How to analyse customer interviews and reviews with AI to sharpen your messaging
- Level: Beginner
- A few hours to set up
The short answer
Gather interview transcripts, reviews and call notes, remove personal details, and ask a model to tag each passage by pain, desired outcome, objection and exact phrase. Check the themes against the quotes, then rewrite your headline and proof points in the words customers actually use.
On this page
At a glance
- The marketing problem
- Messaging written from inside the company tends to describe features. Customers describe problems and outcomes in their own terms. Reading dozens of transcripts by hand is slow, so the analysis is often skipped.
- Who this is for
- Product marketers, content leads and founders rewriting positioning, landing pages or ads.
What you need
Before you start
- Permission to use the recordings, reviews and notes for analysis
- A way to remove names, emails and company details before anything goes to a model
- An AI tool your company allows for customer data, with its data-retention settings checked
What goes in
- Interview or sales-call transcripts
- Public reviews (for example G2, Capterra, Trustpilot or app stores)
- Support tickets or free-text survey answers
- Your current headline and value proposition, for comparison
How it works
- 01 · ToolCollect and anonymise the sources
- 02 · AI stepTag each passage
- 03 · AI stepGroup the tags into themes
- 04 · You decideCheck the themes against the quotes
- 05 · OutputRewrite the message in customer language
- Tool
- AI step
- You decide
- Output
Step by step
01 Tool
Collect and anonymise the sources
Put transcripts, reviews and notes in one spreadsheet, one passage per row, with an ID, the source and the date. Remove names, emails and company names before using a model.
02 AI step
Tag each passage
Ask the model to tag every passage with the same fields and to quote rather than paraphrase. Work in batches if the file is long.
Prompt to copy
You are analysing customer research for [product]. For each passage below, return a table row with: passage ID, pain (what is going wrong), desired outcome, objection or hesitation, trigger (what made them look for a solution) and the most striking exact phrase. Quote exact words only. If a field is not present, leave it empty. Do not summarise or paraphrase the customer’s words. Passages: [paste rows with IDs]
03 AI step
Group the tags into themes
Ask for a small number of themes, each supported by several passages. A theme without quotes behind it is an opinion, not a finding.
Prompt to copy
Here are tagged customer passages. Group them into no more than eight themes. For each theme give: a name in the customer’s own words, the passage IDs that support it, and three exact quotes. List passages that fit no theme separately. Do not create a theme supported by fewer than three passages. Tagged passages: [paste]
04 You decide
Check the themes against the quotes
Open the quoted passages for each theme. Merge, split or delete themes that do not hold up, then choose the two or three that matter most for the next piece of messaging.
05 Output
Rewrite the message in customer language
Draft headline and proof-point options from the exact phrases, keep the quote behind each line, and test against the current version before replacing it.
Prompt to copy
Using only these themes and quotes, write five headline options and three proof points for [page or ad]. Each line must reuse a phrase from the quotes. Show the quote next to each line. Do not add claims that are not in the quotes. Themes and quotes: [paste]
Where a person decides
A marketer checks each theme against the original quotes and decides which phrases change the messaging.
What to watch for
- A model can invent a neat theme the quotes do not support; always read the quotes.
- Reviews over-represent very happy and very unhappy customers.
- Counts from a small sample show direction, not proportion.
How to tell it is working
- Every theme lists at least three quotes you can find in the original text.
- Two people tagging the same passages reach similar tags.
- The rewritten message is tested against the current one on a page or an ad.
An illustrative example
Input
Illustrative: anonymised interview transcripts and public reviews for a fictional staff-scheduling tool.
Output
A table of themes, such as “I spend Sunday night building the rota”, with the quotes behind each one, and three headline options written in that language.
Want this running in your team?
AI marketing advisory. An experienced marketing partner for the decisions your team needs to make.
Related workflows
How to research a prospect with AI before a sales call
Turn a new enquiry into a sourced one-page brief with questions for the first call, checked by a person before sales sees it.
How to do competitor analysis with AI
Compare competitors’ positioning, offers and proof side by side with one repeatable prompt, and check every claim against the source page.
How to train AI to write in your brand voice
Turn your best writing into a voice guide that ChatGPT, Claude or Gemini can follow, and test it before the team relies on it.