An AI assistant says something wrong about you.
A former role given as current, a wrong date, a company you left. Before deciding what to do, establish what exactly is wrong, in which assistant, and where the statement comes from.
An inaccurate AI answer is a statement about a person or an organization, produced by an assistant such as ChatGPT, Gemini, Perplexity or Claude, that does not match the facts: a former role presented as current, a wrong date, an outdated description. It is addressed mainly at the sources the answer draws on, since no one can edit the answer itself, and no one can guarantee what the assistant will say afterwards.
How to recognize it
- A former role, employer or title Given as the present one.
- An outdated description A positioning, a product or an ownership that has changed.
- A wrong fact A date, a place, a figure, a qualification.
- An event attached to the wrong period Something past presented as ongoing.
Two neighboring cases are not this one. If the facts belong to another person, see confused with someone else. If the statement is accurate and unwelcome, it is not an error.
Why it can appear
An assistant composes its answer from what the model learned and, when it searches, from the sources it retrieves; the mechanisms are set out in how AI systems form a description. Three situations can lead to an inaccurate statement.
- A source is itself wrong or out of date A profile never updated, an old article, a directory entry.
- Sources disagree The official page says one thing and third-party pages another.
- The model's own knowledge can be older than the facts An answer given without web search rests on what the model learned earlier, which a later change at a source does not alter.
How to verify it
- Write the question down and ask it several times, in each assistant separately. A single answer is a snapshot.
- Keep the full answer and the sources it cites, with the date and whether web search was active.
- Separate what is wrong from what is merely thin. List each incorrect statement on its own line.
- Trace each one. Is the wrong fact on a page the assistant cited? On one of your own profiles? Nowhere you can find?
What to record for each run is listed in what is recorded.
What can be corrected
| Where the wrong fact lives | What can be done |
|---|---|
| Your own site and profiles | Correct it, and state the current fact the same way everywhere |
| A publication, a directory, a data provider | Ask for a correction |
| Your Google knowledge panel | Claim it, as the entity or its official representative, and suggest a change (Google, knowledge panel help) |
| A Wikipedia article | Propose the change on the talk page, with disclosure (Wikipedia, conflict of interest) |
| Personal data held by an organization | A legal route may exist, depending on the jurisdiction; a question for qualified counsel |
| An AI provider | Each provider has its own channel; OpenAI's is described in how inaccurate information is corrected |
The same routes are set out in order for a person in how inaccurate information is corrected and for a company in how a company corrects its record.
What cannot be guaranteed
No one can edit a generated answer. A correction at a source changes what an assistant may retrieve. An answer given without web search can keep repeating the old fact, and no one outside the provider knows when, or whether, that changes. And a source that has been asked is not obliged to correct.
Evidence and sources needed
- One reference version Of the fact in question, on a page you control.
- A document that proves it An appointment, a filing, a dated announcement.
- At least one independent source That states the current fact.
- Consistency The same wording on the profiles and pages that describe you. See entity optimization.
How to measure improvement
Ask the same questions again, under the same conditions, at a fixed interval, and compare: is the wrong statement still present, in which assistants, and from which sources? The metric is accuracy, recorded per engine. A change in one assistant says nothing about the others.
When to ask for an assessment
When the wrong statement appears in more than one assistant, comes back after a correction, or cannot be traced to a source. A private assessment records what each assistant says today, with the sources it cites, and sets out what to address first. Corrections of this kind are part of the engagements for leaders and organizations, which follow The Authority Architecture System.
Understand how search and AI currently interpret your authority.
Request a confidential assessment of your visibility, entity signals, narrative consistency and source authority.
Request a Private AssessmentSelective engagements for leaders and organizations with material reputational stakes.