Accuracy & Errorsยท3 min readยทUpdated 1 April 2026

How to Flag an Incorrect Answer

How to report an AI answer you believe is wrong, what happens after you flag it, and how your flags improve AskBiz for everyone.

Why Flagging Matters

Every flag you submit makes AskBiz more accurate โ€” not just for you, but for all users. Your flags are the primary signal we use to identify systematic errors in the AI's reasoning. An AI that never receives feedback cannot improve. We take every flag seriously, review them individually, and act on patterns.

How to Flag an Answer

In the chat interface:

1. Find the response you believe is incorrect

2. Click the thumbs-down icon (๐Ÿ‘Ž) below the response

3. A feedback panel opens โ€” select the error type:

  • Wrong number โ€” the calculation or figure is incorrect
  • Wrong interpretation โ€” the data is correct but the conclusion is wrong
  • Outdated information โ€” the answer references something that has changed
  • Hallucination โ€” the AI stated something as fact that isn't true
  • Missing context โ€” the answer is incomplete in a way that could mislead
  • Other โ€” anything else

4. Add a note explaining what the correct answer should be (optional but very helpful)

5. Submit โ€” your flag is logged immediately

Via email:

For complex issues or answers that caused a significant business decision error, email support@askbiz.co with:

  • The exact question you asked
  • The answer you received
  • What the correct answer should be and why
  • Any supporting data

What Happens After You Flag

Within 24 hours: Your flag is reviewed by a member of our team. We check the answer against your data to confirm whether it was genuinely incorrect.

Within 5 business days: If the error is confirmed, we:

  • Update our system prompting to prevent the same error type
  • Add the example to our benchmark test suite so future model updates are tested against it
  • Notify you that the issue has been addressed (if you provided contact details)

Monthly: Error patterns from flags are analysed. High-frequency error types are escalated to model-level improvements and, where relevant, reported to Anthropic.

You will not be penalised, deprioritised, or treated differently for submitting flags โ€” we actively encourage them.