Why this is hard to get right
The Discovery Call That Almost Cost a Deal
Marcus is a mid-market AE at a workflow automation company. He carries a quota of $1.2M and runs 8–10 discovery calls a week. He's a solid closer — but his pipeline forecasting is consistently off, and his manager keeps flagging that his deal notes lack decision-maker clarity and urgency signals.
Marcus knows his product cold. What he struggles with is structuring discovery so that every call surfaces the same critical data: economic impact, the real decision process, competitive considerations, and a defined next step. When calls go well, it's because the prospect is talkative. When they don't, he leaves with vague notes like "they're interested" and "evaluating a few options."
He tried using ChatGPT to generate a discovery script. His prompt was: "Give me discovery call questions for my SaaS product." The output was a generic list of 20 open-ended questions — some useful, most obvious. None were sequenced to a 30-minute agenda. None accounted for his buyer's specific role (Director of Operations), the industry context (manufacturing), or the fact that his prospects are almost always replacing spreadsheets, not a competing platform.
The AI didn't know any of that. So it guessed. And the script it produced felt like something from a generic sales training slideshow.
Marcus spent 45 minutes editing it — cutting irrelevant questions, adding industry-specific probes, rearranging the sequence to match his call flow. By the time he was done, he wasn't sure the result was better than what he'd written from memory.
The real problem wasn't the AI. It was the prompt.
Discovery is not a generic exercise. It's a structured conversation calibrated to a specific buyer persona, a specific business context, and a specific sales methodology. A Director of Operations at a 300-person manufacturing company has completely different pains, language, and decision dynamics than a VP of Engineering at a 50-person SaaS startup. The questions that build trust with one persona can alienate the other.
When Marcus rebuilt his prompt with full context — his ICP, the buyer's current tools, the call length, his methodology (MEDDICC), and the output format he needed — the AI produced a script with time-boxed sections, persona-specific pain probes, red flag signals, and a closing CTA sequence. He ran it on his next call and got to decision criteria and budget in under 20 minutes — something that usually took him two calls.
A structured discovery prompt doesn't just save prep time. It enforces consistency. It ensures every rep on your team is asking the right questions in the right order, collecting the same qualification data, and leaving every call with a concrete next step. That's the difference between a pipeline that forecasts accurately and one that doesn't.
Common mistakes to avoid
Omitting Buyer Role and Company Size
Asking for a 'discovery script for SaaS' without specifying whether the buyer is a Director of Operations at a 300-person company or a CTO at a 20-person startup forces the AI to average across all personas. The result is generic language that builds no rapport with your actual buyer. Always include title, seniority, and company size range.
Skipping the Current-State Context
Discovery questions land differently when the prospect is replacing spreadsheets versus replacing a competitor's platform. Without specifying current tools and workflows, the AI can't generate credible follow-up questions or ROI anchors. Include what your buyers use today and what they're trying to move away from.
Not Specifying a Sales Methodology
Prompts that don't reference a qualification framework (MEDDICC, BANT, SPIN, Challenger) produce unfocused output. The AI doesn't know which fields you need to qualify — so it guesses. Naming your methodology maps the output directly to your CRM fields and forecast criteria.
Requesting Questions Without a Time-Boxed Structure
A list of 20 discovery questions is not a script. Without specifying call length and section timing (e.g., 5 minutes on current process, 10 minutes on pain and impact), the AI produces a question bank you can't actually run live. Always define your agenda structure with explicit time allocations.
Ignoring Output Format Requirements
Asking for a 'discovery script' without specifying format — bullet questions, follow-up probes, red flags, time checks — produces a wall of text that's hard to reference during a live call. Define the output format explicitly so the script is scannable in the moment, not just readable in preparation.
Forgetting to Define the Call Goal
Discovery calls serve different purposes: initial qualification, deepening a stalled deal, or expansion into a new department. Without stating what you need to leave the call knowing, the AI generates a generic question set that tries to do everything and masters nothing.
The transformation
Give me discovery call questions for my SaaS product.
You are a senior SaaS AE. Create a 30-minute discovery call script for a Director of Operations at a 200–500 employee manufacturing company evaluating workflow automation. 1) Context: Mid-market, replacing spreadsheets; current tools: email + Excel. 2) Goals: Uncover pains, impact, decision process, budget, timeline. 3) Tone: Consultative, concise, empathetic. Avoid jargon. 4) Structure: Opening (2 min), qualification (5), pain/impact (10), current process (5), buying process (5), next steps (3). 5) Output: Bullet questions, suggested follow-ups, red flags to probe, time checks. End with a summary and a clear 2-step CTA.
Why this works
Role Clarity Removes Guesswork
The After Prompt opens with 'You are a senior SaaS AE,' immediately calibrating the AI's voice, assumptions, and vocabulary. This single line prevents generic sales-training language and produces output that matches how an experienced AE actually thinks and speaks in a discovery conversation.
Buyer Context Anchors the Questions
Specifying 'Director of Operations at a 200–500 employee manufacturing company evaluating workflow automation' and 'current tools: email + Excel' gives the AI a concrete persona to write against. Questions generated from this context reference real workflows, credible pains, and language that resonates with an operations buyer — not a generic decision-maker.
Time-Boxed Structure Enforces Pacing
The After Prompt breaks the call into six named sections with explicit minute allocations — 'Opening (2 min), qualification (5), pain/impact (10)…' This forces the AI to sequence questions strategically and ensures the script is actually usable live, not just readable in prep.
Methodology Alignment Drives Qualification
The After Prompt's goal list — 'Uncover pains, impact, decision process, budget, timeline' — maps directly to qualification frameworks like MEDDICC. This ensures every question serves a forecasting purpose and produces notes that translate cleanly into CRM fields without additional interpretation.
Defined Output Format Makes It Actionable
Requesting 'bullet questions, suggested follow-ups, red flags to probe, time checks' and 'a summary and a clear 2-step CTA' transforms the script from reading material into a live-call tool. The AI structures output for how AEs actually use scripts — scanning, not reading — during a real conversation.
The framework behind the prompt
The Theory Behind Effective Discovery Prompts
Discovery calls sit at the intersection of behavioral psychology, structured interviewing, and sales methodology. Understanding the research behind them explains why prompt structure matters so much.
The cognitive load problem. Research on working memory (Miller's Law, 1956) shows that humans can hold roughly 7 items in short-term memory. A discovery call forces an AE to simultaneously listen, process, respond empathetically, track qualification criteria, and plan the next question — all while managing time. A well-structured script functions as a cognitive offload tool, freeing mental bandwidth for active listening rather than question recall.
Consultative selling frameworks. Neil Rackham's SPIN Selling research — based on analysis of 35,000 sales calls across 23 countries — found that top performers asked significantly more Implication and Need-Payoff questions than average performers. Most reps over-index on Situation questions (what do you have now?) and under-invest in Implication questions (what happens to the business if this isn't fixed?). A structured prompt can enforce this balance by allocating time and question depth by category.
MEDDICC and its variants (MEDDPICC, MEDDIC) emerged from PTC's enterprise sales methodology in the 1990s and remain the dominant qualification framework in B2B SaaS. The framework's value is its field completeness — it forces reps to collect the same data on every deal, making pipeline forecasting more accurate and coaching more targeted.
The Challenger Sale's insight on commercial teaching suggests that the best discovery calls don't just collect information — they reframe the buyer's understanding of their own problem. This means the best discovery questions aren't just diagnostic; they introduce a perspective the buyer hadn't considered. Prompts that include 'reframe' or 'insight' instructions produce scripts closer to this model.
Structured interviewing research from organizational psychology (Schmidt & Hunter, 1998) shows that structured interviews — with consistent, pre-planned questions — produce dramatically more reliable and valid assessments than unstructured conversations. The same principle applies to discovery: consistency drives better qualification data, which drives better forecasting accuracy.
Prompt variations
You are a B2B SaaS founder running early-stage sales calls without a formal sales team.
Create a 25-minute discovery call script for a VP of Marketing at a Series A e-commerce company with 30–80 employees. They found us through a LinkedIn post about content attribution. They currently use Google Analytics and a basic email platform.
Goals: Validate whether content attribution is a top-three priority, uncover budget ownership, identify the internal champion, and determine if there is an active buying cycle.
Methodology: Loosely follow SPIN (Situation, Problem, Implication, Need-Payoff).
Tone: Peer-level, curious, low-pressure. Avoid any enterprise sales language.
Structure: Warm open (2 min), situation and current stack (6), problem and implication (10), fit assessment (5), next step (2).
Output: Numbered questions per section, 2–3 follow-up probes per question, a red-flag checklist, and a recommended close that does not require an immediate commitment.
You are a Senior Customer Success Manager at a B2B project management platform.
Create a 20-minute expansion discovery script for a Director of Engineering at an existing customer account. They have 80 seats in one department. The account is healthy (NPS 8, active usage), but leadership wants to identify expansion into the product team, which currently uses a competing tool.
Goals: Understand the product team's current workflow pain, assess openness to consolidation, identify the internal sponsor for expansion, and surface any risk signals in the existing account.
Tone: Consultative, relationship-first. Reference shared history with the account. Do not pitch. Ask, listen, reflect.
Structure: Relationship check-in (3 min), current state of the product team (7), pain and consolidation interest (6), next step (4).
Output: Bullet questions, transition phrases between sections, expansion risk signals to watch for, and a non-pushy next-step suggestion tied to value already delivered.
You are an SDR at a mid-market HR technology company.
Create a 15-minute qualification call script for an HR Manager at a professional services firm with 150–400 employees. They responded to a cold email about automating onboarding workflows. They likely use a legacy HRIS and manual document processes.
Methodology: BANT — confirm Budget authority, identify Authority (decision-maker), validate Need, and establish Timeline.
Goals: Determine if the prospect is worth passing to an AE. Disqualify early if no budget ownership, no active need, or a timeline beyond 12 months.
Tone: Efficient, friendly, and direct. Respect the prospect's time. Do not over-explain the product.
Structure: Brief intro and permission (2 min), situational context (3), BANT qualification (8), handoff or disqualify decision (2).
Output: Question-by-question script with decision branches — if yes, go here; if no, go here. Include 3 polite disqualification exits and a handoff summary template for AE notes.
You are an Enterprise AE at a cybersecurity SaaS company.
Create a 45-minute multi-stakeholder discovery script for a joint call with a CISO and VP of IT at a financial services company with 1,000–5,000 employees. They are evaluating endpoint detection vendors after a near-miss security incident. Current stack: legacy antivirus and a SIEM they describe as 'underpowered.'
Methodology: MEDDICC — cover Metrics, Economic Buyer, Decision Criteria, Decision Process, Identify Pain, Champion, Competition.
Goals: Quantify the business risk of the current gap, map the full buying committee, understand evaluation criteria and weighting, identify the internal champion, and surface competitive consideration.
Tone: Authoritative, peer-level with technical stakeholders. Use security-specific language where appropriate. Do not oversimplify.
Structure: Agenda alignment (3 min), business and risk context (10), technical current state (10), decision process and criteria (12), competitive landscape (5), next steps (5).
Output: Section-by-section questions with stakeholder-specific variants (CISO vs. VP IT), red flags, a competitive intelligence probe sequence, and a multi-step next-step CTA that advances both technical and executive tracks.
When to use this prompt
Marketing Managers
Build persona-aligned discovery scripts for campaign handoff demos to ensure SDRs qualify leads consistently.
Product Managers
Validate user pains and workflows during problem interviews with structured, non-leading questions.
Sales Leaders
Standardize discovery across the team with role-specific scripts that map to MEDDICC or your methodology.
Customer Success Managers
Run expansion or renewal discovery to uncover new use cases, risk signals, and buying dynamics.
Founders
Lead early-stage discovery calls that surface critical product gaps and buying triggers quickly.
Pro tips
- 1
Customize by specifying your sales methodology (e.g., MEDDICC) so the script maps to your qualification fields.
- 2
Anchor to the buyer’s current tools and processes to generate credible follow-up questions and ROI angles.
- 3
Set strict time boxes to prevent rabbit holes and ensure you cover decision process and next steps.
- 4
Define red flags to probe (e.g., no executive sponsor, competing priorities) to improve forecasting accuracy.
MEDDICC is one of the most widely adopted enterprise qualification frameworks — and it maps cleanly onto a structured discovery prompt.
Here's how each MEDDICC element translates into a prompt instruction:
- Metrics: 'Include 2–3 questions that quantify the business impact of the current pain (e.g., hours lost per week, error rate, revenue affected).'
- Economic Buyer: 'Include questions that identify who owns the budget and what approval process they follow.'
- Decision Criteria: 'Ask what the prospect will use to evaluate options and how they weight each criterion.'
- Decision Process: 'Map the internal steps from evaluation to signature, including procurement, legal, and IT involvement.'
- Identify Pain: 'Surface the primary pain, its root cause, and what happens if it goes unresolved for another 12 months.'
- Champion: 'Identify who internally is motivated to drive this project forward and why.'
- Competition: 'Uncover what else they are evaluating or have evaluated previously.'
When you build these directly into your prompt as named goals, the AI structures questions to fill each field — making your post-call CRM update a matter of copy-and-paste rather than interpretation. Sales leaders can use this approach to enforce methodology consistency across the team without relying on manual coaching.
Most discovery scripts assume a linear call. Real calls rarely are. Here are three advanced prompt techniques that produce more flexible, battle-tested scripts:
1. Branching Logic Instructions Add: 'For each qualification question, provide a branch: if the answer confirms fit, proceed to X; if it signals a gap, include a polite probe and a graceful exit.' This turns a flat question list into a decision tree you can navigate live.
2. Stakeholder Variant Instructions When you know multiple stakeholders will attend, add: 'Provide a variant of each section's questions for technical buyers (focused on integration, risk, process) and business buyers (focused on ROI, timeline, strategic priority).' This prevents you from asking a CFO about API architecture or an engineer about payback period.
3. Late-Stage Re-Discovery Prompts For deals that have stalled, generate a separate script: 'Create a 20-minute re-engagement discovery call for an opportunity that went dark after initial evaluation. Focus on what changed internally, whether the priority shifted, and whether a new champion has emerged.' This is a distinct call type from initial discovery and deserves its own prompt architecture.
The most common objection to scripted discovery is that reps sound robotic. Here's how to use an AI-generated script as a thinking tool, not a reading prompt:
Internalize the structure, not the words. Before the call, review the six sections and the 2–3 priority questions per section. Know the flow by heart. Use the script as a backup reference, not a teleprompter.
Customize the opening. The first 90 seconds should always reference something specific to this prospect — a recent company announcement, a mutual connection, or their LinkedIn activity. No script can do this for you. Add a blank 'personalization hook' line at the top of every script as a reminder.
Use the follow-up probes when you need them. The AI-generated follow-ups ('Can you tell me more about that?' or 'What does that cost you per quarter?') are valuable when a prospect gives a short answer. Keep them visible but don't force them if the conversation is already flowing.
Treat red flags as mental checkboxes. Review the red flag list before the call. You don't need to ask about each one directly — often you'll hear signals organically. Your job is to recognize them when they appear, not to interrogate the prospect.
When not to use this prompt
When This Prompt Pattern Is Not the Right Tool
Not every sales conversation benefits from a tightly structured AI-generated discovery script. Here's when to step back:
When you're in a highly relational, long-cycle enterprise sale. If you've been building a relationship with a prospect for months, arriving with a scripted question flow can feel transactional. In these cases, a lighter prompt — one that generates 5 strategic questions rather than a full script — is more appropriate.
When the call is primarily a demo or a product presentation. Discovery and demo are distinct call types. Using a discovery script on a call the prospect expects to be a demo creates friction and erodes trust. Generate the right script for the right call type.
When you lack basic ICP data. If you genuinely don't know the buyer's role, company size, or context, a generic prompt will produce a generic script. In that case, spend 10 minutes on LinkedIn research before prompting — the output quality depends entirely on the input quality.
When your team is still learning the product. New reps sometimes use AI scripts as a substitute for product knowledge. That leads to awkward moments when a prospect's answer demands a product-informed follow-up. Scripts work best when the rep can improvise within the structure — not rely on it entirely.
Troubleshooting
The script feels too long and covers too many topics for a 30-minute call
Add an explicit constraint: 'Limit the total script to no more than 15 primary questions across all sections. Prioritize questions by must-know (qualification and pain) versus nice-to-know (process detail).' Also instruct the AI to mark each question as 'essential' or 'optional' so you can cut on the fly during a call that runs short.
The AI generates generic questions that don't reflect our specific product or industry
Add a product context block to your prompt: 'Our product automates [specific workflow]. Our buyers typically struggle with [specific pain 1] and [specific pain 2]. Include questions that probe these specific areas and reference the workflows our product replaces.' The more you describe your product's actual use case, the more the AI can write questions that reflect your buyer's real experience.
The closing and next-step section is weak — just 'schedule a follow-up'
Specify your closing mechanics directly: 'The next-step CTA must include two options for the prospect to choose from (e.g., a technical demo or a business case review), a suggested timeline, and language that ties the next step to a pain or goal the prospect stated during the call.' This forces the AI to write a purposeful close, not a generic 'let's find time.'
The tone is too formal and sales-y — my buyers respond to a more peer-level conversation
Add explicit negative constraints to your tone instruction: 'Avoid any language that sounds like a sales pitch, a product feature description, or a scripted transition. Write as if a trusted advisor is asking questions to understand a problem, not to qualify a deal.' You can also provide a single example sentence in your preferred tone and ask the AI to match it throughout the script.
The script doesn't account for multi-stakeholder dynamics — I often have 2-3 people on the call
Add a stakeholder layer to your prompt: 'Assume 2–3 stakeholders are present: a business buyer focused on ROI and a technical buyer focused on integration and risk. For each section, provide a primary question and a stakeholder-specific variant that addresses each audience.' This gives you the flexibility to direct questions at the right person without losing the thread of the conversation.
How to measure success
How to Evaluate the Quality of Your AI-Generated Discovery Script
Before running the script on a live call, check it against these signals:
Structure and pacing:
- Does the script fit the stated call length without cramming?
- Are sections time-boxed with explicit minute allocations?
- Does the sequence move logically from context to pain to process to next step?
Persona fit:
- Do the questions use language your specific buyer would recognize?
- Are references to current tools, workflows, and industry context present?
- Would a prospect in this role find these questions credible and relevant?
Qualification completeness:
- Does the script surface all required MEDDICC (or chosen methodology) fields?
- Are implication and impact questions included — not just situational ones?
- Is there a clear mechanism to identify the economic buyer and decision process?
Actionability:
- Can a rep scan and use this during a live call, not just read it in prep?
- Does the closing section include a specific, two-option next-step CTA?
- Are red flags identified with follow-up probes attached?
A strong script requires minimal editing before use. If you're spending more than 10 minutes revising, the prompt needs more context — not the script.
Now try it on something of your own
Reading about the framework is one thing. Watching it sharpen your own prompt is another — takes 90 seconds, no signup.
Turn your ICP details into a structured discovery script tailored to your buyer, methodology, and call length.
Try one of these
Frequently asked questions
The more specific, the better — but you need at minimum: buyer title, company size range, and industry. If you also know their current tools and what they're trying to replace or fix, include that. A prompt with 'Director of Operations, 200–500 employees, manufacturing, replacing Excel' will produce a far more targeted script than one with 'operations leader at a mid-market company.'
Yes — and you should. Name your methodology explicitly in the prompt: MEDDICC, BANT, SPIN, or Challenger. Then list the specific fields you need to populate (e.g., 'Economic Buyer, Decision Criteria, Identified Pain'). The AI will structure questions to collect that exact data, making it much easier to update your CRM directly after the call.
Add a tone directive that matches the buyer's context. For enterprise: 'Authoritative, peer-level, reference industry benchmarks.' For SMB: 'Friendly, direct, avoid corporate language, keep it conversational.' Also adjust the depth of decision-process questions — enterprise calls need multi-stakeholder mapping; SMB calls often involve a single decision-maker with a faster cycle.
Specify a maximum question count per section in your prompt (e.g., 'No more than 3 primary questions per section, plus 2 follow-up probes'). Also instruct the AI to prioritize: 'If time is limited, which 5 questions must be answered before the call ends?' This produces a tiered script — essential questions first, depth questions if time allows.
Absolutely. Replace the sales role with 'product manager' or 'UX researcher' and adjust the goal from 'qualify a deal' to 'validate a hypothesis' or 'map a workflow.' The structural logic — persona, context, time box, output format — transfers directly to internal interviews, user research sessions, and executive alignment calls.
Add a specific instruction like: 'Include 4–6 red flags that suggest a poor fit or a stalled deal, with a follow-up probe for each.' Common red flags to flag: no executive sponsor, vague timeline, multiple competing priorities, procurement-led process, or a prior failed implementation. Give the AI examples from your experience and it will add them to the script.
Use a persona-based template as your base — one script per major ICP segment. Regenerate when the buyer role, industry, or call goal changes meaningfully. For minor tweaks (adjusting tone, swapping out a section), edit the prompt parameters rather than starting from scratch. Most teams maintain 3–5 core scripts covering their primary ICPs.
Explicitly instruct: 'For the qualification section, use closed diagnostic questions to confirm or disqualify fit quickly, then open-ended questions for sections focused on pain and impact.' You can also ask for branching logic — 'If the answer to X is no, provide a polite disqualification exit and end the call.' This gives you a decision-tree structure, not just a question list.