How to use AI in sales: 5 uses, step by step, with prompts

To use AI in sales, pick one task around your calls, give the AI the context it needs (who the buyer is, what you know and what you want back) and check every fact before anything reaches a buyer. The five uses that pay off fastest for most sellers are account research, first emails, discovery prep, call recaps and objection practice.

This guide walks through each of the five with a prompt you can copy, what to paste in and what to check before you use the result. For where AI fits across the whole deal and what should stay with people, start with our guide to AI in sales.

What do you need before you start?

You need three things: an AI tool your company approved, the material the AI will work from and a few prompts you save and reuse. Most of the quality comes from the second one. AI writes from what you give it, and when you give it nothing, it guesses.

Ask your sales operations or IT team what data you can paste and where. A common rule: public information about a company is fine anywhere, and anything a customer told you (names, numbers, transcripts) goes only into the approved workspace.

What makes a good sales prompt?

A good sales prompt tells the AI who it is helping, gives it the facts, asks for one specific output and sets the format and the limits. Most weak AI output in sales comes from a one-line prompt like "write a cold email to a VP of Sales", which gets the same email every VP of Sales already deletes.

The last row saves the most time. When the AI has to label what it guessed, a confident draft turns into one you can check in two minutes, because you know exactly which lines to look at.

How do you use AI in sales, step by step?

Use AI on the work before and after each conversation, in the order a deal runs: research, outreach, prep, recap and practice. Each step below has a prompt to copy. Replace the words in capitals with your own details.

1. 01

   Research the account

   A one-page brief before the first call.

2. 02

   Draft the first email

   Four sentences built on a trigger.
3. 03

   Prepare discovery

   Questions, and what a strong answer sounds like.

4. 04

   Write the recap

   Email, CRM notes and the gaps, from the transcript.

5. 05

   Practice the objection

   An AI buyer who pushes back before the real one does.

The example running through all five: an account executive at Halcyon Labs, a fictional company that sells onboarding software to sales teams, is working a deal with Northwind, a software company with a 40-person sales team that just posted ten SDR roles.

1. Research the account before the first call

Use it the day before a first meeting. Paste in the company's About page or a recent press release, the buyer's LinkedIn summary and anything already in your CRM.

You are helping an account executive prepare for a first call.

Context:
- I sell ONBOARDING SOFTWARE FOR SALES TEAMS.
- The buyer is MAYA CHEN, VP OF SALES AT NORTHWIND (software, 40 sellers).
- What I found: PASTE THE PRESS RELEASE, HER LINKEDIN SUMMARY AND CRM NOTES.

Task: write a one-page brief with:
1. What the company does, in two sentences.
2. Three things that changed at the company in the last six months.
3. The two priorities Maya most likely has this quarter, and why.
4. One opening line for the call that refers to something specific.

Rules: use only the information I pasted. If you infer something,
label it "guess". Under 250 words.

What to check: every fact against the source you pasted, and every line labeled "guess". The brief is for you, so a wrong guess costs little if you catch it before the call. Rewrite the opening line in your own words, since it is the first thing the buyer hears. For building lead lists the same way, watch the session how to use AI to prospect for leads.

2. Draft the first email from a trigger

Use it when you have a reason to write now: a new hire, a funding round, an expansion, a batch of job posts. Paste in the trigger, a result from a similar customer that you are allowed to share and why the trigger matters.

Write a cold email from me to MAYA CHEN, VP OF SALES AT NORTHWIND.

Trigger: NORTHWIND POSTED 10 SDR ROLES LAST WEEK.
Why it matters: new SDRs often take weeks to book their first meeting.
Proof: A SIMILAR TEAM CUT TIME TO FIRST MEETING FROM 9 WEEKS TO 5.
Ask: a 20-minute call next week.

Rules: 4 sentences, under 80 words. No opening line about hoping
she is well. No buzzwords. Plain words, the way a sales leader talks.
Then give me 3 subject lines of 4 words or fewer.

What to check: the proof point has to be real and approved for sharing. Then rewrite at least one sentence with something only you know, like a detail from a similar customer or from her own team's job posts. The AI saves you the blank page, and your sentence is the one that gets the reply. The session AI informed subject lines shows how to test subject lines the same way.

3. Prepare discovery questions

Use it the day before a discovery call. Paste in the brief from step 1 and the steps of your discovery method or scorecard (MEDDICC, SPIN, GPCT or your own).

I have a discovery call with MAYA CHEN at NORTHWIND tomorrow.
My research brief: PASTE IT.
Our discovery method has these steps: PASTE YOUR SCORECARD STEPS.

Task:
1. Write 8 open questions, at least one per step, in plain language.
2. For each, write the answer that would mean a strong deal and the
   answer that would mean a weak one.
3. List 3 questions Maya might ask me, with a short, honest answer.

Rules: no yes-or-no questions. No question I could answer from
her website.

What to check: cut the list to the four or five you would really ask. Eight questions read from a list sound like a survey, and the best discovery questions come from following the buyer's last answer. The strong and weak answers are the useful part: they tell you what to listen for. The session discovery calls with AI goes further on prep, and the SPIN Selling guide shows how to order the questions.

4. Turn the call into a recap and CRM notes

Use it within the hour after the call. Paste in the transcript from your note-taker (from the approved tool only) and your scorecard steps.

Here is the transcript of my discovery call with NORTHWIND:
PASTE THE TRANSCRIPT.

Task:
1. A recap email to Maya: what we heard, what we agreed and the
   next step with its date. Under 150 words, in my voice.
2. CRM notes: pains (quoted in the buyer's words), people mentioned
   and their roles, budget, timeline, next step.
3. Which of these steps I didn't cover on the call:
   PASTE YOUR SCORECARD STEPS.

Rules: quote the buyer exactly when you use her words. If something
wasn't said, write "not discussed". Don't fill gaps.

What to check: names, numbers and dates. Transcripts mix up names and round numbers, and a recap with the buyer's name spelled wrong undoes a good call. Part 3 is the one most people skip, and it is the most useful: it tells you what to ask on the next call. To turn those steps into a scorecard your team shares, follow our guide to sales call scoring.

5. Practice the objection before the real call

Use it before a call where you expect pushback on price, timing or "we can do this ourselves". Paste in what the buyer said last time.

Play DAVID OKAFOR, CFO AT NORTHWIND. You are careful with money and
you have seen software that nobody used after month two.

Context: your VP of Sales wants onboarding software for 10 new SDRs
at $24,000 a year. You think a sales manager could onboard them instead.

Rules:
- Stay in character. Push back the way a real CFO would, with short
  questions.
- Don't agree until I give you a clear reason tied to cost or time.
- After 8 exchanges, step out of character and tell me what worked,
  where I sounded defensive and the question I should have asked.

Start by asking me why this can't wait until next year.

What to check: AI buyers give in faster than real ones, so keep the rules strict. Run it twice and change one thing the second time, like opening with a question instead of the price. For the answers that work on the most common pushback, see our guide to common sales objections.

How do you check what AI gives you?

Read every draft as if a new hire wrote it: probably useful, possibly wrong on the details. The check takes two minutes when the prompt had good context. If a draft needs ten minutes of fixing, the prompt needs more context next time.

The same check works for internal output, like CRM notes. A wrong budget figure in the CRM ends up in the forecast, and the forecast ends up in front of your manager.

How much time can AI save a seller?

Most sellers get back a few hours a week, all of it from work around the calls. In Salesforce's State of Sales report, 7th edition (2026, 4,050 sales professionals surveyed), sellers said they spend 40% of their week selling and 60% on everything else. The five uses above all come out of that 60%.

Here is an example week for the Halcyon Labs account executive. The numbers are illustrative, so measure your own.

That is about five hours a week. Practice adds time instead of saving it, and it belongs on the list anyway, because it is the use that improves the next call itself.

To measure your own, write down how long each task takes for one week without AI and one week with it. The hours only count if they go back into conversations with buyers. Our guide to sales productivity covers how to protect them.

How can a sales manager use AI with the team?

Managers get the most from AI by using it on many calls at once: finding what top performers do differently, preparing one-on-ones and turning real objections into practice. Three uses that work:

1. Find patterns across calls. Paste a few calls from won deals and a few from lost ones, and ask what changed.
2. Prepare a one-on-one. Paste a person's last three calls and the scorecard, and ask for one strength and one thing to work on, each with a quote.
3. Build practice from real calls. Take the objection that cost the team a deal last month and turn it into the role play prompt from step 5.

A prompt for the first one:

Here are transcripts of 3 discovery calls from deals we won and 3 from
deals we lost: PASTE THEM.

Compare them. What did the seller do in the won calls that didn't
happen in the lost ones? Give me 3 differences, each with a quote from
a won call and the matching moment in a lost call.

Rules: no general sales advice. Only what is in these transcripts.

Six calls are enough for a hint and too few for proof. Check what the AI found against the next ten calls before you build a coaching plan on it. Then coach one move at a time: our guide to sales techniques maps the main moves by stage of the deal, and our guide to AI sales coaching covers how to run coaching with AI across a team.

What mistakes should you avoid?

Most AI mistakes in sales come from skipping either the context or the check. Five to watch for:

1. The one-line prompt. Without the buyer's details, you get the email everyone else got this week.
2. Sending the first draft. Buyers can tell when nobody read the email before it went out.
3. Customer data in a personal account. Use the tool your company approved, every time.
4. Reading AI questions on the call. Use the list to prepare, then follow what the buyer says.
5. Starting all five at once. Pick one use, make it a habit for two weeks, then add the next.

Where to start

Pick the step that matches your next call, copy its prompt into a note and fill it in before the call. After two weeks, compare the time it takes with your old way, then add the next step. For where AI fits across the whole deal and what to keep with people, go back to our guide to AI in sales.