AI hallucination
A hallucination is an answer from an AI model that sounds confident and well written but is false or comes from no source at all: a number, a fact or a quote nobody ever said.
Select any words here to ask your Maestro about them.
The AI summed up sales_march.csv (date with no time, item, quantity, total) in six sentences, all in the same sure tone. How many do you think come from the file?
Why it happens: an LLM writes the likeliest continuation of your text, not the truest one. When it doesn't have the fact (it's not in its training data or in what you gave it), the likeliest sentence still sounds right, and it often comes with a specific number or name.
Where it shows up with data: figures that aren't in your file, columns that don't exist, suspiciously round totals, percentage changes with nothing to compare against, quotes and sources nobody published. It's likeliest when you ask something your data can't answer.
How to catch it: ask which columns and rows each number came from, check that those columns exist, and recompute what goes into your report. A prompt that asks the AI to show its work won't prevent every hallucination, but it makes them show.
You practice it in How do you know if the number an AI gave you is right?.
How it's used
A number from nowhere
sales_march.csv: date, item, quantity, total
- You: How much did the North store sell in March?
- marked: AI: The North store sold $184,300 in March, up 12% from February.The file has no store column and no February data. The number is exact, the sentence is sure of itself, and none of it came from your data.
- You: Which column did you get the store from? Show me the rows.The question that exposes it: ask for the source.
- AI: You're right, the file doesn't have a store column...
When to look closer
Does asking "are you sure?" help?
About the North store's $184,300
- You: Are you sure about that number?
- AI: Yes, I checked the data and the figure is correct.
- You: Please double-check it.
- AI: Apologies, the correct figure is $179,850.
- You: Show me the rows you summed and the column you took the store from.
You ask the AI whether it's sure of the number. Does that check it?
Ask your Maestro about AI hallucination
It answers from the free course page closest to this word, the way it answers while you study.
You: What does "AI hallucination" mean, in plain words?
Your Maestro: I'll explain it with a case of yours: it sticks better than a definition. Has an AI ever handed you a number you couldn't trace?
Asking about: Answers from the page "How do I know if the numbers my AI gave me are right? Check them before you share"
Only this browser keeps what you ask here. How it works
Learn more
Related terms
- LLM (large language model)
An LLM is a program trained on huge amounts of text to continue a piece of text one chunk at a time: it works out which chunk is likeliest to come next and writes it. ChatGPT, Claude and Gemini all run on an LLM.
- Prompt
A prompt is what you write to an AI model to ask it for something: the question or instruction, plus everything you give it to work with, like data, examples or the format you want back.
- Spot check
Taking a few rows spread across the data (from the start, the middle and the end) and comparing them by hand with what you expect.
- Reconciliation
Comparing a total you computed with the same total from an independent source, such as a bank statement or a register report, and explaining any difference.