Finding the pattern in customer feedback

What it is good at

• Can process a large volume of reviews or tickets in one pass that would take hours manually.
• Good at grouping different wordings of the same underlying complaint together.
• Can attach a frequency estimate to each theme so you know what actually matters most.
• Useful for pulling representative quotes you can share with a team without reading everything.

Where it falls short

• Frequency counts are only as reliable as the completeness of the feedback you paste in.
• It can conflate distinct issues that use similar language, spot-check the groupings.
• It won't catch sentiment nuance like sarcasm as reliably as a careful human reader would.

How to use it for creative work

1. 01

   Gather feedback from every source into one place

   Pull reviews, support tickets, survey responses and social comments together rather than analysing each source separately, patterns often only appear once everything's combined.

2. 02

   Ask for themes with frequency and quotes

   Request the recurring themes, how many times each appeared (roughly), and one representative quote per theme, so you can trust the ranking and show real evidence to your team.

3. 03

   Separate the fixable from the unfixable

   Ask which themes are things you could realistically act on this month versus structural issues (price, category limitations) you can't change quickly, this stops action-planning from stalling on the wrong item.

4. 04

   Spot-check the groupings

   Read the quotes behind your top two themes yourself to confirm they're genuinely the same issue and not just similar wording for different problems.

5. 05

   Turn the top theme into an action

   Take the single most frequent, fixable theme and ask for three concrete ways to address it, then pick one to implement and track whether the complaint frequency drops next cycle.

Ready-to-use prompts

Here is a batch of customer feedback from reviews and support tickets: [paste]. Identify the five most recurring themes. For each, estimate how many times it appeared and give one representative quote.

From these themes: [paste], which are things we could realistically address within a month, and which are structural issues (price, category, policy) that need a longer-term decision? Explain the split.

For each of these themes: [paste], tell me whether the tone is mildly annoyed, genuinely angry, or mixed. Volume alone can be misleading, I want to know severity too.

The most frequent fixable complaint is: [state theme]. Give me three concrete, specific ways we could address this within our current setup, ranked by ease of implementation.

Here is feedback from [period A]: [paste] and from [period B]: [paste]. Has the ranking of top themes changed? Did anything we apparently fixed actually reduce complaints about it?

Questions people ask

How much feedback do I need for this to be reliable?

More is better, but even a few dozen entries can reveal a clear top theme; treat results from very small samples as directional rather than definitive.

Should I remove customer names before pasting feedback in?

Yes, strip personal identifying details where you can, especially for anything going into a general-purpose consumer tool rather than an enterprise account.

What if two themes seem to contradict each other?

Ask the model to check whether they come from different customer segments, a contradiction often means two different audiences want different things, not that the data is wrong.