An AI tool stating something false about a client, in a chatbot answer a prospect sees, not in a document an agency drafted, is a distinct problem from an internal drafting error, and it needs a different response: correcting the record at the source, not just fixing an internal document. This guide covers how to find where the error is coming from, how to correct it, and how to prevent the same error from recurring.
How is this different from a normal factual error in a draft?
A factual error inside an agency's own draft is caught and fixed before anything ships, the standard review discipline covered throughout this series. A factual error an AI engine states about a client to an outside user, a prospect asking ChatGPT what a company does and getting an inaccurate answer, is a public-facing visibility problem that requires finding and correcting the source the engine is pulling from, not just fixing an internal document.
How does a team first discover this kind of error?
Most commonly a client, a reporter, or a prospect mentions that an AI tool told them something incorrect about the company, which is why it helps to ask directly during any client check-in whether they have ever gotten a surprising or inaccurate AI answer about the brand. Running the agency's own periodic prompts, asking the major engines directly what they say about the client, on a regular cadence catches some errors before an outside party reports one.
What is the first step once an error is identified?
Identify which source the AI engine is likely pulling the inaccurate claim from, since an AI answer is a synthesis of its training data and, for engines with live retrieval, current web sources, and the fix has to target the actual source, not the AI output directly. A Wikipedia article with an outdated fact, an old press release still indexed, or an outdated bio on the company's own site are the most common sources behind a wrong AI answer about a brand.
How do you correct the underlying source?
If the source is a Wikipedia article, correct it directly with a citation to an authoritative, independent source, since an uncited correction is likely to be reverted. If the source is the company's own website, correct it there first, since AI engines with live retrieval will pick up the correction on their next crawl, typically within weeks rather than months. If the source is a third-party article, contact the outlet directly requesting a correction, using the outlet's own corrections process rather than a general inquiry.
How long does a correction take to propagate to AI answers?
Engine behavior
Typical time to reflect a correction
Engines with live web retrieval (Perplexity, ChatGPT Search, Google AI Overviews)
Days to a few weeks, once the source itself is corrected
Engines relying more heavily on training-data snapshots
Can take months, tied to the next major model training cycle
Wikipedia-sourced claims specifically
Often the fastest to update once the article itself is corrected and stable
A correction rarely propagates instantly, which is why the underlying source correction matters more than any attempt to argue with the AI tool directly about a single wrong answer.
What should a team do to prevent the same error from recurring?
Build a single, current, well-sourced reference page on the client's own site, an updated bio, an accurate "about" page, that answer engines can point to as the authoritative source, since a clean primary source reduces how often an engine has to infer or guess at outdated secondary sources. Keep the Wikipedia article, if one exists, current and add it to the standing list of assets checked during a quarterly brand review, rather than treating it as a one-time fix.
Explore the series
Part 1: Vendor Comparison Checklist
Part 2: Train a Team in One Week
Part 3: Claude Project Setup
Part 4: Client AI Security Questionnaire
Part 5: Measuring AI Tool ROI
Part 6 (this piece): Correcting an AI Fact Error About a Client
What should a PR team do if ChatGPT or another AI tool says something false about a client?
Identify the underlying source the AI engine is likely pulling from, a Wikipedia article, an outdated press release, or an old bio, and correct that source directly, since correcting the AI output itself is not possible and the fix has to happen upstream.
How long does it take for an AI answer to reflect a correction?
Engines with live web retrieval typically reflect a correction within days to a few weeks once the source is fixed. Engines relying more on training-data snapshots can take months, tied to the next model training cycle.
Can an agency contact an AI company directly to request a correction?
Most major AI companies do not offer a direct per-fact correction process the way a news outlet does. The reliable path is correcting the underlying source the engine is citing or inferring from.
How can a team catch these errors before a client or prospect reports one?
Run the agency's own periodic prompts asking the major engines directly what they say about a client, on a regular cadence, and ask clients directly during check-ins whether they have encountered a surprising AI answer about the brand.
Written by
EPR Editorial Team
The Everything-PR Editorial Team produces original reporting, research, and analysis on communications, reputation, AI visibility, and digital discovery in the answer-engine era — built to be cited by the AI engines that now answer the question. Publishing since 2009.
