returns the answer people are shown: the text, the
sources it cites, the product and place cards, and the brands the engine itself marked. Every engine
returns the same type, so tracking what ChatGPT says about a brand works the same way as tracking Google AI Mode.

```bash
curl -X POST "https://gateway.apinoa.com/v1/chatgpt/answer" \
  -H "x-api-key: $APINOA_API_KEY" -H "content-type: application/json" \
  -d '{"prompt": "What are the best running shoes for beginners?", "country": "US"}'
```

## Engines

Every engine has one operation, `answer`.

| Engine | Countries | Conversation | Photos | Usually under |
|---|---|:-:|:-:|---|
| [ChatGPT](/docs/api-reference/chatgpt) | US DE GB AT KR AU JP CA FR IT | ✓ | ✓ | 30 s |
| [Gemini](/docs/api-reference/gemini) | US DE GB AT KR AU JP CA FR IT | ✓ | | 15 s |
| [Google AI Mode](/docs/api-reference/google-ai-mode) | US | ✓ | | 35 s |
| [Google AI Overviews](/docs/api-reference/google-aio) | US | | | 10 s |
| [Naver AI Tab](/docs/api-reference/naver-ai-tab) | KR | ✓ | | 40 s |
| [Naver AI Briefing](/docs/api-reference/naver-ai-brief) | KR | | | 10 s |

An answer takes seconds, not milliseconds. Allow up to 120 seconds per call.

## Inputs

| Input | Meaning |
|---|---|
| `prompt` | the question, as a person would type it (required) |
| `country` | the market it is asked from. Each engine lists its countries above, and the first is the default |
| `language` | the language to answer in (Google engines). The other engines answer in the prompt's language |
| `conversation` | an earlier answer's `conversation.id`, to ask a follow-up in the same conversation |
| `photos` | up to four photos to ask about, as https URLs or `data:image/…;base64,` URLs (ChatGPT) |
| `include` | `page`: also return the answer as a self-contained HTML page |

Send the inputs as a JSON body with `POST`. An input an engine does not take is refused with
`400 INVALID_PARAMS`, and the message gives the reason. It is never silently ignored.

## Outcomes

`outcome` says what the engine did:

- `answered`: it answered.
- `noAnswer`: it showed no answer for this prompt. Google shows an AI Overview for only some searches, and
  Naver shows a briefing for only some. The surrounding results are still returned in `results`, and the
  call is billed, because the page was read.
- `refused`: it declined; its words are in `text`.
- `partial`: it stopped early, with some text.

## What Is the Same Everywhere

- **One type.** Every engine returns an [`Answer`](/docs/api-reference/answers/response-schema#answer).
  Facts only one engine has are in `ext`.
- **Text, markdown and blocks agree.** `text` and `markdown` are rendered from `blocks`, and `citations`
  point at character ranges of `text`.
- **Sources, cited or not.** `sources` lists everything the engine showed. `cited` tells the sources the
  answer links to from the ones it only read.
- **Mentions are the engine's own.** `mentions` are the brands, products and places the engine itself marked.
  They are never guessed from the text.
- **Price.** $0.0012 per answer.

## Errors

The AI Answers API uses the same envelope and codes as the rest of the API. See [Errors](/docs/errors). An
engine that does not answer in time returns `502 UPSTREAM_TIMEOUT`. Failed calls are not billed.

## Reference

- [Response Schema](/docs/api-reference/answers/response-schema): every field of the answer
- [SERP API](/docs/api-reference/serp): the search engines, whose Google results carry the same AI Overview
- [Errors](/docs/errors): every error code