How to Fix Wrong Brand Information in ChatGPT and Gemini: Trace the Source, Correct It, Test Again

You cannot edit a ChatGPT or Gemini answer directly. What you can fix is the source: a website page, a business profile, or an old article the AI read. Trace where the wrong information came from, correct it there, then ask the same question again. The feedback button helps, but it is not a correction channel.

This guide follows that order: check what is wrong, pick the riskiest errors, trace the source, fix it in six places, then test again. If you do not yet know which questions customers ask AI, start with how to find the questions customers ask AI.

Feedback and reporting steps were checked against OpenAI and Google help pages on 15 November 2026. Button labels can change.

Why does AI get your brand information wrong?

Usually for one of four reasons:

• The model's memory is out of date. A model is trained on data up to a certain date. A price or address that changed after that is unknown to it.
• Another page is wrong or old. A directory listing, a marketplace store nobody updated, or a three-year-old news story is still being read.
• Confusion with a brand of the same name. Common with generic names or names shared with a brand in another country.
• Your own site is unclear. The price is only in an image, key facts sit in a PDF, or two pages give different numbers.

It also helps to know the two paths an AI answer can take. One is model memory: the answer comes from what the model learned in training. The other is live retrieval: the AI searches the web at that moment and summarises what it finds. Google calls the second technique grounding: retrieving relevant, up-to-date pages from its Search index to improve the accuracy of AI Overviews and AI Mode. Which path is used decides how fast your fix shows up. More on timing below.

How do you check what is wrong?

Prepare ten questions customers typically ask: price, address, opening hours, products, return policy, contact, certifications, founder, competitors, and "are you still open?". Ask each one on ChatGPT, Gemini, Google AI Overview and Perplexity, three times per AI. Answers can change between attempts, so one try is not enough to draw a conclusion.

Log the results in a table like this (the rows below are illustrative):

Question	AI	Answer	What is wrong	Source cited	Risk
How much is Kopi Senja's basic package?	ChatGPT	Rp45,000	2024 price, now Rp55,000	Old blog review	High
What time does Kopi Senja open?	Gemini	08:00–17:00	Now open until 22:00	None	Medium
When was Kopi Senja founded?	Perplexity	2019	Should be 2018	Business directory	Low

Ten questions are enough for a first check. To monitor more seriously, read how many prompts you need to track.

Which errors should you fix first?

Sort them into three levels. The examples are illustrative.

• High: price, legal status, product safety, contact details, and "permanently closed". Example: AI says your shop has closed when it is still open. Buyers go straight to a competitor.
• Medium: features and product descriptions. Example: AI says you do not deliver outside your city when you do.
• Low: founding year and history details. Example: the founding year is off by one. Fix it, but later.

How do you trace the source of the error?

1. Open the citations in the AI answer. ChatGPT search, Perplexity and AI Overview usually show source links. Open each one and look for the wrong sentence.
2. Search the wrong claim on Google in quotes. For example "Kopi Senja" "Rp45,000". The pages that come up are the main suspects.
3. Check the usual suspects: Google Business Profile, marketplace store pages, business directories, old press coverage, and outdated pages on your own site.

If you cannot find any source, the wrong information most likely comes from model memory. The fix is the same: make sure the correct version is clearly available on the web. Only the timing differs.

What are the steps to fix it?

Fix it in these six places, starting with the ones you control most:

1. Create a "brand facts" page on your site. One page with price, address, opening hours, contact, legal details and main products, written as plain HTML text (not an image or PDF), with an "updated" date at the top. For writing pages that get cited, see how to get your articles referenced by ChatGPT.
2. Update or redirect old pages. Old promo pages or last year's price list: update the numbers, or 301-redirect them to the correct page.
3. Align your profiles elsewhere. Google Business Profile, Merchant Center, marketplace stores and social bios. Google itself says Merchant Center and Business Profile can help products and services be visible in AI responses. If you have no Google profile yet, see how to add your business to Google Maps.
4. Ask third parties to correct it. Send a short email to the directory, publication or blog carrying the wrong information. A template is below.
5. Make sure AI crawlers can read your site. Check robots.txt. According to OpenAI's documentation, sites that block OAI-SearchBot will not be shown in ChatGPT search answers. GPTBot is different: it is for model training, and you can set it separately.
6. Send feedback inside the AI product. In ChatGPT, click the thumbs-down icon under the answer. In the Gemini app, choose "Bad response" and pick a reason. In AI Overview, click thumbs down, choose "Report a problem" and describe the issue. This helps, but does not guarantee the answer changes.

Email template for third-party corrections:

Subject: Correction to [Brand Name] information on [page name]

Hi [publication/directory] team,

We are [Brand Name]. The page below:
[page URL]

says "[wrong sentence]". The correct information as of [date] is:
"[correct sentence]".

Official source: [URL of your brand facts page]

Could you please update it? Thank you.

[Name], [Role]
[Contact]

How long until the AI answer changes?

To be honest: no AI provider publishes an official correction timeline. If you have seen a "3–6 months" figure online, we could not find its source.

What we can say: answers built on live search can change after the source page is re-crawled. Answers from model memory can persist until the model is updated, and that schedule is not announced. For robots.txt, OpenAI says its systems take about 24 hours to adjust search results after the file changes, but that is about crawl permission, not answer content.

What does not work?

• Arguing with the chatbot. Correcting the AI in your own chat only affects that conversation. ChatGPT's Memory is tied to each account, so your correction does not show up in answers for other people.
• Creating an llms.txt file. Google's official guide says you do not need AI text files, special markup or Markdown to appear in Google Search, including its AI features.
• Buying or mass-posting mentions. Google also writes that seeking inauthentic "mentions" across the web is not as helpful as it might seem.

When do you need PR or legal routes?

If a claim is defamatory or dangerous (for example, your product is said to contain a banned ingredient), document it first: screenshots, dates, the question used, and which AI answered. Then use the official routes:

• Gemini: tap "More" under the answer, then "Report legal issue".
• Google Search and AI Overview: the Report content on Google tool, which has an option for AI-generated content.
• ChatGPT: the thumbs-down menu has an option for safety or legal concerns, and OpenAI has a Report Content form.

For serious cases, speak to a lawyer. This article is not legal advice.

How do you stop it happening again?

Repeat the ten-question check every month, and after every price or product change. Track the share of correct answers over time. That number is more useful than "were we mentioned or not". How to read it is in how to measure GEO performance. If the problem is that your brand is not mentioned at all, read how to get your brand mentioned in ChatGPT.

To understand why AI can sound confident while being wrong, see our guide to spotting AI hallucinations.

AI repeats what it finds. So make sure what it finds is the right version.

Start with the ten-question check above. To see several AIs' answers at once, try the free AI visibility audit. Disclosure: Intura builds Data AI Search, the platform that runs this audit.

Sources

• Google Search Central — Optimizing your website for generative AI features on Google Search
• Google Search Help — AI Overviews in Google Search (feedback)
• Gemini Apps Help — Send feedback or report a problem
• Google Legal Help — Report content on Google
• OpenAI — Overview of OpenAI Crawlers
• OpenAI Help — How your data is used to improve model performance
• OpenAI Help — Memory FAQ
• OpenAI — Report Content