25 AI Prompts for Discovery Calls (Sales)

TL;DR: The most useful AI prompts for discovery calls generate questions, not pitches: pre-call research primers, permission-based openers, problem-surfacing question trees that branch on the answer, impact-quantifying follow-ups, and post-call summaries. Below are 25 prompts across that full arc, plus how to record and take notes without creating a legal or confidentiality problem.

What Are the Best AI Prompts for Discovery Calls?

The best discovery call prompts don't write your talk track. They write the questions that get someone to explain their own problem back to you, and the follow-up that comes next depending on what they actually say. That's the difference this list is built around: 25 prompts across the real arc of a call, from the research you do before you dial in to the internal note you write after you hang up.

Most "AI prompts for sales calls" content quietly turns into a pitch generator: talking points, objection rebuttals, closing scripts. That's not an accident, a pitch is easier to prompt for than a question tree, and it's easier to judge as "done." A discovery call isn't any of those things yet. It's the 20 to 30 minutes where you find out if there's a real problem, who else feels it, what it's costing them, and whether there's an actual process to buy something for it, before you've proposed a single thing. If you're a founder running these calls yourself alongside everything else, our list of 40 AI prompts for startup founders covers the rest of that workload.

The table below shows the shift in what you'd actually type into a generator at each stage.

How Do You Prepare for a Discovery Call With AI?

Pre-call research prompts should turn scattered public signals, the company site, recent hires, a funding announcement, into a short list of hypotheses to test live, not a dossier you read back to the prospect. Use them to walk in with sharper questions, not more facts to recite. The failure mode to watch for is the opposite of under-preparing: showing up having memorized so much that the first ten minutes turn into you narrating their own company back to them instead of asking anything.

1. Company snapshot

Summarize {{prospect_company}} in under 150 words for someone about to run a discovery call
with {{contact_name}}, {{contact_title}}. Cover: what they actually sell (not their tagline),
how big the team looks from public signals, and one thing about their market that would matter
to someone in the {{contact_title}} role. Flag anything you're inferring rather than found stated
directly.

2. Signal-based hypothesis

Based on the following signals about {{prospect_company}}, write 3 hypotheses for why they
might be looking at {{category_or_problem_area}} right now. For each hypothesis, write the one
question I could ask on the call to confirm or kill it.

Signals: {{paste job postings, funding news, leadership changes, recent announcements}}

3. Stakeholder mapping

A {{contact_title}} at a company like {{prospect_company}} (industry: {{industry}}, team size:
{{team_size}}) is joining a discovery call about {{category_or_problem_area}}. Write what this
role likely cares about, what they're likely NOT the right person to answer, and one question
that would reveal which is true in the first five minutes.

This is persona prompting applied to one real stakeholder instead of a hypothetical one. We go deeper on the technique in Persona Prompting: Make ChatGPT Think Like an Expert.

4. Three things I don't know yet

Based on everything I've researched about {{prospect_company}} below, list the 3 most important
things I still don't know that this call needs to answer. Write each as a question, not a
statement to make.

Research so far: {{paste research notes}}

What's a Good AI Prompt for Opening a Discovery Call?

A discovery call opener has one job: get permission to ask real questions, not to demonstrate energy. Both prompts below skip anything that sounds like a cold-open pitch, because the moment a prospect hears a pitch cadence in the first thirty seconds, they mentally file the whole call as a pitch and answer accordingly for the rest of it.

5. Permission-based opener

Write 3 short openers (2 sentences each) for a discovery call with {{contact_name}} at
{{prospect_company}}. Each should state how much time I'm asking for, name one thing I noticed
in my research, and ask what's actually on their agenda for this call rather than assume it
matches mine. No hype language, no "I'm excited to."

6. Rapport tied to a real signal

{{contact_name}} at {{prospect_company}} recently {{specific signal, e.g. "spoke on a panel
about X" or "posted about Y"}}. Write one sentence I could genuinely say to open the call that
references this without flattering them or turning it into small talk that eats the first five
minutes.

What AI Prompts Actually Surface a Prospect's Real Problem?

Problem-discovery prompts should produce open questions and a plan for what to ask next, not a list of things to tell the prospect. The prompt that does the most work here is a question tree: one core question, and a different follow-up depending on which of a few likely answers you get.

7. Broad problem-surfacing opener

Write 4 open-ended questions to start the problem-discovery part of a call about
{{category_or_problem_area}} with someone in the {{contact_title}} role. None should be
answerable with yes or no, and none should mention {{our_product_category}} yet.

8. Implication follow-ups

Neil Rackham's SPIN framework, built from a study of 35,000-plus sales calls at Huthwaite International and published in 1988, found that top performers spent disproportionate time on "implication" questions, ones that ask what a problem causes elsewhere, not just how big it is. This prompt generates that layer:

For each answer type below, write one implication-style follow-up question, the kind that asks
what happens elsewhere in the business because of this problem, not just how big it is.

Problem the prospect described: {{paste their answer here}}

9. The branching question tree

Generate a two-level discovery question tree for the topic below. Start from one open core
question. For the 3 most likely categories of answer, write the exact follow-up question I'd
ask next, and one line on what that follow-up is testing for. If an answer doesn't fit one of
the three categories, tell me to ask you rather than guessing.

Core question topic: how the prospect currently handles {{process_or_task}}
Context: {{one or two sentences on the prospect's company, role, and what I already know}}

Here's what one branch of that tree looks like worked out, so the shape is concrete rather than abstract:

Core question: "Walk me through how your team handles [X] today."

• If they say "we do it manually in spreadsheets" → ask "how many hours a week does that take across the team?" This tests whether the pain is time, not capability.
• If they say "we have a tool but people don't really use it" → ask "what made it stall, the tool itself, the rollout, or something else?" This tests whether a new tool would actually fix anything, or whether the last one died from a change-management problem a new purchase won't solve either.
• If they say "we don't really have a process, it's ad hoc" → ask "what happens when two people handle it differently in the same week?" This tests whether the real cost is inconsistency, not the absence of a system.

Three different answers to the same opening question, three genuinely different follow-ups, and each one is diagnosing a different deal.

10. What have you already tried

Write 3 versions of "what have you tried already" that don't sound like an interrogation,
tailored to a prospect who described this problem: {{paste their description}}. Each should make
it easy for them to admit something half-worked or nothing worked, without feeling judged for
not having solved it already.

11. Status-quo cost bridge

Write one question that gets {{contact_name}} to describe, in their own words, what happens six
months from now if {{problem}} stays exactly as it is today. It should not suggest an answer or
use words like "risk" or "cost" that put the framing in my mouth instead of theirs.

How Do You Get AI to Help Quantify the Cost of the Problem?

A quantifying-impact prompt should get the prospect to say a number, not have you assert one. Chain these as a short sequence and each answer sets up the next question, a version of chain-of-thought applied to a conversation instead of a single completion. The reason this matters more than it sounds: if you say the number first, the prospect either lowballs it to protect themselves or rounds it up because disagreeing with a stranger feels awkward. Either way, the number becomes yours, and yours won't hold up later when a champion has to defend it to their own manager without you in the room.

12. Cost-of-inaction calculation

{{contact_name}} said the following about {{problem}}: {{paste quote}}. Write one follow-up
question that would get them to estimate the cost in a number they say themselves (hours,
dollars, deals lost, whatever fits), rather than a number I supply.

13. Time and hours-lost sequence

Write a short sequence (max 3 questions) that walks from "how often does {{problem}} happen" to
"how long does it take to fix when it does" to "who else has to stop what they're doing to
help." Each question should build on the likely answer to the one before it.

14. Ripple-effect finder

{{contact_name}} in the {{department}} team described {{problem}}. Write 2 questions that find
out which other teams or people are affected by this same problem, without assuming the answer
is "everyone."

15. Benchmark-anchored quantifying

Write one question that asks {{contact_name}} to compare {{problem}} against a time when it
wasn't a problem, whether at this company, a previous job, or a different team, so the size of
the gap comes from their own memory rather than a number I introduce.

What AI Prompts Uncover Budget and Decision Process Without Sounding Pushy?

These questions map the path to a decision, not the amount. Asking "what's your budget" straight out tends to get a defensive non-answer, a number picked to end the question rather than one that reflects what's actually been set aside. Asking about process instead, who signs off, what else competes for the same money, what happens by when, gets you most of the same information without the friction, and it holds up better later when you're trying to remember why a deal stalled.

16. Decision-process mapping

Write 2 questions that map out who else needs to be involved before {{prospect_company}} could
move forward on {{category_or_problem_area}}, phrased so {{contact_name}} explains the process
rather than just naming titles.

17. Budget reality, indirectly

Write one question that surfaces whether {{prospect_company}} has already allocated money
toward solving {{problem}}, without using the word "budget" and without making {{contact_name}}
feel pre-qualified.

18. Timeline-forcing-function

Write one question that finds out what, if anything, is forcing {{prospect_company}} to solve
{{problem}} by a specific date, versus this being a "someday" priority. Make it easy to answer
"nothing, really" without that feeling like a wrong answer.

How Do You Use AI to Surface Objections Before They Kill the Deal?

The goal here is surfacing an objection, not answering one. A prompt that writes rebuttals assumes you already know what's wrong, and most of the time you don't, the objection that actually kills a deal is rarely the one said out loud in the meeting. It's the one a prospect keeps to themselves and cites internally three weeks later as the reason the project quietly died. These prompts are built to get it said in the room, while there's still a chance to address it or, just as usefully, to find out it's a real dealbreaker before you spend another month on the deal.

19. "What would make you say no"

Write one direct, low-pressure question that asks {{contact_name}} what would make them decide
NOT to move forward with something like {{our_product_category}}, framed as helping me not
waste their time rather than as a trap.

20. Silent-objection excavation

{{contact_name}} has gone quiet or noncommittal after I described {{what_we_discussed}}. Write 2
questions that invite them to name a hesitation they haven't said out loud yet, without
pressuring them to justify it on the spot.

21. Past-vendor-failure

{{prospect_company}} previously tried {{a_tool_or_approach_if_known}} for {{problem}} and it
didn't stick. Write one question that finds out whether that failure was about the tool, the
rollout, or the problem being harder than it looked, since each points to a different objection
I'll need to address later.

What AI Prompts Help Design the Next Step, Not Just "Follow Up"?

A next step that only you propose is a next step the prospect can quietly ignore, because they never actually agreed to it, they just didn't object while you were talking. These prompts build one they help design, so the dates and owners came from both sides, not just your calendar.

22. Mutual action plan

Based on this call summary, draft a mutual action plan: 3 to 5 steps, each with an owner (me or
{{contact_name}}'s team) and a proposed date, ending in a clear decision point. Write it as
something I'd ask {{contact_name}} to edit, not something I'd hand them as final.

Call summary: {{paste summary}}

23. Multi-threading

Write one question that finds out who else at {{prospect_company}} should see a recap of this
call before the next step happens, phrased so it doesn't sound like I'm trying to go around
{{contact_name}}.

What Should an AI-Generated Post-Call Summary Actually Include?

A good summary is short enough that a manager or teammate reads the whole thing, and specific enough that it's still useful in a month. Two prompts cover the two audiences: the prospect and your own team.

24. Prospect-facing recap email

Write a recap email to {{contact_name}} after today's call. Cover: the problem in their own
words, what we agreed the next step is, and the date attached to it. Under 150 words. No recap
of things they already know about their own company.

Call notes: {{paste your own notes, no names or figures under NDA}}

The same instinct, get the reader's own words back to them, carries over to follow-up copy generally; see 45 AI Prompts for Email, Ads & Landing Pages That Convert for the broader set.

25. Internal CRM or Slack note

Turn these call notes into a structured internal note for {{deal_stage}} in our CRM: problem (1
to 2 lines), quantified impact if any, decision process and who's involved, objections raised,
next step and date, and a one-line gut read on likelihood to close. Bullet points only, no
narrative.

Call notes: {{paste your own notes}}

Once you've settled on the versions of these 25 that actually fit how you sell, save them as a prompt template instead of retyping the scaffolding before every call. Our guide on building a personal AI prompt library covers the setup, and if you're already reusing one library across investor updates, job posts, and calls, The Founder AI Workflow covers keeping all of it in one place.

Can You Safely Record and Paste Call Notes Into a Public AI Model?

Recording: check the law in both your state and the prospect's before you rely on a verbal heads-up, because roughly a dozen US states require consent from everyone on the call, not just you. Pasting: strip the company name, the deal figures, and anything under NDA before you paste real call notes into a public chat interface, because you don't control what a consumer-facing model retains by default.

On recording, plainly: as of August 2026, eleven US states are commonly cited as requiring all-party consent to record a call, California, Delaware, Florida, Illinois, Maryland, Massachusetts, Montana, Nevada, New Hampshire, Pennsylvania, and Washington. A few more, Connecticut, Michigan, Oregon, and Vermont, show up inconsistently across legal summaries, so don't treat any list, including this one, as the final word. Two people on a call in two different states means two sets of rules can apply at once. If your call tool auto-records or auto-transcribes, that still counts as recording, and a line buried in a terms-of-service page isn't the same as telling the person on the other end. Say it out loud at the start: "This call is recorded for notes, is that alright with you?" Then wait for the actual answer.

On pasting notes into AI: a named prospect, a specific deal size, and anything under NDA are exactly the details that shouldn't go into a public chat window verbatim. Three habits fix most of the risk without slowing you down. Swap the real company name for a placeholder like {{prospect_company}} before you paste notes anywhere, the way every prompt in this piece already does. Strip dollar figures and headcounts that would identify the deal even with the name removed. And check whether your AI tool of choice trains on your inputs by default, many consumer tools do unless you turn it off in settings, versus running under a business or enterprise agreement that contractually excludes your data from training. None of this is a reason to avoid AI for call prep. It's a reason to write your notes the way you would if a competitor could plausibly read them back tomorrow.