How do you use Claude to synthesise customer interviews?

Quick Answer: Paste your customer interview transcripts into Claude with a structured prompt, and it extracts themes, Jobs-to-be-Done, and a usable quote bank in minutes. The process works for 1 transcript or 50, and produces output you can take directly into a PRD, positioning doc, or sales deck.

You finished 8 customer interviews. You have 6 hours of recordings, rough notes, and a growing sense of dread about turning all of it into something your team can actually use.

That synthesis bottleneck is where most product and research work stalls. The interviews were good. The insights are in there somewhere. But extracting them, organising them, and presenting them in a format that drives decisions takes longer than the interviews themselves.

Claude cuts that process down significantly. This guide shows you exactly how to use Claude for customer interview synthesis: what to paste in, what prompts to use, and what outputs to expect.

What Is Customer Interview Synthesis (and Why It Breaks Down)?

Customer interview synthesis is the process of moving from raw transcript data to structured insight. Done well, it produces three things your team needs: recurring themes across customers, JTBD (Jobs-to-be-Done) statements that explain what customers are trying to accomplish, and a quote bank of verbatim evidence you can use in decks, docs, and pitches.

Done manually, synthesis takes 2-4 hours per interview. At 8 interviews, that is a full work week before you have written a single recommendation.

The breakdown happens because synthesis requires holding multiple conversations in your head simultaneously, spotting patterns across them, and writing up findings in a format that non-researchers can act on. That is exactly the kind of multi-document pattern recognition Claude handles well.

What You Need Before You Start

Before opening Claude, get these three things in order:

• Clean transcripts. Auto-generated transcripts from tools like Otter.ai, Fireflies, or Zoom work fine. Remove filler words if possible, but it is not required. What matters is that each speaker turn is clearly labelled.
• Anonymised data. Strip customer names, company names, and any identifying details before pasting into Claude. Replace them with codes: Customer A, Customer B, or Segment 1, Segment 2.
• A clear research objective. Know what question you are trying to answer before you synthesise. "What is stopping mid-market buyers from expanding their seats?" is a better anchor than "learn about customers."

Step 1: Set Up Claude with a Synthesis Brief

Do not paste transcripts cold. Open with a brief that tells Claude what role to play, what you are looking for, and what format to return.

Here is a prompt template that works:

You are a B2B SaaS product researcher. I am going to give you [X] customer interview 
transcripts. Your job is to synthesise them and return:

1. The top 5-7 recurring themes, each with a 1-2 sentence summary
2. 3-5 Jobs-to-be-Done statements in the format: 
   "When [situation], I want to [motivation], so I can [outcome]"
3. A quote bank: 8-12 verbatim quotes organised by theme

My research objective is: [paste your objective here]

Here are the transcripts. I will paste them one at a time and say "NEXT" 
between each one. When I say "SYNTHESISE", analyse all of them together 
and return the output above.

Ready?

This primes Claude before any data arrives. It knows what output format to produce, so you are not cleaning up a wall of unstructured text at the end.

Step 2: Paste Transcripts Using a Consistent Format

For a small batch (1-5 transcripts), paste everything in a single message after your brief. For larger batches (6-50+), use the "NEXT" approach from the prompt above to feed them sequentially.

Label each transcript clearly:

--- TRANSCRIPT: Customer A | Segment: Mid-Market | Role: Head of Ops ---
[paste transcript here]

NEXT

The label matters. When Claude surfaces themes and quotes, you want to know which customer segment they came from. That segmentation becomes useful fast when you are comparing enterprise buyers to SMB buyers or champions to economic buyers.

For 50+ transcripts, the Claude Code approach works better than the standard interface. If you are also thinking about how AI-native search and answer engines change content workflows, SaaS teams often pair this kind of research workflow with broader GEO and AEO support for B2B SaaS. You can write a simple script that loops through transcript files and feeds them to the API with a consistent system prompt. The output quality is comparable; the manual effort drops to near zero.

Step 3: Run the Synthesis

Once all transcripts are in, send one word: SYNTHESISE.

Claude will return a structured output across your three requested formats. Here is what good output looks like for each:

Themes

Each theme should have a name, a 1-2 sentence description, and a frequency indicator (how many customers mentioned it). For example:

Theme: Reporting credibility gap Mid-market ops leads trust their own data pulls more than automated reports. They describe spending 30-60 minutes per week manually verifying numbers before sharing them upward. Mentioned by: Customers A, C, D, F

JTBD Statements

These should be grounded in specific language from the transcripts, not paraphrased into generic product-speak. A strong JTBD statement from Claude will often include a phrase a customer actually used. For example:

"When I am preparing the Monday morning ops review, I want to know the numbers are right without checking them myself, so I can present with confidence instead of caveats."

If Claude returns something generic ("When I need data, I want it to be accurate"), push back with: Make these more specific. Use language from the transcripts where possible.

Quote Bank

The quote bank is the most immediately useful output. Verbatim quotes, organised by theme, give you evidence for every claim you make in a positioning doc, a PRD, or a sales deck. This is especially useful if your team is already investing in B2B SaaS content marketing and needs direct customer language to strengthen messaging. Ask Claude to flag the 3-4 strongest quotes per theme and note which customer segment each came from.

Step 4: Push Deeper with Follow-Up Prompts

The first synthesis pass is the floor, not the ceiling. Use follow-up prompts to extract more specific insight.

For tension and contradiction:

Are there any themes where different customer segments gave conflicting answers? 
Describe the tension and which segments held each view.

For priority signals:

Based on the emotional intensity of the language used, which 3 themes seem 
to cause the most frustration or urgency for customers?

For positioning language:

Pull out any phrases customers used to describe their ideal outcome. 
These should sound like something a customer said, not a product feature.

For sales enablement:

Write 3 objection-handling responses based on concerns that came up 
across multiple interviews. Use customer language in the responses.

Each follow-up takes under 30 seconds and produces output that would take a researcher an hour to write from scratch.

Step 5: Validate and Package the Output

Claude's synthesis is a strong first draft, not a finished deliverable. Before sharing it with your team, do three things:

1. Spot-check quotes. Find 3-4 quotes in the output and verify them against the original transcripts. Claude is accurate, but verbatim accuracy matters when quotes go into external documents.
2. Sense-check the themes. Do the themes match what you heard in the room? If a theme feels off, go back to the transcripts and probe it with a targeted prompt.
3. Add your judgement. Claude tells you what customers said. You decide what it means for your roadmap, your positioning, or your next sprint. That layer of interpretation is yours.

Package the final output as a research summary document with four sections: objective, methodology (number of interviews, segments covered), key themes with quotes, and JTBD statements. That format is readable by a founder, a PM, or a sales lead without any translation.

How Many Transcripts Can Claude Handle at Once?

Claude's context window (200K tokens on Claude 3.5 and above) fits roughly 20-30 average-length interview transcripts in a single session. For 50+ transcripts, either use Claude Code with the API, or run synthesis in batches of 15-20 and then run a second-pass synthesis across the batch outputs.

The batch approach produces cleaner themes because each batch forces Claude to identify patterns within a smaller set before you look for patterns across sets.

What Makes Claude Better for This Than Other AI Tools?

Claude handles long, nuanced text better than most alternatives. Its instruction-following is precise, which matters when you need output in a specific format. It also handles ambiguity in customer language well, distinguishing between a customer venting frustration and a customer describing a genuine workflow problem.

For B2B SaaS specifically, the JTBD output quality is noticeably stronger than generic summarisation tools because Claude can hold the "situation, motivation, outcome" frame consistently across multiple documents. If your next step is turning those insights into pipeline growth, it can also help to compare this workflow with support from B2B SaaS digital marketing agencies or specialist B2B SaaS SEO agencies.

FAQs

What is the best way to use Claude for customer interview synthesis? The most effective approach is to prime Claude with a structured brief before pasting any transcripts, specify the three output formats you need (themes, JTBD statements, quote bank), then use follow-up prompts to push deeper into specific areas. This produces usable output in 10-15 minutes for a batch of 8-10 interviews.

How many interview transcripts can Claude process at once? Claude 3.5 and above supports a 200K token context window, which fits approximately 20-30 average-length interview transcripts in a single session. For larger batches, run synthesis in groups of 15-20 and then synthesise across the batch outputs in a second pass.

Is it safe to paste customer interview transcripts into Claude? Anonymise transcripts before pasting them. Replace customer names, company names, and any identifying details with codes (Customer A, Company X) before the data goes into any AI tool. Check your organisation's data handling policy and, if relevant, your customer agreements before processing interview data externally.

Can Claude replace a professional UX researcher for synthesis? Claude accelerates the mechanical parts of synthesis: pattern recognition, quote extraction, and formatting. It does not replace the researcher's judgement about what themes mean strategically, which insights to act on, or how to present findings to different stakeholders. Use it to eliminate the time-consuming parts so you can spend more time on interpretation.

What format should interview transcripts be in for Claude to process them? Plain text works best. Speaker labels (Interviewer / Customer A) on each turn help Claude distinguish between questions and responses. You do not need clean formatting, but removing timestamps and filler words improves output quality slightly.