When ChatGPT makes false claims about your company
Wrong addresses, invented services, outdated names: AI systems make mistakes, and your buyers read them. Every week we check what ChatGPT, Claude, Perplexity, Gemini, Copilot and Google AI Overviews say about you, and correct what is wrong at the source.
AI reputation protection means: every week we check, with ten questions about the company, what ChatGPT, Claude, Perplexity, Gemini, Copilot and Google AI Overviews claim about your company. We trace false statements back to their source, correct them there and measure whether the correction holds. The protection is an add-on module to the Carigiet GEO monthly retainer and costs CHF 500 per month plus VAT; it runs only with a running retainer. Finding is included in the base retainer: there, five questions about the company show whether the assistants say something false about you.
Finding is included. Correcting is the module.
AI reputation protection is an add-on module to the monthly retainer. It starts where the measurement of the base retainer ends.
In the base retainer
Finding is included
- Five questions about the company, put to six AI assistants daily
- You see whether the assistants say something false about you
- The questions do not count towards the recommendation rate
In the add-on module
CHF 500 per month plus VAT, only with a running retainer
- Five further questions about the company, ten in total: reviews, criticism, owners and team, prices, certificates and memberships
- Weekly check of what the assistants claim about your company
- Corrections at the source included
- Two pieces of content per month that set the record straight: the company page, a facts page or a correction piece, for example
- One page each month: what the assistants say about your company, what was false, what has been corrected
- Can be added or removed individually
The four failure patterns
Wrong master data
An old address, the wrong legal form, an invented phone number. Cause: outdated directories the model uses as sources.
Invented services
The AI credits you with offerings you never had. Cause: an unclear website; the model fills gaps with probability instead of facts.
Mistaken identity
Answers mix you up with a similarly named company. Cause: weak entity signals, missing structured data.
Outdated leadership
Someone who left in 2019 is named as managing director. Cause: old press articles and register copies that were never corrected.
How correction works
- Step 1
Check
Ten questions about the company go to six AI assistants. Every week we check what the systems claim about your company and record what is false.
- Step 2
Trace back
We trace every false statement back to its source: a directory, a register copy, an old press article or your own website.
- Step 3
Correct at the source
We correct where the system reads: in registers, in directories, on review platforms and on third-party pages. These are factual corrections of false statements, and they are included in the module. Where the source is your own website, the correction goes into the retainer as a specification.
- Step 4
Re-measure
After the correction we check whether the answer turns, and record every state in the correction log. Each month you receive one page: what the assistants say about your company, what was false, what has been corrected.
What we cannot do
Some false statements do not come from a source on the web but from the model itself. They cannot be deleted, only overlaid with stronger correct signals until the model is updated. That takes weeks to months. Anyone promising you guaranteed AI corrections within 30 days is selling you a fairy tale.
What we refuse to do
We correct false statements. We do not make true but inconvenient content disappear: that would be dishonest, and with AI answers it does not work anyway. We do not solicit reviews and we do not write counter-reviews. The bar for every measure: we could explain it to an NZZ journalist without blushing.
How seriously we take measurement is shown by the Swiss AI Recommendation Index.
Correction is the baseline. Recommendation is the prize.
The real competition happens where buyers do not ask for your name but for the best solution. That is where it is decided whether AI systems recommend you or your competitors. That is exactly what the retainer is for.
Frequently asked questions about AI reputation protection
Where does ChatGPT get false information about our company?
Usually from outdated directories, contradictory statements on the web or an unclear website of your own. When a reliable source is missing, the model fills the gap with probability instead of facts.
Can false AI answers be corrected?
Yes, when the false statement comes from a source: we correct it where the system reads. If it comes from the model itself, it can only be overlaid until the model is updated. We measure both.
What does AI reputation protection cost?
CHF 500 per month plus VAT, as an add-on module to the monthly retainer: it can be added or removed individually and runs only with a running retainer. It is deliberately not available as a standalone subscription. In the base retainer, five questions about the company show you whether the assistants say something false about you; the module checks ten questions about the company every week and corrects what is false at the source, corrections and two pieces of content per month included.
Do you also delete negative reviews or critical articles?
No. We correct what is false; we do not suppress what is true. Reviews and press coverage are a matter for the platforms and editorial teams; where statements are factually wrong, we support the correction at the source.
How quickly does a correction take effect?
In our experience source corrections take effect within weeks, not days; model-internal statements take longer. We do not guarantee deadlines, but we do provide a correction log that documents every state.
How is this different from classic reputation management?
Classic online reputation management pushed unwelcome results to page two of the search results. AI answers have no page two: there is exactly one answer, and it cannot be pushed down, only corrected at the source. That is why we measure answers instead of nudging rankings.