What is an AI hallucination?

By the Variask team · Updated

Definition

An AI hallucination is a confident statement from an AI model that is false or not supported by any source, such as a wrong price, a feature that does not exist or an invented quote.

Also called Confabulation, Made up facts

With Variask: Variask checks AI answers against the facts you confirm, flags wrong claims and lists the likely sources, so your team knows what to fix.

Why AI models hallucinate

  • Models predict likely words, not verified facts, so a fluent wrong answer can come out as easily as a right one.
  • Gaps in training data, or old information past the knowledge cutoff, push the model to guess.
  • Search results can be outdated or wrong, and the model may repeat them or mix two sources.
  • Similar names, products or companies get blended together.

Common brand hallucinations

  • Old or invented prices and plans.
  • Features you never had, or ones you removed.
  • Wrong founders, locations, integrations or availability by country.
  • Confusing you with a company that has a similar name.

How to reduce hallucinations about your brand

  1. 1Find them: ask the questions buyers ask across engines and markets, and read the answers.
  2. 2Trace them: look at the sources cited next to the wrong claim.
  3. 3Fix the source: update your own pages and ask third party sites to correct old information.
  4. 4Make facts easy to find: a clear pricing page, an about page, structured data and an llms.txt file.
  5. 5Check again over the following weeks to confirm the answer changed.

What it means for brands

  • Hallucinations about your price or features can cost sales without you ever seeing it.
  • You cannot edit a model, but you can fix the pages it reads.

Questions

AI hallucination:
common questions

No. They can be reduced. Grounding answers in good sources and keeping your facts consistent across the web help most.

For serious errors, most AI companies have feedback options. Fixing the sources the model reads is usually faster and lasts longer.

Related terms

Keep exploring

How AI search works

Grounding

Grounding is when an AI model bases its answer on specific sources, such as live search results or a set of documents, instead of relying only on what it learned in training.

Read

How AI search works

Knowledge cutoff

A knowledge cutoff is the date after which an AI model has no information from its training data. Anything that happened later is unknown to the model unless it searches the web or is given the facts.

Read

Measurement

AI brand sentiment

AI brand sentiment is the tone in which AI assistants describe your brand: positive, neutral or negative, and the reasons they give.

Read

Measurement

Sources vs citations

Sources are all the pages an AI engine read or retrieved while writing an answer. Citations are the subset it shows to the user as links. A page can shape an answer without being cited.

Read

How AI search works

Knowledge graph

A knowledge graph is a database of entities, such as companies, people, products and places, and the relationships between them. Search engines and AI systems use knowledge graphs to understand who is who and what is what.

Read
Full glossary