AI describes your company incorrectly: what to check before responding publicly

Save the error before trying to explain it

Copy the question and answer, retain visible links and record the date, engine and language. If the error depends on an earlier conversation, record that context. An isolated screenshot may lose important detail when the team tries to reproduce what happened.

Then write the exact incorrect claim. Saying AI misunderstands us is insufficient. Did it invent a feature, use an old address or confuse two people? Precision allows a specific review instead of a general reputation discussion.

Find current information and what is still circulating

Review your own pages first. If the homepage, a PDF and a product listing give different accounts, start there. Decide which information is correct and who can validate it. Do not replace one contradiction with another without responsible-team review.

Then open the answer's sources. The link may not support the sentence or may concern an earlier period. Distinguish historically accurate publication from an article containing an error. The former may need current context; the latter may justify a correction request supported by evidence.

Correct the right place

A product-page error belongs on that page and in materials contradicting it. Publishing an article defending the brand does not replace the correction. If information is already clear, use available feedback or reporting mechanisms on the platform producing the answer, without assuming a guaranteed change.

For a hypothetical example, a tool stopped offering a feature. Update its current description and contextualize an earlier linked article. History need not be erased, but buyers should not mistake a past capability for part of a new purchase.

Check the correction without promising a deadline

Keep an error list with an owner, validated source and completed action. Repeat the discovery queries and record changes. A page may be corrected before a system's answers change; these are different facts to report separately.

This avoids turning every uncomfortable answer into a crisis. It also distinguishes serious problems from tonal preferences. Wiwo approaches AI Visibility as reading claims and sources; priorities depend on consequences for customers and others, not merely whether a result flatters the brand.

Corrections should serve people researching the company. They are not about hiding legitimate criticism or manufacturing consensus around a commercial description.

• Observed error: the exact sentence and its context.
• Validated information: the fact and supporting evidence.
• Action: which source changed or which review was requested.
• Verification: what later observations show.