# If AI gives false information about your brand: a guide to detecting and correcting it

> If AI says something false about your brand: how to spot it, which source to fix, how to report it to the engines and how to measure the result again.

Published: 2026-10-07
<https://geomono.com/en/blog/yapay-zeka-yanlis-bilgi-tespit-ve-duzeltme>

If AI is giving false information about your brand, the way to spot it is to ask questions that include your name and compare every claim in the answer with the information on your own site. A lasting correction is made not in the engine but at the source the error came from; the engines' feedback buttons complement this but do not replace it. Then the same questions are asked of the same engines again and the result is recorded.

## Where does false information come from?

The engines themselves state plainly that they can make mistakes. Claude's help page says Claude may sometimes produce incorrect or misleading answers. Google's [AI Overviews help page](https://support.google.com/websearch/answer/14901683?hl=tr) also notes that AI-generated responses can contain errors. According to Google's announcement, AI Overviews and AI Mode began rolling out gradually in Turkey from 18 February 2026; in other words, these answers can now appear in Turkish-language searches too.

There are three common sources of errors about a brand:

- **Outdated or conflicting pages.** An "About us" page on your site that was never updated, an old price list or the address of a branch that has closed may be the information the engine reads.
- **What others have written.** News sites, directories, complaint sites and forums may carry outdated or incorrect information about you.
- **Namesakes.** A page about another company, a person or a brand in another country with the same name may be read as being about you. The engine may present its founding year, headquarters or products as if they were yours.

Sometimes the error comes from no page at all: the engine may have generated a piece of information without any source it read. That is why you also need to record the sources the answer shows; if there is no source, there may be no page to correct either.

## How do you detect the error?

When measuring visibility, you ask questions that do not name the brand; when checking accuracy, by contrast, you need questions that include your name. The aim is to see what the engine knows about you. We explain the difference between the two measurements in [how to measure brand visibility](/en/blog/yapay-zeka-marka-gorunurlugu-nasil-olculur).

### 1. Prepare a fact sheet

List the information published on your own site that could be stated correctly or incorrectly: founding year, head office address, cities with branches, scope of products and services, returns and warranty terms, opening hours. This list is the basis of the comparison. You cannot compare information that is not written on your site.

### 2. Ask questions that include your name

Ask questions such as "When was X founded?", "Where is X's head office?", "Which cities does X serve?" and "How do I return a product I bought from X?" separately on ChatGPT, Gemini, Perplexity, Claude and Google. Record each answer, its engine, its date and the source addresses it shows.

### 3. Compare the claims one by one

Compare every verifiable claim in the answer with your fact sheet. Value judgements such as "user friendly" or "high quality" cannot be compared; set them aside. When you find a mismatch, look at your own site first: sometimes the outdated information is not in the engine but on your page.

## "No errors found" can mean three different things

The result of an accuracy check does not split into two, it splits into three. A report that writes two of them with the same sentence ends up telling a brand it could not check that "there is nothing false about you".

| State | What it means | What to do |
|---|---|---|
| The check could not be run | The engine did not answer, the call broke off or the answer could not be read | No result is written; the check is tried again |
| Nothing to compare against | An answer came back, but your site has no matching information or your site could not be read | "Could not be measured" is written; it does not count as clean |
| Checked, nothing false | The claims were compared with the information on your site and matched | This is a real result, and it is good news |

The second row is the one most often overlooked. For a brand whose site blocks AI crawlers, there is no information that can be read for comparison. Such a brand is among those most at risk of being described wrongly: the engines may be unable to read it from its own site and lean on other people's pages. A check that comes back empty and a brand at risk are often the same brand. You can see whether your site can be read with [our scan tool](/en/tarama).

> [!UYARI]
> When you read the sentence "we found nothing false about you", ask whether the check actually ran and how many claims were compared with information on your site. If zero claims were compared, the result is not clean; it means it could not be measured.

{{cta}}

## Correction: fix the source, not the engine

A lasting correction is made on the page the engine reads. The source the answer shows tells you where to start.

**Outdated information on your own site.** Update the page and make sure the same information is not written differently elsewhere on your site (the footer, the contact page, old blog posts). Google's [AI features and your website](https://developers.google.com/search/docs/appearance/ai-features?hl=tr) page lists, among its recommended practices, making sure structured data matches the visible text on the page and keeping Merchant Center and Business Profile information up to date.

**Incorrect information on another site.** Write to the owner of the page, together with a source showing the correct information. Respond to an outdated claim on a complaint site where it is read, in a calm tone.

**An error in a Wikipedia article.** Do not edit the article yourself; disclose your connection and propose the correction on the talk page with an independent source. We set out the details from Wikipedia's own guidelines in our post [Turkish Wikipedia or English Wikipedia](/en/blog/vikipedi-mi-ingilizce-wikipedia-mi).

**Namesake confusion.** State your full legal company name, your city, your founding year and what you do clearly on your site. For the engine to tell you apart from another company, the distinguishing information has to be readable.

**Information generated without a source.** If the answer shows no source at all, make sure the correct information is written on your site clearly and in an easy-to-find form, and use the engine's feedback channel.

## How do you report it to the engines?

Every engine has a feedback route below the response. The information below was taken from each company's own help page (accessed 7 October 2026). None of them promises that a report will change a particular answer.

| Engine | Official channel |
|---|---|
| ChatGPT | The thumbs-down icon below the response; a content reporting form for legal matters |
| Gemini | "Kötü yanıt" (Bad response) below the response; for legal matters, "Yasal sorun bildir" (Report a legal issue) under "Diğer" (Other) |
| Google AI Overviews | The thumbs-down icon, then "Sorun bildir" (Report a problem) |
| Perplexity | The flag icon below the answer, or a support request |
| Claude | The thumbs-down button below the response |

**ChatGPT.** According to OpenAI's [content reporting page](https://help.openai.com/en/articles/10245791-reporting-content-in-chatgpt-and-openai-platforms), you can report by tapping the thumbs-down icon below a message and choosing the problem; for legal matters there is a "Safety or Legal concern" option. The same page says reported domains may be reviewed by OpenAI's model quality team and that filters may be applied to keep ChatGPT from relying on untrustworthy sources. There is also a separate form for trademark infringement.

**Personal data.** If the false information concerns not a company but a person (your founder, for example), OpenAI's [personal data removal page](https://help.openai.com/en/articles/20001057-right-to-be-forgotten-and-personal-data-removal-from-chatgpt) explains that, under certain privacy laws such as the GDPR, a request can be made through the Privacy Portal. Requests are assessed individually, and information removed from ChatGPT is not thereby removed from other sites or search engines.

**Gemini.** According to Google's [Gemini help page](https://support.google.com/gemini/answer/13275746?hl=tr), you can select "Kötü yanıt" (Bad response) below the response and add a reason and a description.

**Google AI Overviews.** Google's help page describes clicking the thumbs-down icon and choosing "Sorun bildir" (Report a problem) for an incorrect or problematic overview. Your latest search query and the related search results are also attached to the feedback you send.

**Perplexity.** Perplexity's [help page](https://www.perplexity.ai/help-center/en/articles/10354902-how-can-i-report-incorrect-or-inaccurate-answers) asks you to report through the flag icon below the answer or through a support request; the report should include the link to the query and a description of the error. Incorrect and outdated information is listed among the issues they look at.

**Claude.** Claude's [help page](https://support.claude.com/en/articles/8525154-claude-is-providing-incorrect-or-misleading-responses-what-s-going-on) recommends reporting an unhelpful response with the thumbs-down button.

## Measure again

After fixing the source, ask the same engines the questions you used in the first check again, and record the results with their dates. If the questions change, the difference between the two checks may come from the questions rather than the correction.

Do not expect the result to show immediately. In the troubleshooting section on preview controls, Google writes that recrawling a page can take anywhere from a few days to a few months. We did not find a published timeframe for the other engines.

GeoMono's [accuracy audit](/en/urun/dogruluk-denetimi) runs inside the scan-based analysis: it compares the verifiable claims in the responses with your site's own pages, adds any contradiction to the analysis roadmap as a step, and does not write up a case where the audit did not run as if it were a clean result. This section is not currently included in registered measurement reports; the report says so explicitly.

## Frequently asked questions

### Can I get AI to correct false information?

The engines have no correction button that changes a particular answer. Each engine has a feedback channel, but none of them promises that a report will change the answer. The lasting route is to fix the source the error came from and then check again with the same questions.

### The check found no errors at all. Is that good news?

It depends. If the check actually ran and the claims in the answer were compared with the information on your site and matched, yes. If the check could not be run, or there is no information on your site to compare against, this is not a clean result; it means it could not be measured.

### What happens if my site blocks AI crawlers?

The engines may be unable to read you from your own site and lean on other people's pages; the audit cannot compare against the information on your site either. This is the situation where the risk of being described wrongly is high while the check comes back empty. The first thing to find out is which crawlers cannot reach your site.

### When will the result of the correction show?

There is no fixed timeframe. Google writes that recrawling a page can take anywhere from a few days to a few months; we did not find a published timeframe for the other engines. So after the correction you need to ask the same questions again at regular intervals and record each result with its date.

## Sources

- [Reporting Content in ChatGPT and OpenAI Platforms](https://help.openai.com/en/articles/10245791-reporting-content-in-chatgpt-and-openai-platforms), OpenAI Help Center, accessed 7 October 2026
- [Right to be forgotten and personal data removal from ChatGPT](https://help.openai.com/en/articles/20001057-right-to-be-forgotten-and-personal-data-removal-from-chatgpt), OpenAI Help Center, accessed 7 October 2026
- [Gemini Uygulamaları'nda geri bildirim gönderme veya sorun bildirme](https://support.google.com/gemini/answer/13275746?hl=tr), Gemini Apps Help, accessed 7 October 2026
- [Google Arama'daki AI Bakışı özelliğiyle aradığınız bilgilere çok daha hızlı ve kolay ulaşma](https://support.google.com/websearch/answer/14901683?hl=tr), Google Search Help, accessed 7 October 2026
- [How can I report incorrect or inaccurate answers?](https://www.perplexity.ai/help-center/en/articles/10354902-how-can-i-report-incorrect-or-inaccurate-answers), Perplexity Help Center, last modified 3 September 2026
- [Claude is providing incorrect or misleading responses. What's going on?](https://support.claude.com/en/articles/8525154-claude-is-providing-incorrect-or-misleading-responses-what-s-going-on), Claude Help Center, updated 16 March 2026
- [Yapay zeka özellikleri ve web siteniz](https://developers.google.com/search/docs/appearance/ai-features?hl=tr), Google Search Central, accessed 7 October 2026
- [Türkiye'de AI Modu ve AI Bakışı Dönemi Başlıyor](https://blog.google/intl/tr-tr/urun-duyurulari/turkiyede-ai-modu-ve-ai-bakisi-donemi-basliyor/), Google Turkey Blog, 18 February 2026