Finding the patterns hiding in your reviews and callbacks
A shop owner reads reviews one at a time and reacts to whichever one stung most this week, so the same recurring complaint — techs running late, confusing invoices, a specific install that keeps generating callbacks — never gets named as a pattern. AI is genuinely useful at grouping a pile of feedback into themes. It's a summarizer here, not a judge: it tells you what people said, and you decide what to fix.
You are analyzing customer feedback for the owner of an HVAC company. I'll paste a batch of reviews, survey replies, and callback notes. Summarize the themes — do not evaluate whether we're a good company or invent explanations. Feedback (names and addresses removed): {{feedback_batch}} Time period: {{time_period}} Produce: 1. The top 3-5 recurring themes, each with how many mentions and one representative (paraphrased) quote. 2. A split of what customers praised versus what frustrated them. 3. Anything that looks like a repeat operational issue (scheduling, communication, pricing clarity, callback on a specific job type). 4. Two or three concrete questions I should ask my team based on what's here. Hard rules: - Summarize ONLY what's in the text I pasted. Do not invent complaints, causes, or fixes, and do not guess at a diagnosis for any technical issue mentioned. - If there isn't enough feedback to call something a pattern, say so instead of stretching. - Neutral and factual — you're organizing what customers said, not grading my business.
Fill in your details and the prompt updates live — then copy.
Top themes (40 items): 1. Punctuality — 11 mentions, mostly positive: "tech texted and showed up right on time." 2. Invoice clarity — 6 mentions, mixed: several said the final bill had line items they didn't expect. 3. Callbacks on ductless installs — 4 mentions: two customers needed a return visit within a week of a mini-split install. Praised: friendliness, on-time arrival, clean work areas. Frustrated: surprise charges, phone hold times. Questions for the team: Are ductless installs getting a final commissioning check? Is the estimate clear about what's not included? Not enough data yet to judge weekend response times.
The full workflow
- Export or copy reviews and survey replies, and strip names and addresses first.
- Run the prompt and sanity-check the counts against the source.
- Treat the themes as leads to investigate, not verdicts.
- Take the two or three questions to your next team huddle.
Watch out for
The model can miscount or over-generalize from a handful of reviews — verify the tallies and don't act on a 'pattern' that's really two complaints.
De-identify the feedback before pasting; a batch of reviews still contains names and job details that don't belong in a consumer AI tool.
If a review mentions a technical failure, AI's guess at the cause is worthless — reopen the job and inspect, since callback diagnosis is licensed work.
Where this comes from
Every use case on this site is grounded in real reports from working hvac technicians — not invented by us.