What is an AI hallucination? Definition, causes and brand risks

An AI hallucination is a response from an AI model that states something false, unsupported or invented as if it were fact, such as a product feature that does not exist, a wrong price or a source that was never published.

Hallucinations are not a rare glitch. They follow from how language models write, and when the subject is your brand, the person reading the answer usually has no way to tell.

Why AI models hallucinate

A large language model generates the most plausible next words, not verified facts. It has no built-in check that tells it whether a sentence is true. Most hallucinations come from one of these situations:

• Gaps in training data. When a model knows little about a brand, it fills the gap with patterns from similar brands: a typical price, a typical feature list.
• Outdated knowledge. Anything that changed after the model’s knowledge cutoff is unknown to it unless it searches.
• Retrieval errors. The engine retrieves an old page, a page about another company, or mixes facts from two passages.
• Ambiguous names. A brand that shares its name with another company, a place or a common word is easy to confuse.
• Pressure to answer. Models are trained to be helpful, so they often answer instead of saying they do not know.

Hallucinations that affect brands

• Invented features or products that you do not offer.
• Wrong prices or plans, often old ones or a competitor’s.
• Outdated facts, such as a discontinued product, a previous name or old locations.
• Confusion with another company that has a similar name.
• Misattribution: a competitor’s feature, review or controversy assigned to you.
• Fabricated sources, such as links that lead nowhere.

Why hallucinations matter for your brand

AI answers read with the same confidence whether they are right or wrong, and users rarely check every claim. A made-up price can end a sale before it starts, and a missing feature can take you out of a shortlist. Unlike a wrong article, there is no single page to correct: the same error can be repeated to many people asking similar questions, in slightly different words each time.

How to detect hallucinations about your brand

1. List the questions that matter. Pricing, features, alternatives, reviews, “is it legit” and comparisons with your main competitors.
2. Ask them regularly on several engines. Answers vary between runs and between engines, so one check is not enough.
3. Read the full answer. Being named is not the same as being described correctly.
4. Follow the sources. If the answer cites a page, check whether the error comes from it.
5. Log each false claim and see whether it persists, spreads to other engines or disappears.

How to reduce hallucinations about your brand

• Publish the facts plainly. A current pricing page, a features page and a clear “about” page give engines something correct to retrieve.
• Keep them consistent. Align your site, profiles, directories and marketplace listings.
• Fix the third-party source. If a review site or comparison carries old information, ask for an update.
• Mark up key facts. Structured data does not stop hallucinations on its own, but it states facts unambiguously. Check yours with the schema validator.
• Disambiguate your name. Say what you are and where, especially if other companies share your name.
• Use feedback options. Where an engine lets users flag an answer, report clear factual errors.

How Mencoro helps you spot hallucinations

Mencoro stores the full text and sources of every AI answer for the prompts you track on ChatGPT, Perplexity, Google AI Overview and Google AI Mode, so you can read exactly what each engine says about you and which pages it cites. Each mention is classified by tone, and negative ones (criticism or a warning) lower your Favorability. Mencoro does not judge whether a claim is true: that check is yours, and having the answers side by side over time makes it fast. Mencoro guards against the same problem in its own data: a language model finds the brands in each answer, and every mention is checked against the visible text, so a brand the answer never named is not counted. See how Mencoro works and the AI brand monitoring feature, or check a single answer with the free ChatGPT rank checker.

Related glossary terms

• AI brand sentiment: the tone AI engines use when they mention you.
• Grounding: tying answers to sources, which makes errors traceable.
• AI reputation management: monitoring and improving how AI describes your brand.
• Knowledge cutoff: why models repeat outdated facts.