AI Prompts for Discovery Calls: Pre-Call Research, Scoring, and Recap

Most agency sales teams are running discovery calls with some pretty poor uses of AI tools. Many don’t use an AI pre-call prep tools or they use a generic "summarize this prospect" prompt.

The agencies that use AI well in discovery calls aren't using better tools. They're using better prompts. Four specific prompts cover the four moments where AI moves the conversation: pre-call research, in-call question generation, post-call scoring, and recap email drafting. Each prompt has the same shape (role, context, structured output) and produces output the team can deploy without rewriting.

Why Generic AI Prompts Fail at Sales Call Prep, Scoring & FollowUp

Three failure patterns repeat across every agency I've seen try to add AI to their sales process:

1. No role definition: "Summarize this prospect for me" produces an executive summary that reads well and contains no actionable signal. Without a role, the model defaults to generalist analyst tone, which is the wrong tone for sales prep.
2. No structured output: Free-form prose is where models ramble and reliability dies. Without an explicit output structure, the output is hard to scan and impossible to standardize across reps.
3. No rubric: "Score this prospect" without scoring criteria produces a number in the 7-to-9 range with three reasons that sound smart and predict nothing. The rubric is what makes the score correlate with actual outcomes.

The four prompts below fix all three. Each has an explicit role, a structured output format, and (where applicable) a rubric the model has to apply.

Prompt 1: Pre-Call Research

This AI prompt takes raw inputs (company name, person name, links to LinkedIn or articles) and produces a structured brief.

You are a senior sales researcher at a B2B agency. Your job is to
produce a focused pre-call research brief for the rep who will run
discovery in the next 24 hours.

The brief should be skimmable in 90 seconds and produce three
actionable outputs: a working hypothesis about the prospect's
problem, three signal-based questions the rep can ask, and one
specific recall trigger the rep can mention to prove they did the
research.

# Inputs
- Company name: [insert]
- Prospect name and role: [insert]
- Company website URL: [insert]
- Prospect LinkedIn URL: [insert]
- Any other relevant links: [insert]
- Our agency's focus: [insert one sentence about who we are and
  what problem we solve]

# Research depth
Limit research to:
- The company's recent news and press releases
- The prospect's role tenure, prior roles, and any public content
  (posts, podcasts, talks)
- Job posts from the company in the last 90 days
- Visible product/service offering and customer base
- Competitive context only as needed to understand positioning

Do not include:
- The company's full leadership team
- Historical M&A activity older than 18 months
- Investor information beyond the most recent funding event

# Output format

Return a markdown document with these sections:

## Company snapshot
- One sentence: what they actually sell
- One sentence: company stage and trajectory
- One sentence: visible signals from the last 90 days

## Person snapshot
- One sentence: tenure and background
- One sentence: priorities implied by recent role and content
- One sentence: what pain is implied by role plus company stage

## Working hypothesis
A specific, falsifiable claim about what the prospect is actually
trying to fix right now. Lead with "I think they're trying to fix
X because of Y." If evidence is thin, say so.

## Three signal-based questions
Specific questions tied to actual evidence found in research. Avoid
generic discovery questions.

## One recall trigger
A specific reference (from their LinkedIn post, their podcast
appearance, their job description, their press release) that the
rep can mention naturally in the first five minutes of the call to
prove they did the work.

## Missing information
List any inputs the rep should gather before the call if possible.

The prompt takes about 90 seconds to run with Claude or GPT and produces output the rep can paste into their notes app. The 90-second runtime is the part that matters. If pre-call research takes 15 minutes per call, the agency stops doing it. At 90 seconds, the agency does it for every call.

Prompt 2: In-Call Question Generation

Use this if a discovery call drifts into territory the rep wasn't prepared for. Open the prompt in another tab during the call and feed it the pivot.

You are a senior sales coach. The rep is in the middle of a
discovery call and the conversation has pivoted to a topic they
didn't anticipate. Generate 3 specific follow-up questions the rep
can ask to dig deeper on the pivot.

# Context
- Prospect company: [paste from pre-call brief]
- Prospect role: [paste from pre-call brief]
- Original working hypothesis: [paste from pre-call brief]
- The pivot: [the rep types one sentence describing what the prospect
  just said that changed the direction of the call]

# Output

Return exactly 3 questions in this format:

1. [Question] (Why: [one-line rationale for what this question
   reveals])
2. [Question] (Why: ...)
3. [Question] (Why: ...)

Each question should:
- Build on the pivot, not redirect away from it
- Surface motivation or constraint, not just facts
- Be open enough to invite a real answer (not yes/no)
- Be short enough to remember (under 20 words)

This prompt runs in 10-15 seconds and gives the rep three usable questions to deploy in the next minute of the call. It's the AI version of having a senior salesperson whispering in your ear.

Prompt 3: Post-Call Scoring

Use this within 30 minutes of the call. The rep dictates or types a brief recap of what happened and the model scores it against the pitch scorecard.

You are a sales performance analyst. The rep just finished a
discovery call and is logging the recap. Score the call against
the agency pitch scorecard.

# The scorecard

Score each dimension 1-5:

1. Problem framing (did the rep frame the prospect's problem more
   clearly than the prospect framed it themselves?)
2. Fit articulation (did the prospect leave understanding why this
   agency specifically is the right team?)
3. Risk handling (did the rep surface and reduce the prospect's
   risk concerns inside the call?)
4. Urgency creation (did the prospect leave with a sense that
   acting now is better than acting later?)
5. Follow-through plan (did the call end with a specific,
   agreed-upon next step?)

# Inputs
- Prospect company: [insert]
- Prospect role: [insert]
- Call recap from rep: [the rep types or pastes 2-3 paragraphs
  describing what happened in the call, what the prospect said,
  what the rep said, and any specific moments worth flagging]

# Output

Return JSON in this exact shape:

{
  "scores": {
    "problem_framing": <1-5>,
    "fit_articulation": <1-5>,
    "risk_handling": <1-5>,
    "urgency_creation": <1-5>,
    "follow_through_plan": <1-5>
  },
  "total": <5-25>,
  "strongest_dimension": "<which dimension scored highest, with
    one-sentence reason>",
  "weakest_dimension": "<which dimension scored lowest, with
    one-sentence reason>",
  "specific_risk_for_this_deal": "<one specific risk this call
    surfaced that should be addressed in follow-up>",
  "recommended_next_move": "<one-sentence specific action the rep
    should take in the next 48 hours>"
}

The output is structured enough to drop straight into the CRM. Over time, the scores accumulate and reveal patterns (the team is consistently weak on risk handling, for example) that point to systemic process fixes.

Prompt 4: Recap Email Generation

Use this within 60 minutes of the call. The model drafts the follow-up email based on the call recap and the scorecard output.

You are a senior account executive. The rep just finished a
discovery call and needs a recap email drafted that they will edit
and send within the next hour.

The recap email should:
- Confirm the prospect's stated problem in their language
- Restate the agreed-upon next step with specific dates
- Surface (not hide) the most important risk that came up
- Avoid generic agency language ("partnership," "white-glove,"
  "strategic")
- Run under 200 words

# Inputs
- Prospect name: [insert]
- Prospect company: [insert]
- Call recap: [paste the recap the rep wrote for the scoring prompt]
- Agreed next step from the call: [insert specifically what was
  agreed and by when]
- Top risk to address: [insert the risk-handling note from the
  scorecard output]

# Output

Return a complete email in plain text:

Subject: [subject line, 50 chars or less, specific to the
conversation]

[Email body, under 200 words, written in the rep's voice (assume
mid-level senior tone, direct, no jargon)]

[Standard signature placeholder]

Avoid:
- Generic openers ("Great talking with you today")
- Agency jargon ("partnership," "white-glove," "strategic")
- Promises beyond what was discussed in the call
- More than one CTA in the email body

The output is a draft, not a final email. The rep edits in their voice and sends. Drafting time drops from 15 minutes per recap email to 90 seconds plus a quick edit.

How to Chain the Prompts

The four prompts work better as a system than as individual prompts. The chain looks like this:

Before the call: Run prompt 1 (pre-call research) on the prospect. The rep walks in with a working hypothesis and three signal-based questions.

During the call: Keep prompt 2 (in-call question generation) in a browser tab. If the call pivots to unexpected territory, the rep types the pivot and gets three follow-up questions in 15 seconds.

Within 30 minutes after the call: Run prompt 3 (post-call scoring). The rep types a 3-paragraph recap and gets a structured scorecard plus a recommended next move.

Within 60 minutes after the call: Run prompt 4 (recap email). The output from prompt 3 feeds the input for prompt 4. The rep edits the draft and sends.

The total AI runtime across the four prompts is about 5 minutes per discovery call. The total rep time saved is about 45-60 minutes per call (mostly in pre-call prep and post-call recap). The math gets dramatic across a team running 20+ discovery calls per week.

What To Do With This

If you're already using AI in your sales process, audit your current prompts against these four. Most teams have prompt 4 (recap drafting) and skip the other three. The other three produce more leverage.

If you're not using AI in sales yet, start with prompt 1 (pre-call research). It produces the most immediate value with the lowest risk. Once the team trusts the output, layer in the other three.

If you'd rather have this already done for you, check out, Discovery Lab & Discovery Lab Pro - two pre-discovery call AI-powered prep tools built specifically for agencies. (And you can feed your discovery call transcript right into Call Lab or Call Lab Pro & get ai-powered sales call coaching & follow up suggestions based on the WTF Sales Method that is at the heart of SalesOS.)