What is the best ChatGPT prompt for sales?

There's no single magic prompt, but the best ones share a shape: who you sell to, what you sell, the stage of the deal, the one outcome you want and the facts the model must use. 'Write a cold email to a head of operations at a 200-person logistics firm, under 90 words, one ask: a 15-minute call, based on these notes' beats 'write a sales email' every time.

Can AI write cold emails that actually get replies?

It can write the draft, but replies come from relevance: a real reason you're contacting this person and one easy ask. Give the model a specific fact about the prospect, ban the clichés, keep it short and edit before sending. Mass-sending unedited AI emails hurts both your deliverability and your reputation.

Is it safe to paste customer data into ChatGPT?

Only if your company allows it and the tool's data settings fit. Business and enterprise plans usually offer stronger data protection than personal accounts. When in doubt, remove names, contact details and contract values and use placeholders instead, and remember that privacy laws such as the GDPR still apply.

Will AI replace salespeople?

Not the parts that close deals. AI is good at research, first drafts, summaries and practice. Buyers still want a person who understands their problem, earns their trust and negotiates in good faith. The reps who gain most use AI to win back time for those conversations.

AI Prompts for Sales: 12 Templates From Prospect to Close

Ask a salesperson where their week goes and "selling" is rarely the first answer. Salesforce's State of Sales research found that reps spend just 28% of their week actually selling. The rest disappears into research, data entry, internal admin and writing the same follow-up for the fortieth time. That's exactly the kind of work AI is good at.

The trouble is that most AI prompts for sales produce the email everyone deletes: "I hope this finds you well", three adjectives about innovative solutions, a calendar link. Buyers spot it instantly, and they're using AI too. In HubSpot's 2025 State of Sales survey, 74% of sales pros said AI makes it easier for buyers to research products on their own. Generic messages don't survive that. The 12 templates below work the other way round: you supply the facts and the judgement, the model does the drafting.

What every sales prompt needs

A sales prompt fails for the same reason a sales call fails: not enough discovery. Before the model writes a word, it needs six things from you.

What you sell: the product in one sentence, plus the outcome customers buy it for. "Route-planning software that cuts fuel costs for delivery fleets" works. Your homepage tagline doesn't.
Who you're selling to: role, type and size of company, and their situation. "Head of Operations at a 200-person logistics firm that just lost a big contract" gives the model something to aim at.
The stage: first touch, discovery, proposal, negotiation or renewal. The same buyer needs a very different message at each one.
One goal: the single next step you want, whether that's a reply, a meeting, the decision-maker's name or a signed order form.
Your facts: proof points, customer results, prices and terms. Paste them in. If you don't, the model will invent them, and convincingly.
Constraints: tone, length, banned phrases, language. "Under 90 words, no buzzwords, sounds like a person" is a constraint.

Write these down once and keep them as a reusable block at the top of your prompts. For a quick structure to start from, the AI prompt generator builds the skeleton for you, and the guide to improving your prompts covers the principles behind it.

Prospect research in minutes, not an hour

Good outreach starts with knowing why this account, and why now. AI can turn a pile of raw material (the prospect's about page, a recent press release, a job ad, a LinkedIn post) into a usable brief in a minute. The key words are "raw material": paste the sources in rather than asking the model what it knows about a company. A chatbot without web access will happily make up a funding round. If you need fresh information, use a tool that searches the web and cites its sources, such as Perplexity, and click the links.

1. Account research brief

You are helping me prepare to contact a prospect. Use ONLY the sources I paste below. If something isn't in them, write "not in sources". Don't guess. WHAT I SELL: [one sentence + the outcome customers buy it for] Give me: 1. What the company does, in two sentences 2. Three things that look like current priorities, each with the line from the sources that suggests it 3. Two problems we could plausibly help with, clearly labelled as hypotheses 4. One specific, non-flattering reason to contact them this month SOURCES: [paste the about page, recent news, job ads, posts]

The "labelled as hypotheses" line matters. It stops a guess from sliding into your email as a fact, which is how reps end up congratulating a prospect on an acquisition that never happened. Our guide to AI hallucinations explains why models do this.

2. Buyer persona map

Before you contact a whole segment, map the person you're selling to: what they're measured on, what frustrates them and the words they actually use.

I sell [product] to [job title] at [type and size of company], mostly in [region]. Describe this buyer the way an experienced sales manager would brief a new rep: - What they're measured on (KPIs) and who they report to - Their top 5 day-to-day frustrations, in the words they'd use themselves - What would make them ignore a cold message - The 3 objections they're most likely to raise, and what usually sits behind each one - Who else typically gets involved in a purchase like this Flag anything that varies a lot by industry or region.

Treat the output as a hypothesis, not market research. The most useful thing you can do with it is correct it after your next five calls, then reuse the corrected version in every prompt that follows.

Cold emails and LinkedIn messages that don't read like templates

Cold outreach lives or dies on the first line, and that line has to prove you're writing to this person for a reason. AI is excellent at this as long as you give it the reason. Feed it a batch of prospects with one real observation each, and ask for openers rather than whole emails.

3. Personalised first lines, in batches

Write one opening line for a cold email to each prospect below. RULES: - Max 25 words each - Build each line on the specific fact I give. Don't add facts. - No compliments ("love what you're doing"), no "I hope this finds you well", no "I came across your profile" - Connect the fact to a problem we solve: [problem] - Plain, conversational English PROSPECTS: 1. [name, role, company]: [fact, e.g. hiring 4 dispatchers in Leeds] 2. [name, role, company]: [fact] 3. [name, role, company]: [fact]

The rest of the email (a short body, a single ask and the subject line) is covered in depth in our AI prompts for email, follow-up nudges included. LinkedIn needs a lighter touch: shorter, more personal and with no pitch in the connection request.

4. LinkedIn connection note and first message

Write two messages for [name, role, company]. CONTEXT: [why I want to connect: shared event, their recent post, a mutual client] WHAT I SELL: [one sentence] Message 1: connection request. Under 300 characters. No pitch, no link. Give a genuine reason to connect. Message 2: first message after they accept. Under 80 words. Share one useful insight or resource about [topic], then ask a soft question that's easy to answer. No calendar link. Tone: peer to peer, not vendor to buyer.

If a draft still sounds machine-made, don't keep hitting regenerate. Tell the model exactly what's wrong ("too formal", "too many adjectives", "the closing question is pushy"), or use the rewrite prompt generator to change the tone. There are more fixes in how to make AI writing sound human.

Call prep: cold call openers and discovery questions

AI won't make the call for you, but it can make sure you never go in cold. On a cold call, the hard parts are the first 20 seconds and the first brush-off, so prepare both.

5. Cold call opener and the first brush-off

I'm cold calling a [role] at a [type of company]. I sell [product], which helps with [problem]. Why I'm calling this account now: [specific trigger] Write: 1. An opener that takes under 30 seconds to say: who I am, why I'm calling them specifically, and a request for permission to continue 2. Three natural responses to "I'm not interested", each ending with a short question 3. Two responses to "Just send me an email", aimed at getting one qualifying answer first Make it sound like a confident person talking, not a script. Short sentences, no jargon.

Discovery calls are where deals are really won, and the most common mistake is asking generic questions you could have answered from the prospect's website. Give the model your research brief and let it build questions around this account.

6. Discovery call question plan

I have a 30-minute discovery call with [name, role] at [company]. What I know so far: [paste your research brief or notes] What I sell: [one sentence] Build a question plan: - 2 opening questions to confirm their current situation (nothing I could find on their website) - 4 questions to uncover problems and what those problems cost them - 3 questions about how they buy: who's involved, the budget process, timing - 2 questions to agree on next steps For each question, add one line on what a good answer would tell me. Order them so the conversation flows naturally.

If your team uses a sales methodology such as SPIN, MEDDIC or BANT, name it in the prompt and the model will organise the questions around it. Don't read the plan out loud. Use it as a map, and follow the buyer when they take you somewhere more interesting.

Practise hard conversations with AI role-play

Role-play is the most underrated use of AI in sales. A model can play a sceptical finance director, an overloaded IT manager or a procurement lead who only cares about price, as many times as you like and with nobody watching. The trick is to make it a difficult buyer, not a polite one.

7. Role-play with a tough buyer

Let's role-play a sales call. You are a [role] at a [type of company]. You're busy, sceptical and you've been burned by a vendor before. You currently use [competitor or spreadsheet]. I sell [product]. My goal for this call: [e.g. agree a demo with your team]. Rules: - Stay in character and reply only as the buyer, one turn at a time - Push back realistically: price, timing, "we're fine as we are" - Only warm up if I ask good questions and connect to your actual problems - When I type END, step out of character and score me from 1 to 10 on discovery, listening, objection handling and next steps, then tell me the two things I should change Start by answering the phone.

Run the same scenario three times and you'll start hearing your own habits: the pitch that comes too early, the discount offered before anyone asked. The role prompting guide explains why a detailed persona makes a far more realistic sparring partner.

Handle objections without sounding scripted

The objection you hear is often not the real one. "Too expensive" can mean "I don't see the value yet", "it's not my budget" or "I'm comparing you with something cheaper". A good objection-handling prompt asks the model to separate those possibilities before it drafts a single answer.

8. Objection handling

A prospect said: "[exact objection, e.g. We already use X and it's fine]" Context: [deal stage, who said it, what they care about, what we know about X] 1. List the 3 most likely meanings behind this objection 2. For each one, give me a clarifying question that would tell me which it is 3. Then write 3 short responses for the most likely meaning: one empathetic, one based on evidence (use only these proof points: [paste]) and one that asks for a small next step No pressure tactics, no fake urgency, no badmouthing the competitor.

Tip: Keep an objection log. After every call, paste the objections you heard, word for word, into one document. Once a month, give the whole log to the AI and ask for patterns: which objections are growing, at what stage they appear and which answers led to a next step. Most teams already have this coaching data and never read it.

Demos and proposals built around the buyer's words

A demo that walks through every feature is a demo nobody remembers. Give the model your discovery notes and let it build the agenda around the problems the buyer actually described, in their language rather than your product's.

9. Tailored demo agenda

Here are my discovery notes from a call with [company]: [paste notes] Here is our feature list: [paste] Build a 30-minute demo agenda: - Open by restating their top 3 problems in their own words (quote the notes) - Map each problem to the one or two features that solve it, and leave everything else out - For each section, suggest one question to check it has landed - End with a proposed next step and who should be in that meeting Flag any problem from the notes that our features don't solve. I'd rather know now.

Proposals work the same way. The model is quick at turning messy notes into a clear executive summary; your job is to make sure every number in it is real. If you're turning it into slides, our AI prompts for presentations show how to keep the deck tight.

10. Proposal executive summary

Write the executive summary of a proposal for [company], one page maximum. Use only these inputs: - Their situation and goals: [paste discovery notes] - Our proposed solution: [paste scope] - Pricing and terms: [paste] - Relevant customer results: [paste real case studies, with numbers] Structure: their challenge → what success looks like for them → our approach → expected outcomes → investment → next steps with dates. Write for a busy executive who will read only this page. Wherever a number or date is missing, write [TO CONFIRM] instead of estimating.

Follow-ups that keep deals moving

Deals rarely die from a "no". They die from silence after a good meeting. The fix is a recap that goes out the same day, plus a follow-up plan that adds something new at every touch instead of "just checking in".

11. Meeting recap and follow-up sequence

Here are my notes from today's call with [name, role, company]: [paste notes or transcript] Part 1: Write a recap email I can send today. Under 150 words. Include their key priorities as they described them, what we agreed, the next steps with owners and dates, and the one question they still need to answer. Part 2: If they go quiet, plan 4 follow-ups over the next 3 weeks. Each one must add something new: a relevant insight, a customer example, an answer to an open question or a simple yes/no choice. The last one is a polite close-the-loop email. Under 80 words each. No "just bumping this up".

Save the sequences that work in a prompt templates library so the whole team can reuse them. Recaps for existing customers and renewals also benefit from the tone rules in our AI prompts for customer service.

Let AI handle CRM notes and pipeline admin

This is where a lot of lost selling time goes. If your calls are recorded and transcribed, with everyone's consent, a single prompt can turn a 40-minute transcript into structured CRM notes and flag the risks you'd rather not see.

12. Call transcript to CRM notes and deal risks

From the transcript below, fill in these fields. Use only what was said, and write "not discussed" if a field isn't covered. - Pains (in the customer's words) - Business impact of those pains - Decision process and the people involved - Budget and timing signals - Competitors or alternatives mentioned - Agreed next step, owner and date Then list up to 3 deal risks (for example: no access to the decision-maker, no agreed date, a competitor already preferred) and one action to address each. TRANSCRIPT: [paste]

Plenty of tools now do part of this automatically: Teams and Zoom can summarise meetings, and most CRMs are adding AI assistants. A prompt like this still earns its place because you decide which fields matter. In Microsoft 365, Copilot can recap your Teams calls, and if you export your pipeline to a spreadsheet, our AI prompts for data analysis help you find the deals that are stalling.

Build a reusable sales assistant

Once you're pasting the same context into every prompt (product, ideal customer, tone, proof points, words you never use), it's time to turn it into a custom assistant. A custom GPT in ChatGPT, a Claude project or a Gemini Gem can hold that context permanently, so every new chat starts fully briefed.

Give it your product and pricing basics, your ideal customer profile, three to five real customer stories, your most common objections with approved answers, your brand voice and a short list of things it must never do, such as promising discounts, inventing customer names or committing to delivery dates. The system prompt generator gives you a solid structure, and our step-by-step guide to custom GPTs shows how to set one up.

What to keep away from the AI

AI makes a good sales assistant and a bad sales manager. Here are the lines not to cross:

Personal and confidential data: don't paste customer names, contact details, contract values or call transcripts into a personal AI account. Use the tools your company has approved, check that customers agreed to call recording, and swap details for placeholders when in doubt. Privacy laws such as the GDPR apply to AI tools too.
Invented proof: every number, quote, customer name and claim in a sales message must be true and checkable. If the model writes "teams like yours save 40%", either you have the case study or the line goes.
Volume over relevance: AI makes it easy to send a thousand "personalised" emails, and spam filters and buyers both notice. Ten well-researched messages beat a thousand lightly edited ones.
The judgement calls: pricing, discounts, contract terms and anything you promise a customer stay with you. The model drafts, you decide.

Used this way, ChatGPT prompts for sales don't replace the conversation with the buyer; they clear the road to it. Start with the one task that eats most of your week, use its template for a fortnight and refine it with what you learn. For the wider picture of AI at work, read our AI prompts for business.

AI Prompts for Sales: 12 Templates for Every Stage of the Deal

What every sales prompt needs