Turn customer feedback into themes

A folder of reviews, survey responses, or support notes that you know contains something useful, if only you had time to read all of it closely. This pulls out the patterns without you reading every line.

Redact first

Remove customer names, emails, order numbers, or anything else identifying before pasting feedback anywhere. The wording and sentiment are what you need; who said it specifically almost never is.

Do this

1. Gather the feedback, as much as you reasonably have. More data means real patterns instead of noise.
2. Remove identifying details.
3. Ask for themes, not a rewritten list, using the prompt below.
4. Read the representative examples for each theme yourself, in the customers’ actual words, before deciding what to do about it.

Here is customer feedback (names and identifying details removed):
[paste it]

Group this into themes: recurring sentiments or issues that show up
more than once, in different words. For each theme:
- Roughly how often does it appear?
- One or two representative quotes, in the customer's own words
- A possible action, if there's an obvious one, otherwise say
  "no clear action, worth discussing"

Don't force an action onto every theme. Some are just worth knowing.

Check this before you trust it

• Every representative quote, against the original feedback. A pulled quote should mean the same thing in isolation as it did in context.
• Whether a “theme” is actually several people, or one person restating the same complaint. Ask it to note distinct customers behind each theme if that distinction matters to you.
• Any suggested action. A plausible-sounding fix based on a handful of comments is a hypothesis, not a roadmap decision.

If it goes wrong

Everything gets grouped into one enormous theme. The grouping was probably too broad. Ask it to split the largest theme into more specific sub-themes.

The representative quotes all sound the same, flattening real differences. Ask for quotes that show the range within a theme, not just the clearest example, so you don’t miss a milder or more severe version of the same issue.

A theme with real volume gets marked “no clear action.” That’s an honest answer, not a gap to fill artificially. Some patterns are worth knowing about without there being an obvious fix; discuss it with your team rather than manufacturing a response.

The feedback set is too small to show real patterns. Say so rather than forcing themes onto a handful of comments. A small batch is a sample, not a dataset, and treating it as more than that risks reacting to noise.