28 APR 2026 · 9 MIN READ
Ten customer-interview promptsI run before every call.
Most founders walk into customer calls under-prepared and walk out with vague feedback. Here is the exact prompt sequence I run, on every call, in under 12 minutes.
Most founders walk into customer calls under-prepared. They have a vague hypothesis, a Notion doc with three bullet points, and a hope that the customer will say something useful. They walk out with feedback that is either too polite to be true or too narrow to be actionable.
It is not a calls problem. It is a preparation problem. And it is the easiest thing in the world to fix with AI, if you have a sequence to run.
This is the sequence I run before every customer call I take. It takes about 12 minutes. The prompts are deliberately short — these are not "prompt engineering" exhibits, they are working tools. Use the names of real people, real companies, real problems. Paste your actual context.
Before the call (8 minutes, three prompts)
Prompt 1 — The 90-second brief. This is the only prompt I would die on a hill for. It runs first, every time.
Brief me on [first name] [last name], [their role] at [company]. I have a 30-minute call with them tomorrow about [topic]. In 200 words: what should I know about how they make decisions, what their company is currently struggling with, and what's the most likely real reason they took this call.
The trick here is the last clause — "the most likely real reason." Customers rarely tell you why they are actually on the call. They tell you the surface reason ("we're evaluating tools") rather than the underlying reason ("my boss is pressuring me to make a decision by Friday and I need someone to sanity-check my shortlist"). Asking the AI to guess the second one primes you to listen for it.
Prompt 2 — Question generation. I ask for too many questions on purpose, then cut.
Generate 25 questions I could ask [name] in this call. Group them into: (1) state of the world today, (2) what they've already tried, (3) what would make this a great year for them, (4) the trap they're walking into. Make question 25 the bravest one I'm too polite to ask.
Always read question 25 first. It is usually the question worth asking. The other 24 are warm-up.
Prompt 3 — The decision-tree pre-mortem. Before the call, I want to know what each kind of answer means.
If [name] tells me [hypothesis A] in this call, what does that mean for our roadmap? If they tell me [hypothesis B], what does that mean? What would I be wrong to conclude from each?
This stops me from over-indexing on a single line of feedback. By the time I'm in the call, I already know which kinds of answers update my model and which don't.
During the call (zero prompts)
You don't run prompts during the call. You listen. The prep work has done its job — your questions are sharper, your ear is tuned for the second-order signal, and you know what each kind of answer means before you hear it.
What I do bring into the call: the answer to prompt 1 on a Post-it next to my laptop. Three sentences. That's it.
Immediately after the call (4 minutes, four prompts)
This is where most founders lose the value. The call ends, you have ten minutes before your next thing, and the lessons evaporate. The next four prompts run in the cool-down.
Prompt 4 — The brain-dump capture.
I just got off a call with [name]. Here are my raw notes [paste]. Pull out: the three most useful things they said, the one thing that contradicts what I assumed going in, and the one thing they hinted at but didn't say outright.
The third clause is the one most founders skip and the one that pays the highest dividends. People rarely state their objections directly. They hint at them, expecting you to ask the follow-up. The AI is excellent at spotting the hint.
Prompt 5 — The proposal seed.
Based on this call, draft the first 100 words of a proposal email to [name]. Lead with the specific phrase they used to describe their problem. End with a concrete next step.
You will rewrite this. The point is to capture their language while it is fresh. Two days later you will not remember the exact phrasing they used, and the exact phrasing is what makes proposals land.
Prompt 6 — The follow-up question generator.
What three questions should I have asked but didn't? Format as: question, why it matters, when I should ask it (next call, follow-up email, never).
The "never" option is important. Some questions look smart on paper but would actually erode trust. Letting the AI flag those saves you from the over-clever follow-up email.
Prompt 7 — The synthesis log.
Add this call to a running synthesis. Across the [N] customer calls I've now done on [topic], what's the strongest pattern? What's the weakest assumption I'm still holding?
For this one, paste the AI's output from previous calls into the same chat thread (or your prompt-library doc). The pattern emerges around call 6 or 7. Founders who skip this step keep "discovering" the same insight on every call without realising they have already heard it three times.
The three prompts I save for after the day is done
The last three prompts run at the end of the day, ideally with a glass of wine and your laptop closed for everything else.
Prompt 8 — The honesty test.
Reading my notes from today's [N] customer calls, where am I lying to myself? Be direct.
This one will sting. That is the point. The AI does not have to maintain your relationship with yourself. It will tell you the thing you already half-suspected and were trying to talk yourself out of.
Prompt 9 — The non-buyer prompt.
Based on these calls, who is NOT a customer? Who would I be wasting time on? What signals did I miss that would have told me earlier?
Founders spend an enormous amount of energy chasing buyers who were never going to buy. This prompt builds the negative profile — the person who looks like a customer in the first 5 minutes and is not by the end. Save the output. Read it before your next batch of calls.
Prompt 10 — The roadmap nudge.
Given today's calls, what one thing should I move up on the roadmap, and what one thing should I move down? Don't equivocate. Pick one of each.
Customers will tell you 30 things. The job of the founder is to pick which one matters. The "don't equivocate, pick one" framing forces the AI to stop being diplomatic. You can argue with the answer. You cannot argue with a fence-sit.
The most common mistake
The mistake I see founders make with sequences like this is using them once, finding them useful, and never running them again. Prompt sequences are like exercise routines — the value comes from the cadence, not the individual session. Run this every time you take a customer call for two months and the cumulative effect is staggering. Run it once and you'll have a slightly better Tuesday.
Three things this sequence will not fix
The sequence above is a multiplier, not a magic wand. Three honest caveats:
1. It cannot replace doing the call. If you have not run thirty customer calls yourself, no amount of preparation will give you the instincts that come from being in the room. Run the calls. Use the sequence to make the next call sharper, not to skip the boring middle.
2. It will not work if your context is thin. The prompt is only as good as the context you feed it. "Brief me on John from Acme" returns nothing useful. "Brief me on John, head of growth at Acme, ARR ~£3M, just hired their first SDR, considering switching from HubSpot, took the call after I sent them our pricing benchmark" returns gold.
3. The synthesis only works if you actually stop and read it. The fastest way for this sequence to become useless is to run it, save the output to Notion, and never look at it again. Block 10 minutes after the call to read prompt 6 and prompt 7. That's the loop.
Where the rest of these live
If you want this sequence ready-made — plus the variants for sales calls, hiring calls and partner calls — that's the customer-research chapter coming in Bible 04 (AI Agency Accelerator), drafting now for Q3 2026. Subscribers get an early read.
If you want to start using prompts like these today: grab the free chapter. It's a full chapter from Bible 02 (Digital Publishing Empire), covers the AI Newsletter Business model end-to-end, and the customer-research workflow inside it ports cleanly to any one-person business.
Either way: the next call you take, run prompts 1–3 first. Twenty minutes. You'll feel the difference before you ask question one.