Reputation guide
What to do when AI gives wrong information about your brand
If an AI assistant gets facts about your brand wrong (old prices, features that don’t exist, confusion with another company), complaining to the platform is not enough. What works is documenting the error, finding the source it comes from, fixing it on your site and third-party sites, strengthening your official facts, and measuring again. GeoRush catches these errors by checking every answer against your official facts.
Updated on

Why AI gets your brand wrong
AI assistants make mistakes, and their makers say so: Google, for example, warns in its help pages that AI Overviews can make mistakes and recommends double-checking important information. When the mistake is about your brand, the damage is concrete: a customer who thinks you lack a feature, cost twice as much, or shut down.
Errors usually come from one of these causes:
Outdated training data. The model learned your pricing or offering from a year or two ago and answers from memory, without searching.
Wrong third-party sources. A comparison, directory or forum has old or wrong data, and the assistant cites it.
Your own site is confusing or contradictory. Two pages with different prices, an old page still indexed, an outdated PDF.
Entity confusion. Another company with a similar name, a former brand, a subsidiary.
Fabrication. The model fills gaps with something plausible but false, especially when there is little public information about you.
Error types and how to tackle them
| Error | Likely cause | What to do |
|---|---|---|
| Old price or plan | Old indexed page, outdated comparison, or model memory. | One current, dated pricing page; redirect or remove old ones; ask comparisons that cite you to update. |
| A feature you don’t have (or says you lack one you do) | Ambiguous documentation or wrong third-party sources. | A clear features page stating what you do and don’t do; fix the cited sources. |
| Confusion with another company | Similar name, weak entity data. | Organization markup with sameAs, an unambiguous description, verified profiles, the same fact sheet everywhere. |
| Company “closed” or at another address | Old directories or profiles. | Update your Google Business Profile and the main directories. |
| Old review or controversy presented as current | Few recent sources to balance it. | Respond with facts and dates; publish current information AI can find. |
| Invented fact with no source | Little public information about you. | Publish the missing facts and earn presence in sources AI reads. |

A 7-step correction protocol
- 1
Document the error
Save the exact question, the engine, the date, the answer text and the cited sources. Repeat the question several times: an error that shows up once in ten tries is not as urgent as one that shows up every time.
- 2
Find the origin
If the answer cites sources, open them: is the wrong fact there? If it cites nothing, search for the fact yourself on Google and the main sites in your field. Check your own site too, including old pages and PDFs.
- 3
Fix your own site first
One official version of each fact, in visible text, with a date. Remove or redirect obsolete pages. Make structured data match the text.
- 4
Fix third-party sources
Write to the author or manager of the comparison, directory or profile with the correct fact and evidence (a link to your official page). Keep it short and specific.
- 5
Strengthen your entity
Identical facts on your site, social profiles, directories and Google Business Profile. If you have a Google knowledge panel, claim it and suggest changes from there.
- 6
Notify the platform
Use each assistant’s rating and reporting channels (below). It helps, but it doesn’t replace fixing the sources.
- 7
Measure again
Repeat the same questions over the following weeks. On engines with search the fix can show sooner; in training-based answers it can take months.
Correction channels by platform
Official channels documented as of November 2026. Menus change often: check each platform’s help pages.
| Platform | Channel | What to expect |
|---|---|---|
| ChatGPT (OpenAI) | Thumbs down under the response; “Report” for problematic content; the model behavior feedback form. | OpenAI says reported content may be reviewed; it does not promise a specific change or timeline. |
| Google AI Overviews | Rate the overview as bad and “Report a problem”, or use the overview’s three-dot menu. | Google collects the feedback; it doesn’t publish correction timelines. |
| Google knowledge panel | Claim the panel by verifying an official profile, then suggest changes with links that confirm the fact. | Feedback from verified entities gets priority, according to Google. |
| Google Business Profile | Edit the profile (hours, address, category, website, description). | Fixes the source Google uses for local business data. |
| Other assistants | Most include controls for rating answers in their interface. | Useful as a signal; the lasting fix lives in the sources. |
Template: official fact sheet
Keep an internal sheet of your official facts and use it to review your site, profiles, JSON-LD and every correction you request. Publish on your site, as visible text, the facts you want AI to repeat correctly:
# Official facts for [brand] (updated: [date])
Name: [exact name, no variants]
What it is: [one sentence: category + who it is for]
Official website: [URL]
Official profiles: [LinkedIn, YouTube, etc.]
Countries we operate in: [list]
Pricing: [from X / plan, date] -> [pricing URL]
Main features: [list]
What we do NOT do: [list, avoids confusion]
Not to be confused with: [similarly named companies]
Public contact: [contact URL]If the error is about a person
When the false information concerns an identifiable person (for example, the founder or a spokesperson), data protection law comes into play. In the European Union, the General Data Protection Regulation grants a right to rectification of inaccurate personal data (Article 16). Check each provider’s privacy policy for how to exercise it and, if the case is serious, get legal advice. That right protects people, not companies: for brand facts, follow the protocol above.
How GeoRush helps
In brand perception, Claude checks the claims each engine makes about you against your official facts and flags the ones that don’t match, alongside tone per engine and the themes that get praised or criticized. Your official facts live in brand identity. Every answer is stored word for word with its sources, so you can document the error and see where it comes from. And alerts notify you when negative mentions increase.
Common mistakes
- Reacting to a single answer without checking whether the error repeats.
- Reporting to the platform without fixing the original source.
- Leaving old pages with outdated prices or terms accessible and indexable.
- Having different facts on your site, in your JSON-LD and on your profiles.
- Not measuring again after the fix.
What does AI say about your brand?
Run the free check and see how AI describes you when your customers ask about your category.
Sources
Official documentation and primary sources checked on November 17, 2026. These policies change often: check the source before making important decisions.
- Google Search Help: AI Overviews (how to send feedback)
- OpenAI Help Center: reporting content in ChatGPT
- OpenAI: model behavior feedback form
- Google Knowledge Panel Help: get verified and suggest changes
- Google Business Profile Help: edit your Business Profile
- Google Search Central: spam policies for Google web search
- Google Search Central: Organization structured data
- EUR-Lex: General Data Protection Regulation (EU) 2016/679, art. 16 (rectification)
Frequently asked questions
Keep reading
Brand perception
Tone per engine, praised and criticized themes, and AI claims checked against your official facts.
Fix what AI says
Find false AI claims about your brand and fix them at the source.
Schema for AI
Which structured data actually helps, what Google says, and copy-ready JSON-LD examples.
Digital PR for AI
How to earn mentions in the sources AI cites, without practices Google treats as spam.