Direct questions · No. 24 / 31
24. What should we do when AI gives incorrect pricing, integration or security information about us?
Last updated
Create an evidence-linked accuracy register and prioritise claims by buying consequence. Correct the underlying information you control, then pursue appropriate publisher or platform feedback routes without promising an immediate answer change.
Use this proposed register:
| Field | Required entry |
|---|---|
| Claim | Exact checkable statement, not a general complaint |
| Context | Prompt, surface, timestamp and complete answer |
| Evidence | Dated product documentation and responsible approver |
| Classification | Supported, contradicted, outdated or unverifiable |
| Recurrence | Number of comparable observations containing the claim |
| Priority | Commercial or compliance consequence |
| Action | Owned-page change, external correction request or feedback |
| Verification | Scheduled recheck and closure criteria |
Hypothetical example: an assistant says an integration is available on every plan, while approved documentation restricts it to a particular plan. Record the plan distinction and supporting evidence; do not merely label the answer “negative sentiment”.
Treat supplier-provided product facts as approved reference material, not independent certification. Escalate sensitive security, health, financial or legal claims to the relevant specialist.
Allow for source-refresh delays. Google’s recrawling guidance explicitly describes variable timing, so a page correction and a changed answer should remain separate milestones. (developers.google.com)
Sources cited on this page
- developers.google.com developers.google.com/search/docs/appearance/ai-features