Skip to main content
Mobius
Beginner

What is hallucination

Also known as: AI hallucination, hallucinations, confabulation

Definition

An AI hallucination is an answer an AI model makes up and states with confidence, such as an invented price, rule or source, because it predicts likely words.

Why it matters

A made up answer that reaches a customer becomes the business's answer: a wrong price, a rule you do not have, a promise you did not make.

Common mistakes

Reading a confident tone as proof that the answer is correct.

Related terms

Worked exampleIllustration

Parking and a child, at a fictional pilates studio

A fictional neighbourhood pilates studio answers questions on WhatsApp with an AI assistant. The question and the answers are invented for this example.

  1. The question

    A customer asks: "Is there parking, and can I bring my child to class?"

  2. Without the sheet

    The assistant answers: "Yes, free parking in the lot next door, and children are welcome in the Friday class."

  3. The check

    Against the studio's policy sheet, the parking is invented, since the sheet says nothing about it, and the child line contradicts "Classes are for adults only".

  4. The fix

    The owner attaches the sheet and adds: "If it is not in the sheet, say you will check."

  5. With the sheet

    The same question now gets: "Classes are for adults only. About parking, I'll check with the studio and get back to you."

The invented answer sounded exactly as sure as the true one. Confidence is not a check.

Illustration

Flat odds, a sure sentence

CustomerIs there parking, and can I bring my child to class?

AssistantYes,free parkingin the lot next door

Step 3 of 3: the likely next pieces

  • in the lot next door23%
  • nearby21%
  • behind the building17%

At 23%. Three near guesses in a row became one confident sentence, and none of it came from the studio.

Asked about something it was never given, the model still picks a next piece. The shares are almost flat, the sentence is not.

Spot the invented parts

CustomerIs there parking, and can I bring my child to class?

Without the sheetYes,free parking in the lot next door,and children are welcome in the Friday class.
  • Parking: the sheet says nothing about it, so this was invented.
  • Children: the sheet says classes are for adults only.

Two invented facts, written as surely as a true one. Nothing in the sentence tells you which parts to doubt.

Without the sheet, the assistant filled the gaps with likely words. The amber parts are not in the studio’s sheet. One switch gives it the sheet.

Definition

A hallucination is content a language model produces that is false or made up but reads as true: an invented price, opening hour, policy, legal rule, statistic, quote, source or link, written in the same confident tone as the correct parts. It happens because the model is built to produce likely text, not to check facts. Researchers at OpenAI argued in 2025 that the usual training and scoring reward a confident guess over "I don't know", the way an exam that gives no credit for a blank answer rewards guessing. NIST's Generative AI Profile of 2024 names the same risk confabulation.

Why it matters

Hallucinations cannot be switched off. They are reduced by giving the model your own documents to answer from, and by a rule that a person checks anything that reaches a customer.

Directly related: RAG, LLM, Prompt.

Improvement tips

  • Give the tool your own documents to answer from, and tell it what to say when the answer is not there: "I'll check and get back to you".
  • Ask for the source of every fact that matters, and open it before you use the fact.
  • Have a person check anything that reaches a customer: prices, dates, policies, allergens, legal or medical wording.
  • Send anything legal, medical or financial to a qualified person, not only to whoever knows the business.

Common mistakes

  • Trusting AI citations, statistics or links without opening them.
  • Believing one setting or one prompt can switch hallucinations off, when they can only be reduced.
  • Letting a customer facing assistant answer questions its documents do not cover, instead of passing them to a person.

Related terms

AI and DataAdvanced

RAG

RAG (retrieval-augmented generation) is a method where AI answers from your documents: it first finds the relevant passages, then writes the answer from them.

AI and DataIntermediate

LLM

An LLM (large language model) is AI trained on masses of text to predict the next piece of a sentence. That is how chat tools write answers, right or wrong.

AI and DataBeginner

Prompt

A prompt is the instruction you give an AI tool: the question, the task, the context and the format you want. A clearer prompt gets a more useful answer.

AI and DataBeginner

Generative AI

Generative AI is AI that creates new text, images, audio or code from a request. What it writes is a likely draft built from patterns, not a checked fact.

AI and DataIntermediate

Context Window

A context window is how much text, in tokens, a model can see at once: instructions, files, the chat so far and its reply. What falls outside, it cannot use.

AI and DataIntermediate

AI Model

An AI model is the trained file at the heart of an AI tool: it takes an input, applies patterns learned from data, and returns a prediction, a label or text.

Quick check

A studio's assistant tells a customer there is free parking next door. The policy sheet says nothing about parking. What is this?

Choose an answer

Frequently asked questions

What is an AI hallucination in simple words?
It is when an AI tool makes something up and says it as if it were true: a price, a rule, a quote, a source. The invented part sounds just as sure as the correct part, and that is what makes it risky.
Why do AI models hallucinate?
Because they are built to write the most likely next words, not to check facts. A 2025 paper by researchers at OpenAI argues that the usual training and scoring reward a confident guess over admitting uncertainty, like an exam that gives no credit for a blank answer.
What does a hallucination look like?
An invented price, opening hour, policy, legal rule, statistic, quote, source or link, written in the same tone as the true parts of the answer. Links and sources are the easiest to check: open them.
Can I switch hallucinations off?
No. You can reduce them: give the tool your own documents to answer from (RAG), tell it to say "I'll check" when the answer is not there, and have a person check anything that reaches a customer.
Why does my customer chatbot invent policies?
Usually because it was asked something its documents do not cover and was not told what to do then. Attach the current policy sheet, tell it to answer only from the sheet, and to pass the chat to a person when the answer is missing.
Is a hallucinating AI broken?
No. This is how these models work: they produce plausible sentences, and sometimes a plausible sentence is false. The same model that invented one detail can answer the next question correctly.
Will an AI tool lie to me on purpose?
No. The model has no intentions. It writes likely text, and likely is not the same as true. The result can still mislead, so treat every answer as a draft until it is checked.
Who should check AI answers?
Anyone who knows the business can check everyday answers: hours, services, policies. Anything legal, medical or financial goes to a qualified person before it reaches a customer.
What goes wrong if a hallucination reaches a customer?
The customer acts on it: arrives when the studio is closed, expects a discount you never offered, relies on a rule you do not have. The result is complaints and refund demands, and the customer hears it as the business's answer, not the tool's.
Why is it also called confabulation?
NIST, the US standards institute, uses "confabulation" in its 2024 Generative AI Profile for confidently stated but erroneous or false content, and notes that it is known colloquially as hallucinations. Both words name the same risk.
How can I spot an invented detail quickly?
Look first at the specifics: numbers, dates, names, quotes and links. Ask the tool where each one came from, and check the ones a customer will rely on against your own documents.
Does connecting the AI to my documents solve the problem?
It reduces it, it does not end it. The model answers from what it retrieves, so an old or wrong document gives an old or wrong answer with the same confidence. Keep the documents current and keep a person's check on what reaches customers.

Sources: Kalai, Nachum, Vempala and Zhang, "Why Language Models Hallucinate", OpenAI, September 2025 (arXiv 2509.04664), read 2026-09-25, NIST AI 600-1, "Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile", 2024, section 2.2 Confabulation (doi.org/10.6028/NIST.AI.600-1), read 2026-09-25

Last reviewed: 2026-09-25

I have worked with

What Clients Say

From different fields, at different stages of business

Dan Manto

Eclipse Capital

Real Estate Investment and Finance, USA

Arty McLabin

GameReady

Game Development Education and Outsourcing, International

Anna, Beautician

Anna

Beautician

After I finished a cosmetology course at a leading company, I was confused. I didn't know how to open a business, what to sell from home, or how to bring in clients. Since I reached out to Mobius, everything has changed. We built an organized plan with clear steps and no unnecessary risks. Today I see results, and I'm growing and developing every month.

Mark, Massage therapist, Gan Yavne

Mark

Massage therapist, Gan Yavne

When I finished a massage course, I imagined a different world. I found out it's a hard field, and financial stability isn't easy to find. When I came to Mobius, something changed. We started an organized process and built a business plan. Without Mobius? I probably would have given up. Today I believe in myself and in my own path.

Wondering where AI actually fits your business?

Alex helps Israeli businesses use AI and data where it pays off, and skip it where it does not. The first call is free.

+972 055-248-6151. Better on WhatsApp: I am in meetings most of the day and answer as soon as I am free.