Using AI feedback summaries

Reading every response stops being realistic somewhere around the fiftieth. Summaries mean you don’t have to, without losing what the responses actually said.

What a summary gives you

For a body of feedback — a survey’s responses, a period of reviews — you get: the main themes, what’s praised, what’s criticised, notable changes since last time, and a suggested next step.

The suggestion is the part worth paying attention to. “Sentiment is 72% positive” is a fact. “Three of the last ten negative responses mention the booking confirmation email being unclear” is something you can fix on Tuesday.

Themes

Themes are topics appearing repeatedly across responses, grouped by whether they’re praised or criticised.

Two things make them useful:

They cross the individual/pattern line. One person mentioning parking is noise. Nine people mentioning parking is a decision about signage.

They find things you’re blind to. Everyone who works there knows where the toilets are, knows the wifi password, knows which door is the entrance. Customers don’t, and they’ll tell you — but only in ones and twos across months of feedback, which is exactly the pattern a human reading responses individually will never assemble.

Reading a summary well

Check the volume behind it. A summary of eight responses is a summary of eight opinions. The themes aren’t stable yet.

Look at what changed. More useful than the current state. A theme that’s new this month is more urgent than one that’s been steady all year.

Don’t stop at the summary for anything serious. If a theme involves safety, staff conduct, or a legal issue, read the raw responses. Summaries compress, and the specifics matter for those.

When to read the originals

• Anything involving a complaint you’re going to act on formally
• Any theme you’re about to spend real money fixing
• Any feedback from a customer you’re about to contact personally
• When the summary says something surprising — verify before acting

Asking follow-up questions

For anything the summary doesn’t cover, you can query your feedback directly rather than reading through it.

→ Asking the AI Assistant