Key takeaways
- Separate factual errors from opinions.
- Correct public sources with evidence.
- Keep dated observations.
Record the exact error
Save the prompt, answer, date and cited sources. Classify the issue: wrong fact, outdated fact, ambiguous identity or opinion. Unfavourable opinion is not automatically a factual error.
Establish the approved fact
Find a current record your team can publish. This might be a dated service page or an approved leadership biography. Resolve internal disagreement before updating multiple public profiles.
Fix sources you control
Update the affected page, relevant structured data and managed profiles. Put corrected information in visible text. Do not publish contradictory facts or hidden statements intended only for models.
Request external corrections responsibly
Use a publisher’s normal correction process and supply evidence. Record the request. Do not create a network of false articles to overwhelm an accurate but inconvenient source.
Retest without a promised deadline
Repeat the original question later and retain both versions. A source update does not guarantee an immediate change in an assistant’s answer. Escalate material errors through the platform’s feedback route.
References and further reading
Primary documentation and product references. Practical checklists and worked examples above are editorial guidance, not claims about private ranking algorithms.