Most discovery calls fail before they start
Not because of bad questions.
Because of bad prep.
Most sellers walk into discovery expecting the conversation to reveal what matters.
That approach is backwards.
By the time the call starts, you should already have a working point of view.
Not a pitch. A hypothesis.
Lately, I’ve been using a custom GPT to prepare for discovery calls before every first meeting. Not to replace judgment. To force clearer thinking earlier.
This issue walks through exactly how to set one up yourself and how to use it in a practical, repeatable way.
No theory.
No BS.
Just mechanics.
Step 1: Decide what your discovery GPT is responsible for
Before you build anything, get specific about the job.
Your discovery GPT is not:
A note taker
A script writer
A demo assistant
Its job is to pressure-test your thinking before you speak to a buyer.
That means its responsibilities should be narrow and opinionated.
At a minimum, it should:
Summarize a company in plain language
Identify likely reasons a buyer would take a meeting now
Surface risks, objections, and disqualifiers
Call out weak or unproven assumptions
Generate a short list of discovery questions that actually matter
If it starts doing more than that, it usually gets worse, not better.
Step 2: Create a Custom GPT and set its posture
In ChatGPT, go to Explore GPTs → Create.
You’ll see two main sections:
Instructions
Knowledge
Start with Instructions.
Most people focus on tasks. That’s a mistake.
You should define posture first.
Your instructions should make it clear that this GPT:
Is direct
Is skeptical
Prioritizes clarity over politeness
Cares more about disqualification than optimism
If the output feels agreeable, you’ve already lost value.
This should feel like a senior seller reviewing your prep, not an assistant trying to be helpful.
Step 3: Give the GPT your baseline context
Before you use the GPT for any discovery prep, you need to tell it a few things about how you sell.
Not every time.
Once.
This is what prevents generic output.
In the Instructions section, write a short block that covers:
What company you sell for
Who you usually sell to
What problems you are typically brought in to solve
What a strong discovery call looks like in your world
What usually disqualifies a deal early
This does not need to be long or polished.
Think of it as a standing assumption set.
For example, you might describe:
Your typical buyer role and level
The deal sizes you care about
Whether you prioritize speed, depth, or risk elimination
The kinds of conversations that usually lead nowhere
The goal is simple.
When you ask the GPT to prepare you for a call, it should already understand:
What “good” looks like
What you tend to overestimate
What you need pushed on, not validated
Without this step, every output will feel slightly off.
With it, the GPT can actually challenge your thinking instead of repeating it.
You can update this block anytime, but you shouldn’t need to touch it often.
Step 4: Upload internal materials that reflect reality
Next, move to the Knowledge section.
This is where you give the GPT context about your business.
Upload documents that represent how your company actually sells:
Product one-pagers
Pitch decks
Internal positioning docs
Battlecards
Case studies
You are not training it to pitch.
You are giving it awareness of:
What you sell
Who it’s for
Where deals usually go sideways
Important rule:
If a document is pure marketing language, don’t upload it.
You want signal, not slogans.
Step 5: Teach it how to do account research
Your GPT should be explicitly instructed to analyze accounts, not just summarize them.
When you give it a company name and website, it should be able to:
Explain how the company likely makes money
Identify who their customers are
Call out operational complexity
Flag scale, growth, or margin pressure
You can improve the output by prompting it to look for:
Recent press or announcements
Funding or expansion signals
Regulatory or compliance exposure
Signs of internal strain or change
If information is inferred, it should say so clearly.
Ambiguity is fine. Unstated assumptions are not.
Step 6: Map the likely buying structure
This is one of the most useful parts of discovery prep.
For any call, your GPT should help you think through:
Who is likely involved beyond the person you’re meeting
Who owns budget versus execution
Where decisions usually stall in similar companies
You do this by providing:
The role you’re meeting with
Company size
Deal type and motion
Then explicitly asking:
“If this progresses, who else will need to care and why?”
This helps you avoid running discovery with someone who can’t move the deal forward.
Step 7: Use it to spot recent change
Good discovery is anchored in change.
Your GPT should be instructed to look for signals like:
New leadership
Role changes
Hiring spikes
Reorganizations
New initiatives or mandates
You can support this by pasting:
Job descriptions
LinkedIn snippets
Press headlines
Then ask a simple question:
“What likely changed that made this conversation relevant now?”
If there’s no clear answer, that’s information too.
Step 8: Generate fewer, better discovery questions
Do not ask your GPT for a long list of discovery questions.
Ask it for:
Questions that change the direction of the call
Questions that surface ownership and urgency
Questions that expose disqualifiers early
Then force it to rank them.
When you review the output:
Delete at least half
Rewrite the rest in your own words
Decide which answers would cause you to stop pursuing the deal
If a question doesn’t help you make a decision, it doesn’t belong.
Step 9: Keep yourself in control
The rule I follow is simple:
The GPT does the prep.
I run the conversation.
I don’t read output on calls.
I don’t follow it word-for-word.
I use it to:
Enter the call with conviction
Ask better first questions
Avoid defaulting to generic discovery
The confidence comes from preparation, not automation.
Final thought
If your discovery prep feels easy, it’s probably shallow.
A good discovery GPT should:
Make weak assumptions obvious
Create tension around unclear value
Force clarity early
That’s the point.
More to come next week in Selling with AI.
If you want the version I actually use
If you followed the steps above, you can build this yourself.
It will work.
It will also take time to get right.
I’ve already done that work.
For paid subscribers, I share:
The exact base instructions I use
The full discovery prep prompt
The structure I use for account research
The way I force assumption testing
The output format I rely on before every first call
You can copy it, adjust it to your business, and use it immediately.
If you prefer building from scratch, you don’t need this.
If you want to skip iteration and start from something battle-tested, it’s there.
Unlock the Execution Layer
You are already inside the system. The Execution Layer unlocks the prompts, checklists, and operating playbooks used to deploy real revenue motions.
Get Vault AccessVault access includes::
- Copy and paste execution prompt packs
- Deal, outbound, and follow-up playbooks
- Operating checklists for every motion
- Members Vault access