Most salespeople open ChatGPT, type a lazy prompt, get a generic answer, and quietly decide AI isn't for them. The reps who win do one thing differently. This is that one thing, with the prompts to prove it.
Use ChatGPT for sales as a fast first draft, never the final word. Feed it tight context and strict rules, then make every output sound human before it reaches a buyer. Done well, it compresses pre-call research, cold emails, discovery questions, and objection prep from hours into minutes, while you keep control of the relationship and the close.
7 min read
Here is the uncomfortable truth. The tool is not the problem. The prompt is.
Most reps treat ChatGPT like a vending machine. They type "write a cold email to a marketing director" and paste whatever comes back. The buyer has read that exact email three times this week, recognises the rhythm in two seconds, and deletes it. The rep concludes that AI doesn't sell. The rep is half right: bad AI doesn't sell.
The data shows how common this is. Only about 19% of sales reps use the AI built into their sales tools; most are copy-pasting prompts into a general chatbot, which misses the context, CRM data, and signals a purpose-built workflow would catch. Yet the upside for the people who use it well is real: 56% of sales professionals now use AI daily, and those daily users are twice as likely to exceed their targets. Sellers who partner effectively with AI are 3.7 times more likely to hit quota than those who don't.
The gap between those two groups is not talent or budget. It is method. So let's fix the method.
of reps use the AI inside their sales tools. The rest paste into a raw chatbot.
HubSpot, 2025more likely to beat targets among daily AI users vs non-users.
LinkedIn, 2025more likely to hit quota when reps partner well with AI.
GartnerGive it constraints. That's the whole game. A vague prompt produces the bland, recognisable "AI voice." A constrained prompt produces something you could actually send. In practice, the constraints do roughly 80% of the work, and which model you use barely matters next to how specifically you prompt it.
Every strong sales prompt carries five things. Memorise this stack and you'll never write a weak prompt again.
That last one matters more than the rest combined. A first name in the greeting is not personalisation anymore. A line about the role change they posted last week is.
Keep this template and fill the brackets. It works for almost any sales writing task.
Write a [type of message] from [my role] at [my company] to a [prospect role] at [prospect company]. About us, in 2 sentences: [what you do and the proof]. Our value in one sentence: [outcome you create]. One specific, recent detail about them: [paste the detail]. Rules: under [N] words, plain text, no emojis, peer-to-peer tone. Open with the specific detail, not a compliment. Ban these phrases: "I hope this finds you well", "just reaching out", "I wanted to", "quick question", "circle back", "touch base". End with one low-commitment ask.
Output quality equals input quality. The more real material you hand ChatGPT, the better and more human the result: the prospect's job posting, a call transcript, three of your own best emails. Vague in, generic out. Every prompt below assumes you bring something specific to feed it.
These are working prompts, not examples to admire. Each one has the constraints baked in. Adjust the brackets to your world and save them somewhere your whole team can reach.
Account research is where reps lose the most time. AI-assisted reps report cutting research and personalisation time by up to 90%. This prompt turns 30 minutes of tab-switching into a focused brief.
Brief me for a sales call with [company]. I'm pasting their [website copy / latest job posting / a recent LinkedIn post] below; use that, plus anything you can verify by searching the web, and label anything you infer as a guess. In under 200 words: what they sell, who their customers are, how they likely make money, their 3 most probable priorities this quarter, and 3 specific problems that [one line: what you do] could solve. Don't invent facts, numbers, or names. End with 2 sharp questions I should open with. [paste the source here]
ChatGPT will state an invented number, client name, or company fact with total confidence. Only rely on a brief that came from a source you pasted or a link it actually searched, and verify any specific claim before you repeat it on a call. One fabricated detail loses the room.
Iterate, don't accept the first draft. Teams that push back on the output three or four times before sending report their reply rates roughly doubling.
Write a cold email from [my role] at [my company] to [prospect role] at [prospect company]. About us: [2 sentences]. Value in one line: [outcome]. First line must reference this specific detail: [paste detail]. Rules: under 90 words, plain text, no pleasantries, no emojis. Ban: "I hope this finds you well", "just reaching out", "quick question". End with one low-commitment ask. Then give me a second version with a different angle.
Keep cold emails between 50 and 125 words, and ask for two subject lines to test, since personalised subject lines can lift open rates 22 to 36%. One caution worth saying plainly: even a perfect email lands in spam without clean sending setup, which is a separate job covered in our AI cold email templates.
I'm meeting a [prospect role] at a [industry] company about [problem area]. Give me 8 open discovery questions that uncover business impact and quantify the cost of their current situation. No yes/no questions. Order them broad to specific. Skip generic questions they've answered a hundred times, like "what's your biggest challenge".
You don't have to walk into the hard conversation cold. Make ChatGPT play the difficult buyer first.
Act as a skeptical [buyer role] who thinks [my product] is too expensive and already uses [competitor or status quo]. I'll pitch; you push back with realistic objections and don't fold easily. After each of my replies, score it 1-10 and tell me in one line what to improve. Stay in character until I say "stop".
Write a breakup email to [name] after [N] unanswered messages. Include one genuinely useful insight or stat relevant to their role, and name a future trigger that should make them re-engage. Under 80 words. Tone: helpful, not passive-aggressive. No guilt-tripping.
This is the safest and highest-return use of all, because ChatGPT is summarising your own call, not inventing anything. It also clears the admin that drags reps down: the average seller now juggles eight tools, and overwhelmed sellers are 45% less likely to hit quota.
Here are my notes, or the transcript, from a call with [name, role, company]: [paste]. Do three things: (1) a 5-line summary of what they need and any objections they raised, (2) clear next steps with owners and dates, (3) a short follow-up email that confirms the next step and references one specific thing they said. Use only what's in my notes; don't add anything.
Even a well-prompted draft carries fingerprints. The model reaches for the same words, the same tidy structure, the same eager tone. Buyers have learned to spot it. Stripping those tells is the difference between a draft and something worth sending, and it takes thirty seconds.
The most powerful move is also the simplest: show ChatGPT your own voice. Paste two or three emails you've written that landed well and tell it to copy them. It will match your rhythm and word choice far more convincingly than any rule ever could.
Here are 3 emails I've written that got replies: [paste them]. Study the voice: sentence length, how formal it is, how it opens and closes. Now write a [message type] to [prospect] about [topic] in that exact voice, under [N] words. Match me; don't sound like AI.
Then tighten what's left. Ban the phrases that scream "a machine wrote this" and keep them out of every message:
Finally, run a humanise pass. After ChatGPT gives you a draft, send it back through this:
Rewrite this so it reads like a busy human typed it in thirty seconds, not like AI. Vary the sentence length. Cut every adjective that isn't doing real work. Remove anything that sounds like a template, especially the "compliment, then pain point, then pitch" pattern. Keep it under [N] words and keep my one specific detail.
The structure itself is a tell. The recognisable cold-email shape (a compliment, a pain point, a pitch, a call to action) is so common that people delete it on sight. Break the pattern. Lead with the specific detail, say the one useful thing, ask for one small step. Stop.
Here is the difference in practice.
Hi John, I hope this email finds you well! I wanted to reach out because I noticed your company is growing fast. We help companies like yours improve sales efficiency. Would love to hop on a quick call to explore synergies. Let me know your thoughts!
John, saw you're hiring three SDRs this quarter. That usually means pipeline is outrunning the team's capacity to follow up. We helped a similar firm cut follow-up time per rep without adding headcount. Worth 15 minutes next week? If not, no problem.
This is the question most "ChatGPT for sales" guides skip, and it's the one that can get you in trouble. Treat a general chatbot like a public room, not a private vault.
Customer names and contact lists, signed contracts, anything under NDA, non-public pricing, internal financials, login credentials, and personal data you don't have permission to share.
Use role-based context instead of real identities: "a procurement lead at a mid-size logistics firm" carries everything ChatGPT needs without exposing your client. If your plan offers a no-training or enterprise setting, switch it on and check it. Consumer plans and business plans handle your data differently, so confirm yours before you paste anything sensitive.
The reps who get the most from ChatGPT spend ten minutes configuring it, then never re-explain themselves again.
And use the features, not just the chat box. Turn on web search for live prospect research, upload files (a job post, a transcript, your call notes) instead of retyping them, use Canvas to edit longer drafts side by side, and keep your team's best prompts in a shared Custom GPT or Project. These features change often, so check what your plan includes.
Ask ChatGPT: "Explain to a 12th grader why [product] saves a [buyer role] money compared to [status quo]." If it can't do that simply, your pitch is too complicated, and no prompt will save it.
One more thing worth saying plainly. ChatGPT Plus is around $20 a month, which is the cheapest leverage in your sales stack right now. The cost was never the barrier. The method was.
ChatGPT is a sharp tool, not a system. It writes the email; it doesn't decide who to email, when, or why. That judgment is yours, and it's where deals are actually won. A good chatbot prompt sits inside a bigger motion: finding the right buyer, building trust, qualifying honestly, and following through after the sale.
That motion is the AI-ENABLE Sales Framework, a six-step approach to using AI across the whole buyer cycle while keeping the human in charge. ChatGPT does its best work in the first two steps, Engage and Nurture, where research and personalised outreach decide whether a conversation ever starts.
Rajiv Sharma and the NLP Limited team train sales teams across the UAE, India, and Africa to use AI with judgment, not hype. Start with a strategy conversation.
Rajiv Sharma is a sales coach, business strategist, and NLP Master Trainer with more than 35 years of experience training teams across India, the Middle East, and Africa. He created the AI-ENABLE Sales Framework and wrote AI-Powered Sales Success: Outsmart the Competition (NLP Limited). More at RajivSharma.me.
Yes, when used as a drafting and research assistant rather than an autopilot. ChatGPT is strong at pre-call research, first-draft outreach, discovery questions, and objection practice. It is weak at judgment, timing, and relationships, which stay with the rep. Used with tight prompts and a human edit, it saves hours a week without lowering quality.
For most reps, ChatGPT Plus at around $20 a month is enough, and it unlocks Custom Instructions and Custom GPTs that keep a team consistent. Larger teams that handle sensitive data should consider Team or Enterprise plans for stronger privacy controls and admin settings. The plan matters far less than how specifically you prompt it.
No. ChatGPT removes repetitive work like research and first drafts, but it cannot build trust, read a room, negotiate, or own a decision. The reps who win treat it as a force multiplier for the human parts of selling, not a replacement for them. Mass AI-generated outreach with no human judgment is exactly what buyers now ignore.
Give it one specific, recent detail about the buyer, set a hard word limit, and ban the giveaway phrases like "I hope this finds you well" and "just reaching out." Then run a humanise pass that varies sentence length and breaks the standard cold-email structure. The specific detail and the human edit do most of the work.
It is safe for research, drafting, and practice as long as you never paste confidential or personal data into a general chatbot. Use role-based descriptions instead of real names, keep contracts and pricing out, and switch on any no-training or enterprise privacy setting your plan offers. Treat it as a public room, not a private vault.
Yes, it can. ChatGPT will state invented details, numbers, or company facts with full confidence if you let it research from memory. Ground every brief in a source you paste or a live web search, tell it not to guess, and verify anything before you repeat it on a call. One fabricated detail can cost you the deal.
Written by Rajiv Sharma, NLP Limited. Part of the AI-ENABLE Sales Framework series. Statistics: LinkedIn State of Sales (2025), HubSpot State of Sales (2024–2025), Gartner. Tool figures and prompt-craft guidance reflect practitioner reporting current at the time of writing and may change.
