# AI Reputation Management: When ChatGPT Gets You Wrong

> AI reputation management is the work of checking how ChatGPT, Perplexity, Gemini, Claude and Google's AI features describe your company, then fixing what they get wrong. You can't edit an answer directly. You find the source the answer leans on in its citations, correct that source, publish the right facts where engines already read, and re-run until the description holds.

By Othmane Khadri, founder of Earleads. Published 2026-10-16. Canonical: https://www.earleads.com/blog/ai-reputation-management/

## What is AI reputation management?

AI reputation management is the work of checking how AI assistants describe your company and fixing what they get wrong. It covers ChatGPT, Perplexity, Gemini, Claude and Google's AI Overviews and AI Mode. The goal isn't to be mentioned more. It's to be described correctly when you are. That means the right price, the right product and the right company, with no scandal from four years ago presented as current news.

Many buyers now read one AI paragraph about you before they reach your site. If that paragraph is wrong, you don't see it happen. There's no bounce, no lost-deal reason in the CRM, no review to answer. The prospect arrives with the wrong idea, or doesn't arrive at all.

The engines say plainly that this happens. OpenAI's [help page on ChatGPT search](https://help.openai.com/en/articles/9237897-chatgpt-search) warns that "search results and citations can be incomplete, outdated, or incorrect." Google's [AI Overviews help page](https://support.google.com/websearch/answer/14901683) goes further. "AI Overviews can and will make mistakes."

Independent testing agrees. In a study coordinated by the European Broadcasting Union and led by the BBC, journalists from 22 public service broadcasters in 18 countries reviewed more than 3,000 answers from ChatGPT, Copilot, Gemini and Perplexity. They found that 45% had at least one significant issue, and 20% contained major accuracy issues "including hallucinated details and outdated information" ([EBU, 22 October 2025](https://www.ebu.ch/news/2025/10/ai-s-systemic-distortion-of-news-is-consistent-across-languages-and-territories-international-study-by-public-service-broadcaste)). That study tested news, not brands, but the mechanism is the one that describes your company too. An assistant summarises sources it found or remembers, and sometimes summarises them badly.

You can't log into ChatGPT and edit the paragraph about you. What you can change is what it reads, and the rest of this guide is how to do that properly.

## The four ways AI gets a company wrong

Most wrong answers about a company fall into four types. Naming the type matters, because each one usually comes from a different kind of source and needs a different fix.

| Error type | What the buyer reads | Where it usually comes from | Where to fix it |
|---|---|---|---|
| Wrong or old pricing | A plan you retired, a price from last year, "no free trial" when there is one | Review and comparison listings, old blog roundups, your own archived pricing page | The listing or article itself, plus a dated plain-text pricing page on your site |
| Old products | A feature you removed, a product you sunset, a missing launch | Launch coverage, old docs, threads from the product's early days | Your docs and changelog, the threads, the roundups that list you |
| Confused entity | A company with a similar name, or a description of the dictionary word your brand is named after | Thin or inconsistent descriptions of you, a louder namesake | Your About page, profiles and listings, all stating the same one-line description |
| Stale controversy | An old outage, lawsuit, layoff or bad review cycle presented as current | Press coverage and forum threads from that period | Dated follow-up content on pages engines read, answered in the original threads where allowed |

**Wrong pricing is the most expensive one.** It's the error sales hears first. In a story that comes up in operator forums, a prospect pastes a ChatGPT paragraph into the chat describing a pricing tier the company killed a year earlier, and the sales rep spends the live call unwinding it. Nobody on the call made a mistake. The answer was simply a year out of date.

**Confused entities are the subtlest.** If your brand name is also a common word, or a bigger company shares it, engines may describe the wrong thing with total confidence. A cousin of this is positioning drift, where an engine files an enterprise product as "a beginner tool". You're visible, but in the wrong drawer.

**Stale controversies outlive their facts.** A thread about an outage three years ago can keep getting cited for "is X reliable" long after the fix shipped, simply because nothing newer answers the same question as directly.

## Brand visibility in LLMs is not the same as brand accuracy

Most AI visibility work counts appearances. Brand visibility in LLMs tells you whether you're in the answer. It doesn't tell you whether the answer is right.

When we track a client's buyer questions, every answer gets read three ways.

1. **Appearance.** Are you named at all, next to which competitors?
2. **Position.** When you're named, are you the main suggestion or one in a list of eight?
3. **Description.** What does the engine say you are, who you're for and what you cost?

The third reading is where AI reputation management lives. A brand that's named often but described as something it isn't can be worse off than one that's absent, because the wrong description travels.

We also separate two kinds of questions. Findability questions ("best payroll software for startups in Germany") test whether you get on the shortlist. Validation questions ("is [your brand] legit", "[your brand] reviews", "[your brand] pricing") test what an engine says once a buyer already knows your name. Stale controversies and old prices mostly show up there, so tracking findability alone misses most reputation problems.

![The AI reputation loop. Detect the error by reading each engine's description and logging what prospects repeat, find the source in the citations, correct it at the source, publish the right version where engines already read, then re-run the same questions and loop back if the description still drifts](/blog/ai-reputation-management/diagram-1.svg)
*Every correction runs the same loop. Finding the source in the citations is the step most teams skip.*

## Step 1: Read how each engine describes you

Start with what buyers actually see. Open ChatGPT, Perplexity, Gemini, Claude and Google, logged out where you can, and ask the questions a buyer would ask once they know your name.

- What is [brand] and who is it for?
- How much does [brand] cost?
- Is [brand] legit? Is it reliable?
- [brand] vs [main competitor]
- What are the downsides of [brand]?
- Does [brand] still offer [product or feature]?

Add five or ten findability questions and read how you're described when you appear.

Write down every factual claim each engine makes about you, one per line. Price, plans, products, leadership, markets, the category you're filed under. Mark each one correct, wrong or stale. This is the core of the AI accuracy check we include in every audit, and nobody can skip it, because a dashboard that counts mentions will happily count one that gets your price wrong.

Two rules make this reliable.

**Use the real apps, not just APIs.** API answers aren't always what a person sees in the app, so we run questions natively in each engine on paid subscriptions. If a tracking tool disagrees with a live run, trust the live run.

**Run each important question more than once.** Answers change from run to run. SparkToro had 600 volunteers run 12 prompts through ChatGPT, Claude and Google's AI a combined 2,961 times and found "a <1 in 100 chance" that two responses would list the same brands ([SparkToro, January 2026](https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/)). Wrong descriptions behave the same way, so look for the claim that keeps coming back.

## Step 2: Ask sales to log every AI claim prospects repeat

Your sales and support teams hear AI answers before you do. A prospect says "I read you don't integrate with Salesforce" or "ChatGPT said you start at $500 a month." That sentence is the most valuable reputation data you'll get, because it's an error that already reached a buyer and shaped a conversation.

Make logging it a habit, not a project.

- Add one field to the call notes or CRM, called "AI claim the prospect repeated". Free text, plus which assistant if they say.
- Give support the same field for tickets that start with "the AI told me".
- Review the log every two weeks with whoever owns AI visibility. Each new claim goes straight into the question list from Step 1.

For consumer brands, customer service and social carry the same signal ("ChatGPT said you ship to Canada"), and so does a "how did you hear about us" answer at signup, which we cover in [AI search analytics](/blog/ai-search-analytics/).

Without this step you fix the errors you happen to find. With it, you fix the ones costing you deals first.

## Step 3: Find the source in the citations

Most wrong answers have a source, and most engines show it. ChatGPT responses that search the web "may include citations," and OpenAI says you can select **Sources** "to view cited sources and other relevant links" ([OpenAI](https://help.openai.com/en/articles/9237897-chatgpt-search)). Google's AI Overviews link to the supporting pages too, and Google tells users to click them to double-check.

For each wrong or stale claim, open every cited source and find the sentence the answer leans on. You're looking for one of five things.

1. **Your own page.** An old pricing page that's still live, a blog post announcing a plan you retired, docs for a feature you removed. More common than anyone admits.
2. **A directory or review listing.** Software directories, review sites, marketplaces, business listings. These update on their own schedule and often keep old prices and plan names for years.
3. **A third-party article.** A "best tools" roundup, a comparison page, a review written before your last change.
4. **A forum or community thread.** Someone's experience from two years ago, still the most direct answer to the question.
5. **No citation at all.** The engine answered from what the model learned in training. This is the hardest case, covered in Step 6.

![Where a wrong answer comes from. An AI answer about your company draws on five kinds of source, namely your own old pages, directory and review listings, third-party articles, forum threads, and training memory with no citation. The first four can be corrected at the source. Training memory, greyed out, only changes as new sources outnumber old ones](/blog/ai-reputation-management/diagram-2.svg)
*Four of the five sources behind a wrong answer can be corrected directly. Memory only moves as the others pile up.*

Record the engine, question, source and sentence next to each claim. That's your correction list, and later it's how you'll tell a fix that worked from a coincidence.

If you want the wider picture of which pages engines cite for your category, not just the wrong ones, the method is in [how to get cited by AI](/blog/how-to-get-cited-by-ai/).

## Step 4: Correct it at the source

Now fix each source, in the order it's costing you. As a rough rule, errors sales has heard go first, then errors on validation questions, then everything else.

**Your own pages.** Update or redirect the page. If an old pricing page must stay live for existing customers, add a clear dated note at the top ("Pricing as of October 2026: see current plans") and link the current page. Remove retired plan names from anywhere they appear as if current. Check that the facts are in the HTML as plain text, not only inside images, tabs or scripts, because a crawler that can't read the correct price will keep repeating the old one.

**Directory and review listings.** Claim every profile you can and update pricing, plans, category, description and screenshots. Use the same one-line description everywhere. Where you can't edit, contact the site with the specific change and a link to your source page.

**Third-party articles.** Email the author or editor with the exact sentence that's wrong, the correct fact and a dated source on your site. Keep it short and factual. If you're one entry in a long list, give the concrete details that make an entry useful (pricing model, integrations, limits), because answers often pull just two names from a list of ten.

**Forum threads.** If the community allows it and the thread is open, add a reply that states the current fact plainly and links the source, posted by a real person who's transparent about working for you. "This was true in 2023, we changed it in 2025, here's the page" is enough. If the community removes vendor replies, leave the thread alone and answer the question better somewhere else.

**Wikipedia.** If an article about your company carries the error, don't edit it yourself. Wikipedia's [conflict of interest guideline](https://en.wikipedia.org/wiki/Wikipedia:Conflict_of_interest) says people with a COI are "strongly discouraged from editing affected articles directly" and "may propose changes on talk pages" using the edit COI template. Post the correction there with an independent source.

**Google's own panels.** If the error shows in Google's knowledge panel, Google says the panel "generates automatically based on public information on the web" and that anyone can send feedback, with verified owners' feedback prioritised ([Google](https://support.google.com/knowledgepanel/answer/7534842)). Claim it if you can.

## Step 5: Publish the correct version where engines already read

Removing the wrong fact doesn't guarantee the right one replaces it. The correct version has to exist on pages engines already retrieve for the same question.

On your own site, that usually means one plain page per fact buyers ask about. A pricing page with plan names, starting prices and what changed, dated. A product page or changelog that says what you no longer offer and when it stopped. An About page that opens with one sentence saying what you are, who you're for, and, if you're often confused with someone, who you're not. We explain the consistent-description work in more depth in [how to get ChatGPT to recommend your business](/blog/how-to-get-chatgpt-to-recommend-your-business/).

Off your site, mentions appear to matter more than links. Ahrefs studied 75,000 brands and found that branded web mentions correlated with AI Overview visibility at 0.664, against 0.218 for backlinks ([Ahrefs](https://ahrefs.com/blog/ai-overview-brand-correlation/)). So the right facts need to show up in the places that write about you: the roundups and comparison pages already cited for your questions, the listings, and the community threads where buyers ask. Start with the pages already cited on your questions, not the ones with the most authority.

If you've renamed or repositioned, write "New Name, formerly Old Name" in every source you control and keep the old name in community content until engines describe the new one unprompted. Dropping the old name overnight throws away everything engines already associate with you.

When a page changes, tell search engines. [IndexNow](https://www.indexnow.org/) is "a simple ping so that search engines know that a URL and its content has been added, updated, or deleted," supported by Microsoft Bing, Naver, Seznam.cz, Yandex and Yep. Resubmit your sitemap in Search Console for Google.

## Step 6: Re-run until the description holds

Go back to the exact questions from Step 1 and run them again, several times per engine over several weeks. A single correct answer might be luck.

The fix holds when the wrong claim stops coming back across runs and engines, and when sales stops hearing it. Until then, keep looping. Re-read the citations, look for a source you missed, and check whether the engine is now citing a page you haven't fixed yet.

Expect different speeds. Answers that search the live web can change once a corrected page is recrawled. Answers that come from training memory, with no citation, change slowly. They move as new, consistent sources outnumber the old ones and new model versions ship. That's why Step 5 matters even after the old source is fixed. The right description has to become the one the web repeats most.

Citations also churn, so keep the accuracy questions in your regular tracking. Drift gets caught on a schedule instead of on a sales call.

![Timeline of a correction. Answers that search the web change after the corrected page is recrawled, while answers from training memory change only as new sources accumulate and new model versions ship, so re-runs continue until the claim stops coming back](/blog/ai-reputation-management/diagram-3.svg)
*Live answers move after a recrawl. Memory moves slowly. Re-run until the wrong claim stops coming back.*

## Should you report the wrong answer to the AI company?

Yes, but don't rely on it. Every engine has a feedback button, and using it costs a minute. Google's thumbs down on an AI Overview lets you choose **Report a problem**, and Google notes that the feedback "includes your most recent search query and its search results" ([Google](https://support.google.com/websearch/answer/14901683)). Most assistants offer a similar rating on each response.

Feedback won't rewrite how a model describes a company on demand, and there's no business listing inside ChatGPT to claim. Report the answer, then do the source work.

## What changes for B2B and B2C companies

**B2B.** The errors cluster around pricing, integrations, security and compliance claims, company size and "who is it for". They hurt most on validation questions, because by the time a buying committee asks "is [brand] SOC 2 compliant" or "[brand] vs [competitor] pricing", you're already on the shortlist. If your pricing isn't public, engines fill the gap with whatever a directory or thread says. A starting price or a plain explanation of how pricing works closes that gap, which we cover in [GEO for B2B](/blog/geo-for-b2b/).

**B2C.** The errors cluster around price, stock, shipping countries, returns, discontinued products and "is this store legit". Product data is the usual source, from feeds, marketplace listings and review sites that kept an old price. Fix the product record at the source and let it flow to every channel, rather than patching one channel by hand. If you sell online, check the product side with [is your store visible to AI shopping agents](/blog/is-your-store-visible-to-ai-shopping-agents/).

## What doesn't work

- **Arguing with the chatbot.** Telling ChatGPT it's wrong in your own chat changes your conversation, not what the next buyer sees.
- **Publishing a "correction" nobody cites.** A press release rebutting AI doesn't help unless engines read it. Put the fact on pages already retrieved for the question.
- **Ranking yourself first in your own "best of" list.** Readers discount it, and it's a weak source for the very facts you're trying to fix.
- **Paying for reviews or placements that look independent.** It's a reputation problem waiting to happen, and it's the opposite of what you're trying to repair.

## AI reputation management tools vs an accuracy check

AI brand monitoring tools are good at counting. They'll show how often each engine names you, against which competitors, and alert you when that shifts. Use one as the record if you have it.

They're weaker at the job this guide describes. They only track the prompts you give them, many query engines through APIs rather than the apps buyers use, and tracing a wrong sentence to its source and getting it changed is human work.

So split the work. A tool for the ongoing count, a monthly hand-run check on the questions that matter, sales logging what prospects repeat, and a correction log that ties every fix to the claim it addressed. That's also what an [AI visibility audit](/blog/what-an-ai-visibility-audit-should-show-you/) should hand you, with an accuracy check that lists each claim engines make about you, what's wrong or stale, and the source each answer leans on.

## Frequently asked questions

### Can I ask OpenAI to correct what ChatGPT says about my company?

Not in the way you'd edit a listing. There's no form that rewrites how ChatGPT describes a business. What you can change are the sources it reads. OpenAI's own help page says search results and citations can be incomplete, outdated or incorrect, so the practical fix is correcting the cited page, then publishing the right facts on pages ChatGPT already retrieves.

### Why does ChatGPT have wrong information about my company?

Usually because a source it trusts is wrong or old. A review listing with last year's pricing, a comparison article written before a product change, a thread describing a feature you've retired, or a company with a similar name. Answers that don't search the web lean on training data, which lags your changes by months or longer.

### How long does it take to correct ChatGPT information?

It depends on where the error comes from. When the wrong fact sits on a cited page, correcting that page can change answers that search the web once it's recrawled. When the error lives in the model's training data, it moves only as fresh sources outnumber old ones and new model versions ship. Re-run the same questions until the description holds across several runs.

### What is the difference between AI reputation management and online reputation management?

Classic online reputation management works on what people find, such as reviews, search results and press. AI reputation management works on what assistants say, which is assembled from those same sources plus forums, directories and comparison pages. The sources overlap, but the job differs, because an AI answer can repeat one stale page as fact to every buyer who asks.

### Do AI brand monitoring tools fix wrong answers?

No. They count and alert. A tool can show that an engine names you less often or describes you oddly, but it only tracks the prompts you give it, and many query through APIs rather than the apps buyers use. Use one as the record, then check the important questions by hand and fix the sources yourself.

### Does AI reputation management matter for consumer brands too?

Yes. Shoppers ask assistants which product to buy, whether a store is legit and what something costs. A wrong price, a discontinued model or an old recall story repeated in an answer costs sales the same way it costs a B2B company a deal. For B2C, product data, review sites and marketplace listings are usually where the error starts.