How to Use Claude for Customer Research Interviews

Most solo operators skip customer research entirely. I used to be one of them. I’d list a property in Madeira, write what I thought buyers wanted to hear, and wonder why some listings sat for weeks while others moved in days. The problem wasn’t the property. It was that I had no real understanding of what my clients were actually thinking during their decision process.

Customer research interviews fix that. But transcribing them, pulling themes, and turning raw conversation into anything useful? That used to eat my entire Tuesday afternoon. Since early 2026, I’ve been running that whole workflow through Claude — and the difference is significant enough that I’ve documented exactly how I do it, step by step.

This guide covers the exact process I use. Not theory. Not a demo walkthrough. The real method, including where Claude falls short and what to do about it.

Why Claude Handles Research Interviews Better Than Other AI Tools

Before the steps, a quick note on why I landed on Claude for this specifically. I tested ChatGPT-4o and Gemini 1.5 Pro on the same interview transcripts. Claude’s context window — 200,000 tokens on the Pro plan ($20/month) — lets me paste a full 90-minute interview transcript and get analysis in a single session without chunking. That matters enormously when you’re dealing with long, rambling client conversations that jump between topics.

Claude also tends to stay closer to what was actually said. It doesn’t editorialize as aggressively as other models. When I’m trying to understand what a real buyer felt, I need the AI to reflect the transcript back accurately — not give me a polished summary that smooths over the contradictions and hesitations that are often the most useful data points.

Step 1: Prepare Your Interview Questions Using Claude Before You Record Anything

Here’s exactly how I do it: before every client call, I open Claude and describe the specific decision I’m trying to understand. For real estate in Madeira, that might be “a Northern European buyer in their 50s who is considering a second home purchase and has been in the research phase for more than six months.”

Then I ask Claude to generate a 12-question interview guide built around the Jobs-to-be-Done framework. I specify that I want questions focused on the timeline of their decision, the moment they first started looking, who else is involved in the choice, and what would make them walk away from a deal.

The prompt I use looks something like this:

“You are a qualitative researcher specializing in high-consideration purchase decisions. Generate a 12-question customer discovery interview guide for a real estate consultant in Madeira, Portugal. The interviewee is a prospective second-home buyer, likely Northern European, aged 45-65. Use the Jobs-to-be-Done framework. Focus on past behavior and specific moments, not hypothetical preferences. Avoid leading questions.”

Claude returns a structured guide in about 15 seconds. I edit maybe 3 questions. The rest I use almost verbatim. What used to take me 45 minutes of staring at a blank document now takes under 10 minutes total.

Step 2: Record and Transcribe the Interview With a Dedicated Tool

Claude does not record or transcribe audio. That’s a real limitation you need to plan around. You need a separate transcription tool before Claude enters the picture.

I use Otter.ai for English interviews ($16.99/month) and occasionally Whisper via a simple web interface for interviews in Portuguese. Both produce transcripts accurate enough for analysis — not perfect, but close enough. A 60-minute interview gives me roughly 8,000-10,000 words of transcript.

One practical tip: always get a clean export as plain text or .txt. Claude handles plain text better than formatted Word documents when you’re pasting large volumes into the interface.

Step 3: Paste the Full Transcript Into Claude and Run a First-Pass Analysis

This is where Claude earns its place in the workflow. I paste the entire transcript and use this prompt structure:

“Below is a verbatim transcript from a customer research interview with a prospective property buyer in Madeira, Portugal. I am a real estate consultant trying to understand the buyer’s decision timeline, key anxieties, and the language they use to describe what they want. Please do the following: (1) Identify the top 5 themes in their own words, (2) Pull 8-10 direct quotes that reveal emotional or practical concerns, (3) Flag any contradictions or unresolved hesitations in their answers, (4) Note what they did NOT say that I might have expected based on typical buyer behavior.”

That fourth instruction — asking for what they didn’t say — is something I added after my third session with Claude. It’s consistently the most useful output. Buyers often don’t mention renovation costs, local bureaucracy concerns, or resale value when they probably should. Those silences tell you exactly where to focus your follow-up conversation.

Step 4: Extract the Exact Language Buyers Use to Describe Their Problem

This step is specifically for improving your marketing copy and property descriptions. After the first-pass analysis, I run a second prompt:

“From this same transcript, extract every phrase the buyer used to describe: (a) what they want their life to look like in this property, (b) what they’re afraid of getting wrong, and (c) how they described their ideal location or neighborhood. Present these as verbatim phrases, not paraphrases.”

The output from this single prompt has directly improved my listing descriptions. Real buyers don’t say “stunning sea views.” They say things like “I want to be able to have my morning coffee outside without getting windblasted” or “somewhere my kids will actually visit us.” Those phrases, used in property descriptions, perform better than anything I could manufacture.

Step 5: Synthesize Multiple Interviews Into a Buyer Persona Document

After running 4 or more interviews through Claude individually, I move to synthesis. I paste summaries from each interview — not full transcripts at this stage — and ask Claude to identify patterns across all of them.

The prompt:

“Below are analysis summaries from 5 separate customer interviews with prospective second-home buyers in Madeira. Identify: (1) Shared themes that appeared in 3 or more interviews, (2) Significant outliers worth noting, (3) The top 3 objections that appear consistently, (4) The triggering events that initiated the buying process in most cases. Then write a 300-word buyer persona document I can use for marketing and client communication.”

Claude produces a persona document in one pass that I’d normally spend a half-day building. I review it, adjust anything that doesn’t match my lived knowledge of the Madeira market, and use it as the foundation for my listing copy, email sequences, and social content for the next quarter.

Step 6: Turn Research Insights Into Actionable Client Communication Templates

The final step is applying what you’ve learned. I ask Claude to take the buyer persona and the top objections and turn them into specific communication assets.

For example:

“Based on the buyer persona and objections above, write 3 email subject lines and opening paragraphs for a follow-up sequence aimed at buyers who have been in the research phase for more than 6 months but haven’t scheduled a viewing yet. Use the language patterns from the interviews. Keep a warm, consultant tone — not a sales pitch.”

The emails Claude produces from this context are noticeably different from generic AI marketing copy. They reference the real anxieties my actual buyers expressed. That specificity is the whole point of doing the research in the first place.

My Real-World Experience Running This Workflow in Madeira

In January 2026, I had a stretch of three weeks where four separate buyer leads had gone quiet after initial contact. All four were Northern European, all four had expressed serious interest, and all four just… stopped responding. I knew something was off in my messaging, but I couldn’t pinpoint what.

I reached back out personally and offered each of them a 20-minute “feedback call” — framed not as a sales call but as me wanting to understand the market better. Three of the four agreed. I recorded all three with their consent using Otter, ran the transcripts through Claude using the process above, and got back analysis within about 40 minutes of starting.

What Claude surfaced was consistent across all three interviews: every single buyer mentioned concerns about the Golden Visa program changes and what they meant for resale liquidity. I hadn’t been addressing this at all in my follow-up emails. I thought it was a non-issue for buyers who weren’t pursuing residency for tax reasons. I was wrong. These buyers were worried about the resale market being affected even though they personally weren’t using the visa program.

I used Claude to write a short 400-word email addressing that specific concern directly, using language pulled from the transcripts themselves. I sent it to all four original leads — including the one who hadn’t agreed to a call. Two booked viewings within a week. One of those closed two months later, a €485,000 property in Ponta do Sol.

The entire research-to-email process took me about 3 hours across two days. Before Claude, I either wouldn’t have done this analysis at all, or it would have taken me most of a week to pull insights from three transcripts manually and turn them into revised messaging. That’s not an exaggeration — I did try to do this kind of qualitative analysis manually in 2023, and I gave up after two interviews because the pattern-finding was too slow and too subjective when I was doing it alone.

The one genuine limitation I hit during this process: Claude occasionally over-indexes on the most emotionally charged moments in a transcript and underweights quieter, practical concerns. One buyer spent about 8 minutes discussing flight connection logistics from their home city to Madeira — a real and practical concern for second-home owners — and Claude’s summary barely mentioned it because the buyer stated it calmly rather than with obvious emotion. I now specifically prompt Claude to “include practical logistics concerns even if stated in neutral language,” which fixes this about 80% of the time.

Comparing Claude to Other Tools for This Workflow

Tool	Context Window	Cost/Month	Best For This Workflow	Limitation
Claude Pro	200,000 tokens	$20	Full transcript analysis, persona synthesis	No audio input; misses calm-stated concerns
ChatGPT-4o	128,000 tokens	$20	Fast first-pass summaries	More prone to adding assumptions not in transcript
Gemini 1.5 Pro	1M tokens	$19.99	Very long transcripts or batch processing	Quote extraction less precise in my testing
Otter.ai	N/A (transcription only)	$16.99	Recording and transcript generation	No analysis capability; needs Claude downstream

Pro Tips That Changed How I Use This Process

• Always ask Claude what’s missing. The “what didn’t they say” prompt instruction consistently produces insights I wouldn’t have caught manually.
• Run the transcript twice with different prompts. First pass: themes and emotions. Second pass: exact language and objections. You’ll catch different things each time.
• Keep a Claude Project for ongoing research. Claude’s Projects feature lets you store your buyer persona context so it persists across sessions. I update mine every quarter as I run new interviews.
• Add a “surprises” instruction. Ask Claude to list anything in the transcript that surprised it given the typical buyer profile you described. This catches outlier data that generic summaries flatten out.
• Don’t skip the verbatim quotes step. Paraphrased summaries are useful for analysis. Verbatim quotes are what you actually use in copy. They’re different outputs and you need both.

Summary: The 6-Step Claude Customer Research Workflow

1. Use Claude to build your interview guide before the call — save 30-40 minutes per interview.
2. Record and transcribe with Otter.ai or Whisper — Claude needs text input, not audio.
3. Paste the full transcript and run a first-pass analysis covering themes, quotes, and silences.
4. Run a second prompt specifically extracting verbatim buyer language for use in copy.
5. After 4+ interviews, synthesize into a buyer persona document with a single Claude session.
6. Convert the persona and objections into real communication templates — emails, listing copy, social content.

The whole system costs me $20/month for Claude Pro plus $16.99 for Otter. That’s just over $37/month to run a research operation that would otherwise require either a research assistant or, more realistically, just not getting done at all.

If you’re running a solo business and you’ve been guessing at what your clients want, this process will be uncomfortable at first — because you’ll find out you were wrong about several things. That’s the point. I’d rather know.

If you want to see the exact prompt templates I use for each step, I’ve documented them in the Solo AI Kit resource library. Start there, run one interview through the process, and adjust based on what your specific clients actually say.