What is an AI hallucination?
An AI hallucination is a statement from an AI model that sounds confident and plausible but is false or made up. It can be a wrong price, a feature your product never had, a quote nobody said or a source link that does not exist.
Hallucinations happen because a large language model predicts likely-sounding text rather than looking facts up in a database. When its information about a topic is thin, outdated or contradictory, it fills the gap with something that fits the pattern. Smaller brands, new products and details that changed after the model's knowledge cutoff are especially exposed.
Web search reduces the problem but does not remove it. An assistant can still misread a source, mix up two companies with similar names or combine facts from pages that disagree.
Why AI hallucinations matter for your brand
To a buyer, an AI answer often looks like a neutral summary. If it says your software lacks an integration you actually offer, or quotes an old price, you may lose the deal without ever knowing why. The error also repeats: everyone who asks a similar question can get the same wrong answer.
Hallucinations can hurt sentiment too. An invented lawsuit is the worst case, but even small errors about returns, delivery times or support hours can make your brand look unreliable next to competitors that are described correctly.
How to find and fix AI hallucinations
Test the questions buyers ask. Run prompts about your pricing, features, policies and comparisons in several assistants and note every factual error.
Trace the source. If the answer cites a page, check it. Outdated reviews, old press releases and directory listings are common culprits.
Correct it at the source. Update your own pages, ask third parties to fix old information and keep facts identical across your site, profiles and listings.
Publish clear facts. A plain FAQ, pricing page and company page, plus structured data, give search-enabled assistants accurate material for grounding their answers.
Keep measuring. Answers based on search can change once pages are recrawled. Answers from a model's memory may only change with a new model version.
Trackbase runs your prompts daily across 11 AI models, so you can spot wrong claims early. Our guide on fixing what AI gets wrong about your brand covers the full process.
Examples of AI hallucinations
An assistant tells users that a B2B payroll software vendor has no mobile app, although the app launched last year.
A Dutch bicycle webshop is said to offer free returns within 60 days, while its real policy is 30 days.
An accounting firm in Utrecht is confused with a similarly named firm in Belgium, including the wrong address and services.
An assistant cites a "2024 industry report" that ranks a marketing agency first, but neither the report nor its URL exists.
Frequently asked questions
Why does ChatGPT make things up about my company?
ChatGPT and other assistants generate the most likely-sounding answer based on patterns they learned. If there is little, outdated or conflicting information about your company, the model may fill the gaps with plausible guesses. Clear, consistent facts on your own site and on trusted third-party sources give it better material, especially when it searches the web before answering.
Can I report an AI hallucination?
Most AI assistants let users rate an answer or flag it as wrong, for example with a thumbs-down button, but there is usually no direct way for a brand to correct a model. The practical route is to fix the information the model relies on: your own pages, third-party listings and outdated articles. Then keep testing to confirm the answer improves.
Does web search stop AI hallucinations?
It reduces them but does not stop them. Grounding an answer in retrieved pages gives the model current facts to work with, yet it can still misread a source, merge details from different companies or rely on an outdated page. That is why you should also check answers from assistants that search the web, not only from models answering from memory.