Templates

AI Customer Research Prompt Pack (Free B2B Framework for Extracting Real Insights)

8 min read · Jun 16, 2026· AO Network Editorial Team

Customer research is the single highest-leverage thing a B2B marketing team does. It is also the activity most teams underinvest in because it takes time to do well.

AI can compress the time without flattening the insights. The five prompts below do specific research work. They are not 'help me think about customers' prompts. They are 'paste 10 transcripts, get specific patterns back' prompts.

Copy any prompt directly from the page using the button in the top right of each card. Fill in the bracketed inputs with your specific context. This is part of the broader AI prompt library and pairs with the ICP and persona worksheet as the destination for what these prompts produce.

What good customer research looks like

Three rules before you run any of these prompts.

Use real transcripts. Recorded calls, written reviews, exported chat logs. Not summaries written by your sales team, not your interpretation. The prompts find patterns; they cannot recover what your interpretation already lost.

Minimum data volume. 5 transcripts for the lightest analysis, 10 for confident pattern detection, 20 if you want to slice by segment.

Verify a sample. Pick three findings and trace them back to the source quotes. AI hallucinates patterns sometimes. Verification catches it. The marketing audit framework is where audit findings get integrated; same discipline applies here.

Prompt 1: Trigger pattern extraction

What changed at the customer's company in the 90 days before they bought? Trigger patterns are the single most actionable input to outbound, targeting, and content.

Prompt

Trigger pattern extraction

Recommended model: Claude Sonnet 4.7

I will paste 10 customer interview transcripts below. Each is a 30 to 45 minute conversation with a recent customer.

Extract trigger patterns: what was happening at each company in the 90 days before they started evaluating solutions like ours?

Return:

1. A ranked list of triggers, with frequency count out of 10 transcripts. Include only triggers that appear in 2 or more transcripts.

2. For each trigger, the exact quote from one transcript that best illustrates it, with transcript number.

3. A 'singleton' section listing triggers that appeared in only one transcript but are unusually specific or actionable.

4. The triggers that surprised you: anything that does not match the assumption that buyers move because of a single named pain point.

Do not invent triggers that are not in the data. Do not generalize across transcripts.

Transcripts:
[PASTE TRANSCRIPT 1]
---
[PASTE TRANSCRIPT 2]
---
[PASTE TRANSCRIPT 3]
---
[PASTE TRANSCRIPT 4]
---
[PASTE TRANSCRIPT 5]
---
[PASTE TRANSCRIPT 6]
---
[PASTE TRANSCRIPT 7]
---
[PASTE TRANSCRIPT 8]
---
[PASTE TRANSCRIPT 9]
---
[PASTE TRANSCRIPT 10]

Prompt 2: Voice-of-customer language mining

Marketers describe products the way they want to. Customers describe problems the way they actually feel them. The gap between those two languages is where messaging fails.

Prompt

Voice-of-customer language mining

Recommended model: Claude Sonnet 4.7 or GPT-5

I will paste 10 first-call sales transcripts below. These are conversations with prospects in the first 15 minutes of the sales process.

Extract:

1. The 10 most-used phrases that describe the prospect's pain, in their exact words. Rank by frequency. Show the phrase and the count of how many transcripts it appears in.

2. The 5 most common ways prospects describe what they wish they had. Their exact words.

3. The 5 most common ways prospects describe the current alternative they are using (spreadsheets, a competitor, internal tooling, doing nothing).

4. Three specific phrases that appeared multiple times that should appear verbatim in marketing copy.

5. Words and phrases that marketers love but customers never use. Surface anything that appears in our messaging but never in the transcripts.

Do not paraphrase. Quote exactly. Cite transcript numbers.

Our current messaging (for the contrast in item 5):
[PASTE 2 to 3 PARAGRAPHS OF YOUR CURRENT WEBSITE/EMAIL COPY]

Transcripts:
[PASTE 10 TRANSCRIPTS, EACH SEPARATED BY ---]

Prompt 3: Buying committee mapping

B2B buying decisions involve multiple roles. Marketing usually addresses one. The other roles silently kill deals.

Prompt

Buying committee mapping

Recommended model: Claude Sonnet 4.7

I will paste 8 customer interview transcripts below. Each interview asked about the buying process explicitly.

Map the typical buying committee.

Return:

1. The roles involved in the buying decision, ranked by how often each appeared. For each role:
   - Title and seniority level
   - Their concern in the buying process (in their language or the buyer's language about them)
   - Their typical entry point to the deal (when they enter the process)
   - Their veto power: would they kill a deal, slow it down, or just need to be informed

2. The 3 most common patterns of how the buying committee assembles. Specific. For example: 'CMO finds it, brings in VP of Marketing, RevOps validates, CFO approves' is one pattern.

3. The most underestimated role: who showed up in the data more than most marketing programs would assume.

4. Specific language each role used about the others. How does the CMO describe their CFO when explaining the buying process?

Do not invent roles. Stick to what appears in the transcripts.

Transcripts:
[PASTE 8 TRANSCRIPTS, SEPARATED BY ---]

Prompt 4: Watering-hole discovery

Where customers actually go for information is rarely where marketers think they go. This prompt surfaces the gap.

Prompt

Watering-hole discovery

Recommended model: Claude Sonnet 4.7 or GPT-5

I will paste 10 customer interview transcripts and 10 sales call transcripts below.

Extract every information source customers reference. Be exhaustive.

Return:

1. Newsletters mentioned by name, with frequency count.
2. Podcasts mentioned by name, with frequency count.
3. LinkedIn voices (specific people) they follow, with frequency count.
4. Communities, Slack groups, or Discord servers, with frequency count.
5. Conferences and events attended, with frequency count.
6. Specific authors, books, or publications cited.
7. Tools or platforms they use to discover content (e.g., 'I save things to Pocket', 'I get all my news from LinkedIn').

For each source, the exact phrasing of how the customer mentioned it and the transcript number.

Then, the 3 most surprising findings: sources that appeared multiple times but you would not expect for this audience.

Customer interview transcripts:
[PASTE 10, SEPARATED BY ---]

Sales call transcripts:
[PASTE 10, SEPARATED BY ---]

Prompt 5: Switcher story extraction

Customers who switched FROM a competitor TO you carry the highest-value signal. They know both products. They can articulate the difference.

Prompt

Switcher story extraction

Recommended model: Claude Sonnet 4.7

I will paste 6 interview transcripts with customers who switched from a competitor to us.

For each transcript, return:

1. The competitor they switched from (name, if mentioned).
2. The specific moment they decided to look for alternatives (in their language).
3. The 2 to 3 reasons they cited for switching, in their exact words.
4. What they expected our product to do that the competitor did not.
5. What they did not expect about our product, positive or negative.
6. The specific phrase they would use to recommend us over the competitor to a peer.

Then, across all 6 transcripts:

1. The 3 most common reasons for switching, ranked by frequency.
2. The 2 patterns in 'unexpected positive' that surprised you.
3. The competitor weaknesses these switchers exposed that we could lean into in our positioning.
4. The specific phrases we should use verbatim in competitor-comparison pages and switcher content.

Do not invent. Quote exactly. Cite transcript numbers.

Transcripts:
[PASTE 6 TRANSCRIPTS, SEPARATED BY ---]

How to gather the input data

Three sources, in order of value.

Customer interviews. The gold standard. 30 to 45 minutes recorded. Use a tool like Riverside or just a Zoom recording. Transcribe with Otter, Descript, or the best analytics tools stack.

Sales call recordings. Already exist in Gong, Chorus, or your call recorder. Marketing rarely listens to these. Listen quarterly. Or run AI prompts on them.

Reviews and post-purchase surveys. G2, Capterra, TrustRadius. NPS responses with comments. Onboarding survey free-text. Lower signal per item but high volume.

What these prompts do not solve

The strategic interpretation. AI returns patterns. You decide what they mean and what to do about them. The patterns alone are not the strategy.

The recruiting. Getting 10 customers to spend 45 minutes with you is the hard part. The analysis is the easy part. Customer success leadership is usually the right partner for recruiting.

The follow-up validation. AI surfaces hypotheses. Real conversations confirm or refute them. Treat AI output as inputs to the next round of interviews, not as final findings.

Operating cadence

Run prompt 1 (triggers) and prompt 2 (voice-of-customer) every quarter. They are the highest-frequency need.

Run prompt 3 (buying committee) once a year. The committee structure does not change fast.

Run prompt 4 (watering holes) every six months. Audiences shift their media habits.

Run prompt 5 (switcher stories) opportunistically. Every time a switcher closes, schedule the interview.

Frequently asked questions

How long should each customer transcript be?

30 to 45 minutes of conversation, which produces 5,000 to 8,000 words. Long enough to surface patterns, short enough to fit in modern AI context windows.

Can I run these prompts on sales call recordings instead of dedicated interviews?

Yes for prompts 2 (voice-of-customer) and partially for prompt 1 (triggers). Less so for prompts 3, 4, 5 which need explicit questions that sales calls do not include. Dedicated research interviews complement sales call analysis.

How do these connect to the ICP and persona worksheet?

Directly. Prompt 1 feeds the trigger field. Prompt 2 feeds the language used throughout. Prompt 3 expands a single persona into a buying committee. Prompt 4 fills in the watering holes field. The worksheet without the prompts requires you to do the synthesis manually.

Which prompt above are you most likely to skip because you have not yet done the work to record the conversations? That is the one to invest the recording infrastructure for.

The Always-On Brief

Weekly strategy, tool picks, and playbooks. 6,000+ marketers subscribed.

Templates

AI Content Brief Template (Free Framework for Briefing AI Tools to Match Your Brand Voice)

Jun 15, 2026

Templates

AI Prompt Library for B2B Marketing Teams (Free + Copy-Paste Ready)

Jun 14, 2026

Templates

Always-On Marketing KPI Dashboard Template (Free + How to Use It)

Jun 14, 2026