Customers

Find what customers keep telling you: spot patterns in feedback with AI

Turn a pile of reviews and messages into clear themes. A copy-paste prompt, how to protect customer privacy, and a quick check so the counts are real.

Say you run a small cafe. Over the past few months, customers have told you things. Some left reviews online. A few sent emails. Others said something at the counter that you scribbled on a napkin. You have a feeling there is a pattern in all of it, but you have never had the time to sit down and find it.

This is a good job to share with AI. Reading a pile of comments and spotting what keeps coming up is something AI does well. The part that needs you is deciding what is true and what to do about it.

What AI is good at here, and where it slips

AI writing tools (ChatGPT, Claude, and Gemini are the well-known ones) are good at grouping similar comments together. If ten people mention the wait, it will usually notice.

But AI also makes things up. Ask it “what do customers think?” and it may sound very sure about a trend that is thin or not there at all. It is also poor at counting. If it says “most customers mentioned parking”, you cannot trust “most” until you check.

So the plan is simple. AI sorts. You count and decide.

Step 1: collect what customers have already told you

Gather real comments in one place. A plain document or note is enough. Good sources:

  • Online reviews (copy the text only)
  • Emails and messages asking questions or complaining
  • Notes you wrote after a conversation
  • Answers from a quick survey, if you have run one

Aim for at least 15 to 20 comments. With only two or three, you do not need AI. You can just read them.

Step 2: remove private details

Before you paste anything, take out full names, phone numbers, email addresses, and anything else that identifies a person. Replace names with “Customer 1”, “Customer 2”, and so on. Customers shared their words with you, not with an AI tool.

Step 3: a prompt you can copy

Paste this, then add your comments underneath:

Below are real comments from my customers. Group them into themes. For each theme, give it a short name, tell me how many comments belong to it, and quote one or two comments that fit it, word for word. Put positive and negative themes in separate lists. Use ONLY what is in the comments. Do not guess what customers meant, and do not add themes that are not there. If a theme only appears once, say so instead of calling it a trend.

[paste your comments here]

Two parts of that prompt matter most. Asking for word-for-word quotes lets you check each theme against the original. Asking it to say when something only appears once stops one loud customer from looking like a crowd.

Step 4: check the counts yourself

This is the step to not skip. Take the biggest theme on the list and look for those comments in your own notes. Did you really find that many? Are the quotes the exact words? AI sometimes tidies a comment so it fits a theme better than it did.

If a number looks off, ask:

Show me every comment you placed under that theme, numbered.

Now you can count them by eye. It takes two minutes and it is worth it.

Step 5: turn themes into one small change

A list of themes does not help your business until something changes. Ask AI to help you think, but keep the choice yours:

Based on the themes above, suggest three small, low-cost things a small business could try this month. For each, say which theme it answers and how I would know if it worked.

You will get ideas, some good and some not. Pick one that fits your budget and your space. Ignore the rest. You know your business far better than the tool does.

Step 6: close the loop

When you make a change, tell people. A short note on your door, your website, or your next email works well:

You told us the weekend wait was too long. We have added a second person on Saturday mornings. Thank you for saying so.

Customers who see their feedback used are far more likely to give it again. You can ask AI to polish the wording, but keep it honest and only claim what you have really done.

A few cautions

  • One source is not the whole picture. People who write online reviews are often the happiest or the angriest. Quiet customers are missing.
  • Do not use it to answer individual complaints. A real person who is upset deserves a reply from you, not a theme on a list.
  • Keep your original comments. If the summary ever seems strange, the source is there.

Try this today

Open your email and your reviews page. Copy 15 or more real customer comments into one note and take out the names. Paste them under the prompt from step 3, then check the biggest theme against your notes. Ten minutes, and you may learn the one thing customers have been trying to tell you.