How do you prep a sales call in three minutes with AI?
Paste your CRM record, the notes from your last call and the latest email thread into one prompt, and ask for four things back: the goal of this call, three questions, the risks in the deal and the next step you will ask for. That takes about two minutes; spend the third checking names, numbers and the one thing the AI guessed. The AI assembles, you decide: a brief you did not check is a brief you will quote wrong on the call.
How we research this: facts come from each vendor’s public pages and named studies, dated where shown; the math is ours and labeled. Vendor prices and limits are as of October 9, 2026.
Updated
Why three minutes, and not thirty?
Reps blow prep two ways (our view, not a study): walking in cold and re-asking what the buyer already told them, or burning half an hour on a dossier and still having no plan. A booked call already has history. Use it.
Never heard of you? That is research, not prep; see how much time to research a prospect.
Our guidance: three minutes for a follow-up call with history, up to ten for a first demo with a new economic buyer.
Why a goal and a dated ask, every time: a call with no target drifts into small talk, and small talk with no date on it is how deals go quiet (our view).
What goes into the prompt?
| Input | Where it lives | What to copy (30-60 seconds) |
|---|---|---|
| Deal record | CRM opportunity | Stage, deal value, expected close, the next-step field, every contact with title |
| Last conversation | Call recorder summary or your notes | What they said the problem is, numbers they gave, who they named, what they promised |
| Last email thread | Inbox | The last 3-5 messages, especially anything they asked for |
| Optional: one news item | Their site, your alert | Only if dated in the last 30 days and you open the link |
Keep out (our rule): pricing exceptions, contract terms, personal data, anything your company has not approved for AI. Force the model to write "unknown" instead of filling gaps, as in AI prompts for sales prospecting. Empty deal record, empty brief: this routine rewards reps who write things down.
What does the three-minute clock look like?
| Clock | You do | Output |
|---|---|---|
| 0:00-0:45 | Copy the three inputs into the prompt | Raw material |
| 0:45-1:00 | Paste the prompt, hit enter | Nothing yet |
| 1:00-1:45 | Read the brief once, top to bottom | A draft goal, questions, risks, ask |
| 1:45-2:30 | Check every name, title and number against your notes; delete any "inference" you cannot back | A brief you trust |
| 2:30-3:00 | Keep one goal, pick two of the three questions, write the next-step ask in your own words | The plan you take into the call |
One row belongs to the AI; four are yours. Running over means your inputs were too long, not your prompt.
What is the prompt?
Our prompt A, for the call brief:
``` Brief me for a sales call that is already booked. Work only from what I paste here. Where it is silent, write "unknown"; where you infer, write "(inference)".
Call: [date, type: discovery/demo, attendees and titles on both sides] CRM: [stage, value, expected close, next-step field] Last conversation: [paste notes or summary] Email thread: [paste last messages]
Return, in this order, under 200 words: 1. A one-sentence goal, plus what has to happen for it to count. 2. Three questions, each tied to something we do not know yet. 3. What in this record could stall or kill the deal. 4. The next step to ask for, with a date and one action the buyer takes. ```
The "(inference)" tag is the money part: it shows which lines to confirm or delete. When the questions come back generic, run our prompt B:
Here are three questions for my call with [title] at [company]: [paste]. For each one, tell me whether the material below already answers it. If it does, replace it with a question about something still unknown, in plain language a [title] would answer in one sentence. Material: [paste notes].
What should come out, and what does good look like?
All examples below are invented, in our words.
| Output | Weak version | Strong version |
|---|---|---|
| Goal | Build rapport with the CFO | Get the CFO to agree to a one-branch pilot with a start date |
| Risks | Budget concerns | The CFO has never been on a call, and the price is above the labor saving they quoted |
| Next-step ask | Follow up next week | "Can we pencil the Northport branch in for November 10?" |
If you already know the answer, cut the question. Three questions start the meeting, they are not the meeting: for how many to ask in a full discovery meeting and how to space them, see discovery call without interrogating. For the questions themselves, by buyer role, see what discovery questions to ask.
The ask needs a date and a buyer action. New economic buyer on the call? Check what is still unqualified: BANT vs MEDDIC.
Our example, invented: a demo call with a CFO joining for the first time
Invented example, every number made up. You sell order-entry software to Coastline Marine Parts, a boat-parts distributor with 3 branches and about 120 staff.
Deal record: "Demo booked", $36,000 a year, closing November 28, next step blank. Keisha Moran (operations) was on both earlier calls; Raúl Mendes (CFO) joins tomorrow for the first time and wants "the numbers before the call." Keisha's notes: each branch loses "about six hours a week" re-keying phone orders; about 1,500 orders a month; "around 2% go out wrong"; a wrong order costs "maybe $45"; the ERP module renews in February.
What the rep changes in the brief:
- Goal. AI: "Earn the CFO's confidence." Rep: get Raúl to agree to a 30-day pilot at one branch, with a start date, before the call ends.
- Questions. Keep the February renewal question and "Who signs a $36,000 commitment?" Drop the roll-out question; Keisha already answered it on call two.
- Risks. No CFO on any prior call (an inference, confirmed); price above the labor saving (to check); no success metric agreed (confirmed).
- Ask. AI: "Propose a follow-up." Rep: "Raúl, if the pilot numbers hold, can Northport go live on November 10, with you on the kickoff?"
The math the rep checks in minute two (our math, invented numbers):
- Labor: 6 hours a week x 3 branches = 18 hours a week. 18 x 52 = 936 hours a year. At a loaded $32 an hour, 936 x 32 = $29,952 a year.
- Errors: 1,500 orders x 2% = 30 wrong orders a month. 30 x $45 = $1,350 a month, x 12 = $16,200 a year.
- Total cost of the problem: $29,952 + $16,200 = $46,152 a year, against a $36,000 price.
The AI flagged "price above labor saving" because it counted only labor. The error line kills that risk and hands Raúl his numbers as slide one. The AI's miss was a missing line, not a wrong sum: that is why "every number" is on the check list.
What must a human still check?
| Check | Why | Seconds |
|---|---|---|
| Names and titles | A wrong title costs credibility | 10 |
| Every number | Models round, merge or invent figures | 20 |
| Anything marked "inference" | Delete it or confirm it; never say it aloud as fact | 15 |
| The goal | Only you know what the deal needs next | 20 |
| The news item | Open the link; if you cannot, do not mention it | 10 |
About 75 seconds, which is why the clock gives you 1:45 to 3:00.
Which tools build the brief for you?
Per their help pages, read 2026-10-06:
| Tool | What its help page says the prep includes | Where it shows up | Needs |
|---|---|---|---|
| Gong meeting prep | Summary with objectives, key topics, prior highlights, action items, suggested questions; "Ask anything" | Homepage, account console, activity page, digest emails, mobile app | Gong Foundation; summary and "Ask anything" need prior tracked activity on the account. Its AI "dry run" rehearsal is listed as available on Enable |
| Granola pre-meeting briefs | 2-3 bullets: open threads, context, agenda items, from earlier Granola notes, calendar, web sources and optionally Gmail | Top of the chat panel when you open the note for an upcoming meeting | Business or Enterprise plan; Google Calendar connected; external meetings only; skipped when information is thin |
Granola's page: "Briefs are gathered for you overnight." Both lean on history inside that tool; a new rep or account gets little, so the pasted prompt wins there. On cost for small teams, see is Gong worth it.
When should you skip each tool?
- Gong meeting prep: skip it if you are not already a Gong customer; its help page lists it as available on Gong Foundation, Gong's core platform.
- Granola briefs: skip it if your meetings are internal or your calendar is not Google; per its help page, the brief is not generated.
- The pasted prompt: skip it when your tool already writes a brief; spend the three minutes checking that one.
How do you know the routine is working?
Track three numbers (our guidance):
- Percent of calls closing on a dated next step. The one that matters. To get the buyer to agree to the steps, see what a mutual action plan is.
- Minutes per prep. It should fall.
- Percent of briefs where you deleted an "inference". High means thin notes.
After the call, log the four lines from the last FAQ and plan how long to wait to follow up after a demo. Deals that stall usually stalled on a call nobody planned: why B2B deals end in no decision.
Sources
Research and vendor pages accessed October 6, 2026.
- Gong Help Center: Get ready for meetings with AI-powered meeting prep
- Granola Help Center: Pre-meeting briefs
- The three-minute clock, the four-part output, the prompts, the check table, the metrics and the Coastline example are our own; no outside source.