How to Fix What AI Says Wrong About Your Brand

A potential client asks ChatGPT about your company before the meeting. The response mentions a case study you never had, attributes a controversial statement to your CEO they never made, or repeats a four-year-old criticism that no longer applies. The client walks into the call with an image of your brand that has nothing to do with reality.

It's not an isolated case. It's what's happening every day to most mid-size brands. AIs combine two-year-old data with poorly interpreted summaries, mix companies with similar names, and repeat errors no one has corrected because no one has done the audit.

The good news is it can be fixed. The bad news is the process isn't like asking a newspaper for a correction: it requires understanding where the AI gets its information from and working on the sources, not on the model.

Why AI Gets Your Brand Wrong

Before fixing anything, you need to understand what kind of error you have. AI reputation problems fall into four categories and each one is fixed differently:

Outdated information. Your company moved offices, launched a new product, shut down a business line, or changed CEO. The AI keeps telling the version from two years ago because its training corpus is old and the sources that rank for your brand haven't been updated either. It's the most common and easiest to solve.

Confusion with another brand. Your company has a name similar to another one operating in the sector. The AI mixes achievements, client cases, controversies, and financial data. If your brand is smaller, you'll be attributed problems from the bigger one. If it's the other way around, you'll be attributed successes that aren't yours but also criticisms that don't belong to you.

Old criticism that won't die. There was a controversy, a viral unhappy customer, a critical article three or four years ago. The matter was settled but that piece of content still ranks well and the AI keeps citing it as relevant. The old source keeps winning over new ones because it has more backlinks.

Pure hallucination. The AI invents a fact. It attributes a partner that doesn't exist, a fictional client case, a wrong revenue figure. There's no source: the model generated it by statistical similarity to similar brands. It's the rarest but also the hardest to fix, because there's no clear origin to address.

How to Run an AI Reputation Audit

You can't fix what you haven't measured. The initial audit gives you the complete map of what's happening with your brand in each model:

List the questions someone would ask about you. Not the ones you would ask, but the ones a potential client, journalist, job candidate, investor, competitor, or supplier would. "Tell me about [your company]". "What's it like to work at [your company]?". "What problems has [your company] had recently?". "Is [your company] reliable?". Ten questions minimum, covering different angles.

Run them through the five major LLMs. ChatGPT, Gemini, Claude, Perplexity, and Grok. Each has a different corpus and will give different responses. Testing only in ChatGPT isn't enough because each model has its own set of errors that repeat.

Document each response literally. Copy the exact text, don't summarize. You'll need the exact quote when you have to evidence the problem and trace where it comes from.

Classify errors by category. The four from the previous section: outdated, confusion, old criticism, hallucination. Each category has a different protocol.

Identify the sources. Ask Perplexity and ChatGPT with search enabled to cite sources for that specific claim. Sometimes it's outdated Wikipedia, sometimes a 2022 press article, sometimes a forum, sometimes the model can't give a source and that confirms it's a hallucination.

How to Fix Outdated Information

It's the most frequent case and the easiest. The process has three levers and it's worth attacking all three at once:

Update your own website first. If the AI says your CEO is someone who's no longer there, review the "team" section, the "about us" page, the footer, legal pages, and blog post bylines. The owned website is the source of truth and the AI eventually aligns with what the brand says about itself when it's consistent.

Update Wikipedia and public databases. If you have a Wikipedia page and it's old, edit it with verifiable sources the AI prioritizes (self-editing without external sources isn't accepted). Crunchbase, company LinkedIn, Glassdoor, business registries where you're listed. AIs give a lot of weight to these structured sources.

Generate new press about the change. A press note in sector publications about the change you want reflected. It doesn't have to be a national newspaper front page: with three or four sector media with authority you already build the new source fabric the AI will start prioritizing.

Realistic timeframe for the AI to reflect the change: between four and twelve weeks. Models with active search (Perplexity, ChatGPT with browse, Gemini with Google) pick it up sooner. Models without search only pick it up when retraining, which can take months.

How to Resolve Brand Confusion

This one is trickier because it's not just your work: it depends on how clear the signals are that differentiate your brand from the other one. The strategy is to generate context that unambiguously separates them:

Define your brand with unique descriptors. If your company is called "Lumen" and there's another "Lumen" in another sector, on your website don't write just "Lumen" but "Lumen, [your specific niche] platform for [your vertical]". The AI learns to associate unique descriptors with your entity.

Reinforce identity signals across all sources. Company LinkedIn, Crunchbase, Wikipedia if you have one, blog posts, client cases. Repeat the same descriptors. Consistency between sources is what teaches the AI that you're distinct entities.

Create comparative content when it makes sense. If the confusion is very frequent and the other brand is relevant, a post like "[Your brand] vs [Other brand]: why we're different companies in different sectors" can clarify the confusion. Only if the other brand is really known, otherwise you're giving them free relevance.

Use Schema.org Organization markup on your website. With sameAs pointing to your LinkedIn, Crunchbase, Wikipedia, official social. It's the most direct way of telling the AIs "this specific entity is me, not the other one".

How to Bury an Old Criticism

There's no magic here: you can't "delete" a four-year-old critical article. What you can do is generate enough new, relevant, and well-positioned content for the AI to prioritize new sources over the old one:

Audit which sources are feeding that criticism. It's usually an article, a Reddit thread, a Trustpilot review, or an old press note. A single source rarely sustains a reputation problem: it's usually two or three that cite each other.

If the criticism was valid and was resolved, tell it publicly. A post on your blog explaining "what happened in 2022 and what we've changed since then". This gives the AI a new, contemporary source that contextualizes the old one. Important: don't deny the problem, contextualize it.

Generate 5-10 new pieces of press or editorial content about your brand over the next 6 months. Case studies, interviews with your team, growth data, sector awards, new integrations. You flood the corpus with recent and positive information. The AI ends up giving more weight to recent volume than to the old piece.

Ask for old content to be updated or retired if it's factually incorrect. Some publications do update old posts when the information they contain is outdated. They don't always say yes, but asking is free.

How to Handle Hallucinations

Pure hallucinations are the most frustrating because there's no clear source to fix. The strategy is preventive, not reactive:

Create the source that should exist. If the AI is inventing that you have a partnership with such company, it may be because there's no clear information about your real partners. Create a "/partners" page on your website listing your real integrations. The AI, when it searches again, will have a concrete source and will stop inventing.

Be explicit about what you don't do. If the AI repeatedly attributes services you don't offer, a page like "what we do and what we don't do" or a FAQ with "do you do X? No, we focus on Y" plants clear signals that the model picks up. Knowing how your brand appears in each model helps you spot where to plant these signals.

Report the error directly when you can. ChatGPT, Gemini, and Claude have a feedback button on each response. Reporting specific errors doesn't guarantee anything but it's input those teams do look at when it's systematic.

It's not worth chasing isolated hallucinations. If one in twenty responses has a different invented error each time, there's no pattern to fix. It's only worth it when the same hallucination repeats consistently: that's when there's an actual model issue or an identifiable wrong source.

How to Monitor Your AI Reputation Continuously

Running the audit once is fine for a diagnosis. But AI reputation changes every month with each model update, each new review, each article published, each Wikipedia update. What you fixed in February can come back in September because someone wrote something new.

Mentio analyzes your brand across the five major LLMs with questions adapted to your sector and market, not just visibility but also sentiment and accuracy of information. The report shows what each model is saying about you, which errors repeat, which competitors are being recommended with more updated data than yours, and which content pieces are sustaining each error.

What was a four-hour manual audit becomes an automated analysis you can run monthly to detect regressions and new errors before they impact your sales.

Audit for free what AI says about your brand →

Your AI Reputation Is an Asset You Maintain, Not Build Once

The difference between brands AI recommends with correct data and those appearing with wrong information isn't size or budget. It's the discipline of auditing regularly and fixing what's broken.

A five-person company with an updated website, well-maintained Wikipedia, and three press notes a year will have better AI reputation than a multinational with a hundred live pages but nobody checking what LLMs say about them.

The cost of not doing it is paid in clients arriving at calls with wrong ideas, candidates rejecting offers based on outdated information, investors worrying about crises already overcome. That cost is invisible until you measure it.

Start free, no credit card required →

Want to know if AI mentions your brand?

Discover your visibility in ChatGPT, Claude and Gemini in minutes.

Analyze your brand for free

Related articles