An AI assistant gives a customer your old address, describes a service you stopped offering or invents a price. Where do you start?
To correct inaccurate business information in AI answers, identify the specific claim and its likely sources, fix the information you control, request corrections from relevant third parties and test the same questions again over time. You cannot directly edit every AI model or guarantee when an answer will change.
This is a practical part of generative engine optimisation (GEO): helping your business be understood accurately, not just mentioned more often.
First, separate a factual error from an unwanted recommendation
“The company is based in Bristol” is a factual claim you can check. “This is the best agency for your business” is a judgement that depends on the question and evidence.
Prioritise errors that could lose a customer or create a complaint:
- An incorrect telephone number, website or trading address.
- Services, delivery areas or opening hours that have changed.
- Prices presented as current when they are outdated.
- Confusion between your company and another business with a similar name.
- Qualifications, guarantees or partnerships you do not hold.
Save the exact question, answer, date, platform and any visible citations. Note whether web search was enabled and whether the answer came from a fresh conversation. A screenshot alone does not show why the error occurred.
1. Find where the incorrect claim appears
Open the sources linked in the answer, if there are any. Then search for your business name alongside the incorrect detail. Check old service pages, PDFs, directory listings, press coverage, partner profiles and duplicate business listings.
For example, an installer might have removed “emergency repairs” from its homepage but left the service in a downloadable brochure and several directories. Changing the homepage again will not resolve those other contradictions.
An answer without citations needs more caution. Do not assume a particular page caused it, or that the assistant retrieved a current web page at all.
2. Make your own website unambiguous
Update the relevant service, contact, pricing and location pages. Put important facts in readable page text rather than only inside an image or a document.
Use a consistent business name and explain any relationship between your trading name and legal entity. If you serve customers remotely or visit their premises, say so clearly. Do not describe a correspondence address as a customer-facing office.
For a discontinued service, an explanatory update can be more useful than simply deleting every mention: “We no longer provide residential installations; our current service is commercial maintenance.” Keep the statement truthful and relevant to visitors.
Structured data should agree with the visible information. It is supporting description, not a private instruction that overrides what an AI assistant says. See our AI SEO audit guide for the broader checks.
3. Correct the important external sources
Update profiles you own and contact the publisher of an incorrect third-party page. Provide the exact correction and an authoritative page that supports it.
A useful message is specific: “Your listing shows our previous telephone number. Please replace it with the number on our current contact page.” Avoid asking a publisher to remove legitimate criticism just because it appears in AI answers.
Start with sources customers actually use or that recur in the answers you recorded. Buying dozens of new directory listings is not a substitute for fixing one influential inaccurate profile.
Keep a correction log: source URL, incorrect claim, requested change, date contacted and whether the correction is live.
4. Check access and freshness without promising instant changes
Make sure the corrected page loads, returns a successful response and is available to the search services you want to reach. Check that canonical tags point to the intended page and that a security challenge is not replacing the content.
Search indexing, retrieval and model knowledge are different processes. A page being accessible does not mean every system has indexed it, and a new search result does not mean an older model response will immediately change.
Where a platform provides an answer-feedback option, report the factual error and include supporting evidence. Feedback is not a guaranteed correction service. For serious impersonation, defamatory statements or sensitive personal information, use the platform’s relevant reporting process and obtain appropriate professional advice.
5. Retest the claim, not just your favourite prompt
Use a small set of realistic questions: “Where is this company based?”, “Does it offer emergency repairs?” and “Which areas does it cover?” Repeat them in fresh conversations on the platforms your customers use.
Record correct, incorrect and uncertain answers separately. One corrected answer is encouraging, but it is not proof that the issue has disappeared everywhere. Recheck periodically and after material business changes.
Our guide to brand mentions in ChatGPT explains the difference between being named and being represented accurately.
Questions business owners ask
Can I pay to have an AI answer corrected?
Be cautious of anyone guaranteeing that they can edit an independent AI system’s answers. An agency can investigate sources and implement legitimate corrections; it cannot promise control over every generated response.
Should I publish a page telling AI to ignore the old information?
Publish a clear factual update for customers instead. Hidden instructions and commands aimed at models are not a dependable correction method.
How will I know the work has helped?
Look for sustained improvement across repeated checks, fewer customer misunderstandings and corrected source pages. Keep those outcomes separate: fixing the source is observable even when answer behaviour remains variable.
Need help identifying the conflicting sources? Book an AI visibility consultation with Panovista to discuss a focused factual-accuracy review.