AI
Hallucinations
A practical guide to understanding why AI generates false information and what to do about it.
What AI hallucinations are, why they happen, how to spot them, and how to design AI products that account for them.
What it is
An AI hallucination is when a language glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term generates information that is factually incorrect, fabricated, or entirely made up — but presents it confidently as if it were true.
Hallucinations happen because language glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term do not retrieve facts. They generate text based on glossaryPatternA reusable solution to a common design problem.Open glossary term learned during training. When asked about something outside their knowledge or in an area where guideTraining DataWhat training data is, how it shapes what an AI model knows and assumes, and what product and design teams need to understand about its role in AI product quality.Open guide was limited, they fill the gap with what sounds most plausible.
This can produce convincing-sounding but completely wrong answers — including invented citations, fictitious statistics, and events that never happened.
Hallucinations are not a sign of a broken glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term. They are a known characteristic of how these glossarySystemA system is a collection of interconnected components that work together to achieve a specific function or outcome.Open glossary term work.
The practical implication is that any AI-generated content that will be used in a real glossaryContextThe surrounding conditions that shape behaviour and decisions.Open glossary term needs to be validated before it is relied on.
When to use it
Understand when hallucinations are a serious risk. They are most problematic when:
They are a lower risk when:
Key takeaway
Assume hallucinations will happen. Design your AI features around that assumption rather than hoping they will not occur.
How it works
Understand the basic mechanism. Language glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term generate glossaryResponseA response is the data or result returned by a server after receiving a request.Open glossary term by predicting the most likely next token given everything that came before it. They do not look up facts — they glossaryPatternA reusable solution to a common design problem.Open glossary term-match from training.
When a glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term does not have reliable guideTraining DataWhat training data is, how it shapes what an AI model knows and assumes, and what product and design teams need to understand about its role in AI product quality.Open guide on a topic, it still generates a confident-sounding glossaryResponseA response is the data or result returned by a server after receiving a request.Open glossary term based on what typically surrounds similar questions. That is where hallucinations come from.
The glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term has no internal mechanism to signal uncertainty in the way a human expert might say they do not know.
What this means for designers and product teams. Every AI glossaryFeatureA feature is a specific piece of functionality within a product that delivers value to users. It represents something users can do or experience as part of the overall product.Open glossary term that surfaces factual information to users carries hallucination risk. That risk needs to be designed around, not ignored.
Design choices that help include grounding the AI in a verified knowledge source (RAG), building in human review before outputs are used, displaying glossaryConfidenceConfidence is the level of certainty in a decision or outcome based on available evidence.Open glossary term glossarySignalsSignals are data points or triggers that indicate changes in user behaviour, context, or external factors.Open glossary term where appropriate, and making it easy for users to verify or question AI-generated content.
What to look for
Focus on:
Where it goes wrong
Most issues come from: If you ship AI-generated content without a validation step, hallucinations will eventually reach your users.
What you get from it
Understanding hallucinations gives you:
Key takeaway
Hallucinations are not an edge case — they are a regular occurrence. The question is not whether they will happen, but whether your design accounts for them.
FAQ
Common questions
A few practical answers to the questions that usually come up around this method.
What is an AI hallucination?
It is when an AI generates information that is factually wrong or entirely made up, but presents it confidently as if it were accurate. The glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term is not lying — it is glossaryPatternA reusable solution to a common design problem.Open glossary term-matching from its guideTraining DataWhat training data is, how it shapes what an AI model knows and assumes, and what product and design teams need to understand about its role in AI product quality.Open guide and producing what sounds most plausible, even when it does not have reliable information to draw from.
Why do AI hallucinations happen?
Because language glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term generate text based on statistical glossaryPatternA reusable solution to a common design problem.Open glossary term, not verified facts. When a model encounters a topic where its guideTraining DataWhat training data is, how it shapes what an AI model knows and assumes, and what product and design teams need to understand about its role in AI product quality.Open guide is limited, it still produces a response — it just has less reliable information to base it on.
Can hallucinations be prevented?
Not entirely, but they can be significantly reduced. Grounding the glossaryModelA model is a system or representation used to process data and generate outputs, often trained to perform specific tasks.Open glossary term in a reliable knowledge source (RAG), using well-structured glossaryPromptA prompt is the input or instruction given to an AI system to guide its output or response.Open glossary term, and building in human review are the most effective approaches. Hallucinations cannot be eliminated, but their impact can be managed.
How do I spot a hallucination?
Look for specific facts, statistics, or citations that cannot be verified. If an AI gives a precise number, names a specific study, or cites a source, check it. Hallucinations tend to be specific and confident — vague answers are usually less likely to be fabricated.
Should I tell users that AI can hallucinate?
Yes. Being transparent about the limitations of AI glossaryFeatureA feature is a specific piece of functionality within a product that delivers value to users. It represents something users can do or experience as part of the overall product.Open glossary term glossaryBuildA build is the process of compiling and packaging code into a runnable application.Open glossary term glossaryTrustUser confidence that a product, service, or organisation will do what it promises.Open glossary term and helps users engage with outputs critically. Hiding the fact that AI can be wrong and then having it get something wrong is much more damaging than being upfront about limitations from the start.
Quick take
AI hallucinations are not glitches — they are a fundamental characteristic of how language models work, and designing around them is part of the job.
Related Services