A model keeps hallucinating about a product it has never seen

I asked our own AI assistant about a feature in our product, and it described one that has never existed in convincing, confident detail.

Beginner to intermediateRead 4 minDoing: under two hours6 steps

▶ Start

What you'll have at the end

A written log of 3 to 5 product questions where the model made something up, each marked confirmed-wrong

Before you start

You need

An AI assistant already answering real questions about your product, plus your own access to the one place the true answer actually lives, such as an internal doc, a codebase, or a ticket.

Not covered

How to stop the model from inventing these answers; this only proves and records that it happens.

Leans on

Catch the model contradicting itself

Pick a fact only your team knows
Ask it once, save the exact words
Reword the question twice more
Let it say it doesn't know

Turn the catch into a record

Check the claim against the real source
Repeat for four more real facts

Where to go next

Find your model's knowledge cutoff blind spot

go there when the wrong answer is about a real event that happened after training, not something that was never in it at all

Checked 15 Aug 2026

Part of the Large Language Models (LLMs) cookbook