AI & Infrastructure
Hallucination

Hallucination

An AI hallucination is an answer that sounds confident but is false or unsupported, such as an invented fact, quote or citation. Language models produce them when they guess instead of admitting they don’t know.
June 27, 2026
Cihan Geyik
Table of Content

Ask an AI assistant about a mid-sized software company and it may quote a price dropped last year, or an integration that never existed. That’s a hallucination: a false claim delivered in the same calm tone as a true one. For brands it’s a reputation problem that reaches buyers before anyone at the company sees it.

Confident and wrong look the same. A model’s tone gives no signal of whether a claim is true. Only a check against a source does.

What is an AI hallucination?

The term covers any output that sounds plausible but is false or unsupported. Some hallucinations are pure inventions. Others are real facts pinned to the wrong company, or facts that were true once and aren’t now.

AI search adds its own variant, the citation mismatch. The link is real, but the page doesn’t say what the answer claims.

TypeExample about a brandUsual cause
FabricationA feature or award that doesn’t existThe model fills a gap with a likely-sounding detail
Outdated factLast year’s price or plan namesTraining data older than the change
Entity mix-upAnother company’s founder or reviewsSimilar names and thin entity data
Citation mismatchA cited page that says something elseA source misread or squeezed too hard
Wrong comparisonA competitor’s limit stated as yoursVague or one-sided comparison pages

Why do language models hallucinate?

A large language model predicts likely text. It has no built-in sense of which predictions are facts. In September 2025, researchers at OpenAI and Georgia Tech argued that training and evaluation reward guessing over admitting uncertainty. A model that guesses scores better on most benchmarks than one that says "I don’t know."

The same paper ties errors to rare facts. Its example: if 20% of birthday facts appear exactly once in pretraining data, expect base models to get at least 20% of birthdays wrong. Details about smaller companies often show up only a handful of times online. That puts them in the same risk group.

How often do models hallucinate?

It depends on the model and on whether it’s allowed to abstain. OpenAI’s summary of the research compares two models on the SimpleQA benchmark. The older o4-mini answered almost everything, with 24% accuracy and a 75% error rate. The newer gpt-5-thinking-mini abstained on 52% of questions, scored 22% accuracy and erred 26% of the time.

Accuracy barely moved. The willingness to say "I don’t know" cut errors by about two thirds.

Why are hallucinations a brand risk?

  • Buyers act on the answer. A wrong price or a missing feature can drop you from a shortlist without a single visit to your site.
  • Errors repeat. A wrong fact can resurface across thousands of similar prompts until its source changes.
  • Your own bots count too. In February 2024, a British Columbia tribunal held Air Canada liable for wrong bereavement-fare rules its website chatbot gave a customer. The case is Moffatt v. Air Canada, 2024 BCCRT 149.

Hallucinations feed straight into your AI reputation, the overall picture answers paint of your brand.

Does grounding stop hallucinations?

It reduces them. Grounding gives the model retrieved sources to write from, so claims can be checked and cited. The retriever can still pick a weak page, though, and the model can still misstate what a source says. Errors tend to survive where sources are thin or contradict each other.

How to detect and reduce hallucinations about your brand

Run buyer prompts across engines → Pull out claims about your brand → Check each claim against your facts → Trace wrong claims to cited sources → Fix the source or publish the correct fact
  • Publish the facts models get wrong in plain text: pricing, plans, integrations, founding year and headquarters. Use the same wording everywhere.
  • Correct third-party profiles that carry old data, such as review sites, directories and partner listings.
  • Describe your company as a consistent brand entity, with matching names and sameAs links across profiles.
  • Re-test after each fix. Retrieval-based answers can change once the corrected page is indexed. Training-based errors wait for the next model version.

Setting up the prompt runs is the first step. Our guide on tracking brand mentions in ChatGPT covers it.

How Ansvisor helps

Ansvisor’s Answer Engine Insights shows how AI answers describe your brand, including recurring themes and the terms tied to it. Filters for platform, topic, region and model narrow the view to the markets you care about. AI Citation Sources lists the domains and URLs that shape those answers, which is where wrong claims usually trace back.

AI Hallucination, LLM Hallucination, Model Hallucination, AI Confabulation, Fabricated Output

Let AI Search Work for Your Growth

Each prompt you win brings more mentions, more citations and more buyers. Ansvisor keeps finding the next one.

AI Answers Mentions Citations Visibility Score AI Share of Voice AI Traffic Revenue
Win Back Search VisibilityTry Ansvisor 14-day Free →✓ Zero risk. Cancel anytime.

FAQ

Frequently asked questions.

What is an AI hallucination?

An AI hallucination is output that sounds plausible and confident but is false or not supported by any source. Examples include a made-up statistic, a fake quote, a product feature that doesn’t exist or a citation pointing to a page that says something else.

Why do AI models hallucinate?

Models are trained to predict likely text and are graded in ways that reward guessing over admitting uncertainty. A 2025 paper from OpenAI and Georgia Tech researchers adds that facts appearing only once in training data are especially error-prone. Details about smaller companies often fall into that group.

Can AI hallucinations be eliminated?

Not with current methods. Grounding answers in retrieved sources, letting models abstain when unsure and better evaluations all reduce them. Even grounded answers can misread a source or attach a citation that doesn’t support the claim.

How do AI hallucinations affect brands?

An AI answer can state wrong prices, invent features, confuse your brand with a similarly named one or repeat outdated facts. Buyers may act on it without checking. In 2024, a Canadian tribunal held Air Canada liable for wrong information its own website chatbot gave a customer.

How can I find hallucinations about my brand?

Run the questions buyers ask about you across several AI engines and repeat them over time. Compare each claim with your own facts, and check whether the cited pages support it. Errors that repeat across engines usually trace back to an outdated or wrong source.

Help us grow the AI Visibility Glossary

New terms are added regularly.

Help us improve the page or suggest a new term →
About the Author
Cihan Geyik

Cihan Geyik

Co-founder at Ansvisor

Cihan Geyik is the co-founder of Ansvisor, an open-source AI Visibility platform for AI Search. With more than 15 years of experience in digital marketing and growth, he writes about AI visibility, AI search, AEO, GEO, citations, and answer engines. He focuses on helping brands understand and improve their presence across ChatGPT, Gemini, Google AI Overviews, Bing AI Mode and other AI answer engines.

Google Preferred Sources
Get more Ansvisor insights in Google

Add Ansvisor as a preferred source and Google will show you more of our AI search, AEO and GEO guides.