Most AI prompts for sales ask for an email from thin context, which is how you get polished emails that say nothing. These 13 plays get the deal context straight and diagnose first, then draft.
The best AI prompts for sales run as a sequence instead of one-off requests. Give the model real deal context (call notes, emails, and the business case), have it diagnose what's going on, make the call on what to do next yourself, and only then ask it to draft. The playlist below has 13 plays and 52 prompts to run that way.
Placeholder answer
Short, practical frameworks for executive messaging, business cases, multithreading, and sales enablement.
Ask any AI tool to write a follow-up after a discovery call, and you'll get something like: "Thanks so much for your time today! It was great learning more about your priorities around operational efficiency, and I'm excited to explore how we can support your goals."
Which looks fine. It's polished, it's fast, and it won't move your deal an inch, because the model had nothing to work with but your request, so it filled the gaps with the most average sentence it could find. It can't work out what's going on in a deal from a request that tells it nothing.
The real material from the deal: call notes or a transcript, the email thread, the CRM fields, your business case draft, and anything the buyer has shared with you. In the playlist, that material shows up as labels like @Deal, @Meeting, @Prospect, and @Account Plan.
Those labels don't connect Claude or ChatGPT to your CRM or your call recorder. They're placeholders for context you paste in yourself, which means what comes out is only as good as what you put in. The plays come from Olli, the agent inside Fluint (the company I co-founded), where they're pre-built and run against a team's connected deals. You can also run them yourself by copying the prompts into whatever AI you use and feeding in the context by hand. It works decent, it just takes more effort, and more editing of the outputs.
Then decide what the prompt is for. "Write a follow-up" gets you tone. "Figure out why this deal stalled, so I can choose the next move" gets you a diagnosis, which is what the follow-up needs.
Each play is four prompts, run in order. In most of them, the draft comes after the diagnosis, so by the time you ask for the email, the model already knows the problem and the gaps. (A few plays, like Pre-Call Prep, never draft at all, because their job is research.)
None of the plays stops to make you choose, though. So add that pause yourself: read the diagnosis, decide what's true about the people involved, and then keep going. That's the step a model can't do for you, since you're the one who knows which hypothesis fits the champion who went quiet.
Thirteen plays, in the order a deal moves through them:
Prospecting and early engagement
Discovery and qualification
Building consensus
Demo and evaluation
Business case and mid-funnel
Account management
Here's Stuck Deal Diagnosis, all four prompts, with @Deal standing in for the context you paste in first (the last few emails, your call notes, the stage history, and who's gone quiet):
1. Deal Breakdown. Based on @Deal, what's the current state of this deal and where did it stall? Look at the last few interactions and tell me what might be going on.
2. Root Cause Hypotheses. Give me 3 hypotheses for why this deal is stuck. Be direct: consider whether it's priority, politics, missing stakeholder, weak business case, or something else.
3. Re-Engagement Options. For each hypothesis, what's one move I could make to test it and get the deal moving again? Give me specific actions, not generic advice.
4. Changed Direction. Draft a "changed direction?" message I can send if nothing else works, something respectful that creates urgency without being pushy. The goal is an honest and direct answer.
The first two prompts give you a read on the deal. The third gives you a test for each hypothesis, and that's where your pause goes: run the test that fits what you know about the people involved. (If the stall is about priority, that might mean finding the compelling event you never confirmed.) The message in prompt four is the last resort, for when the tests come back empty.
The parts that depend on knowing the deal. Let AI research the account, but decide for yourself what you believe is going on, and say it in your own words (the point of view guide walks through that). Let it fill in the numbers on a customer's revenue model, but trace the path from your product to their outcome yourself, the way a driver tree makes you.
And mind what you paste in. Use only the deal material your company allows you to share with an AI tool, keep anything confidential out of tools that aren't approved for it, and ask the model to tell you which parts of its answer came from your notes and which parts it inferred.
I am, and I have. Which is exactly why I know the difference between the AI that builds you, and the AI that quietly replaces you.
If you hand it the wheel, and let it draft every email, build every deck, write every recap, and plan every account, it feels incredible for about a quarter. But you've stopped doing the reps yourself, and you slowly become an average of whatever the model spits out. Running plays like these keeps you in the work: the model does the digging and the drafting, and you do the thinking on your live deals.