An AI-generated report can make a recommendation sound settled. You can still ask where it came from. This short activity gives you something concrete to look for.

You do not need an AI account or any customer data. Everything below is fictional and prewritten. These are teaching examples, not results from a live AI test. Allow about five minutes.

Read the original comments. Imagine a small shop receives these two comments: “I couldn't find the opening hours.” “Are you open on Sunday?”

Now compare two possible recommendations. A: Make the opening hours easier to find. B: Open seven days a week because that is what customers want.

Before reading on, choose which recommendation you would investigate first. What can you point to in the comments to explain your choice? What would you still need to find out?

Compare your answer. A is a reasonable small experiment. One person could not find the hours, and another asked about Sunday. Clearer information might help. The comments do not prove that it will, so you would still watch what happens after the change.

B adds a much bigger conclusion. Asking whether a shop opens on Sunday is not the same as asking it to open every day. Two comments also cannot tell you what all your customers want.

That does not prove seven-day opening is a bad idea. It means you need more evidence before using these comments to justify it.

Turn the recommendation into a test. Try writing one sentence in this shape: Because [something in the original material], we could try [one small change] and watch for [something we can observe].

For this example: Because one customer could not find our opening hours, we could put them beside the contact details and watch whether questions about our hours become less common.

That gives you a reason for the change and a way to judge it. It does not promise an increase in sales.

Try the same check somewhere else. Here is a fictional delivery note: “Delivery costs £4. Delivery is free on orders over £40.”

A prewritten reply says: “You get free next-day delivery when you spend over £40.” Which part cannot be checked against the original note?

The added detail is next-day. The original says nothing about delivery speed. You could remove those words or check the delivery policy before using them. A fluent sentence is not evidence for a promise.

Keep this question: Can you show me which original words support that conclusion, and what you have added?

Use it when checking a summary, a suggested reply or a recommendation. You can do the check yourself; you do not have to give an AI tool private customer messages to practise it. If you try this with your own work, keep names, contact details and other private information out of anything you share.

You are practising the difference between an original fact, an interpretation and an unsupported addition. You do not need to know how the software was built to ask for that connection.

This is a companion to The book that couldn't be opened. In that edition, researchers use different tools to reveal existing writing, and people examine what the results support. Here, you use a small everyday example to practise asking what the source actually says.