The POST /v2/answers/{answerId}/feedback endpoint records your user's rating of an answer: thumbs up or down, optionally with topics that say what went wrong, a comment, and a correction.

Project availability: ✅ Classic projects • ✅ Agentic projects

{answerId} is the answerId returned by the call that produced the answer — see Getting an answerId for where each endpoint returns it.

Basic Usage

Send the body as JSON with Content-Type: application/json. Only rating is required:

{
  "rating": "up"
}

{
  "feedbackId": "b6e2d0f4-91a3-4c57-8e2b-5d7f1a9c3e60",
  "answerId": "3f1c9a52-7d4e-4b8a-9c61-2e5f0b7a8d13",
  "rating": "up",
  "channel": "API",
  "createdAt": "2026-10-27T08:15:30.412Z",
  "updatedAt": "2026-10-27T08:15:30.412Z"
}

A negative rating usually says more:

{
  "rating": "down",
  "topics": ["INCORRECT_DATA_RETRIEVED", "INCOMPLETE_ANSWER"],
  "comment": "Cancelled orders are included in the total.",
  "correction": "Exclude orders with order_status = 'canceled'.",
  "metadata": {
    "endUserId": "u-48213",
    "screen": "sales-dashboard"
  }
}

Keep the feedbackId from the response if your user may change their rating later — see Updating feedback.

Request Parameters

Sending null for an optional field is the same as leaving it out. Lengths are counted in UTF-16 code units, so some emoji count as two characters.

Topics

Any of these values is accepted on any answer:

A value outside this list returns 422 INVALID_INPUT. Anything the list does not cover belongs in comment.

Response Fields

Updating feedback

Send the same request again with the feedbackId you received, and that feedback is overwritten — no second feedback is created:

{
  "feedbackId": "b6e2d0f4-91a3-4c57-8e2b-5d7f1a9c3e60",
  "rating": "down",
  "topics": ["INCORRECT_DATA_RETRIEVED"]
}

{
  "feedbackId": "b6e2d0f4-91a3-4c57-8e2b-5d7f1a9c3e60",
  "answerId": "3f1c9a52-7d4e-4b8a-9c61-2e5f0b7a8d13",
  "rating": "down",
  "channel": "API",
  "createdAt": "2026-10-27T08:15:30.412Z",
  "updatedAt": "2026-10-27T09:02:11.087Z"
}

• An update replaces the whole feedback. Fields you leave out are cleared, not kept — send comment, correction and metadata again if you want to keep them.
• Only a feedbackId this endpoint returned, for the same answer, can be updated. Anything else returns 404 FEEDBACK_NOT_FOUND.
• New feedback is not deduplicated. Sending the same request twice without a feedbackId creates two feedback records.

Sending answerId and feedbackId

Send answerId and feedbackId back exactly as Wren AI returned them. Both are compared as given, not case-folded: Wren AI returns them in lower case, and a value sent in a different case is not found (404).

Send the answerId, not the response id. Only an answerId identifies an answer.

Error handling

Error bodies carry code and error. Unlike most endpoints, they carry no id:

{
  "code": "INVALID_INPUT",
  "error": "rating is required and must be 'up' or 'down'"
}

Three responses differ: 405 returns only error, and 400 (a body that is not valid JSON) and 413 (a body over 1 MB) return a plain-text body instead of JSON.

Checks run in this order, so a request with several problems gets the error of the first one: body size, JSON parsing, method, API key, answer lookup, license, feedback storage, whether the answer can be rated, request fields, feedbackId lookup, saving.

Telling the two 401s apart

Telling the two 404s apart

Telling the three 422s apart

Retrying a 502

FEEDBACK_WRITE_FAILED is retryable, but the feedback may already be partly saved:

• An update can be retried safely with the same feedbackId; the retry overwrites the same feedback.
• New feedback gets a new feedbackId on retry, so the answer may end up with two copies of it.