AI prompts for sales: templates for every deal stage
16 AI prompts for sales that work in ChatGPT, Claude, Gemini or Microsoft Copilot, from account research to upsell, plus real estate and recruiting. Each one says what to paste and what to check, with a filled example, a Word file and an Excel prompt library with a test log.
Made by Anatolii Ulitovskyi 
Fact-checked by RH Fardin Updated 1. Sales AI prompt rules: what to paste, what to check
Start hereThese AI prompts for sales work in ChatGPT, Claude, Gemini or Microsoft Copilot. Paste your facts where you see square brackets, and end any prompt with the rules block, or save it once in your assistant's custom instructions.
Rules block: paste at the end of any promptRULES
1. Use only the facts in this message. If something is missing, write MISSING and what is missing. Do not guess.
2. Do not invent names, job titles, numbers, dates, prices, quotes, customer stories or results.
3. Quote people only with words from my notes, in quotation marks.
4. Mark anything you inferred with (guess).
5. No claims about savings or results unless I gave you the number and its source.
6. No flattery, fake familiarity or fake urgency.
7. Plain words, under word limit words. End with one next step.
The 6 checks before you send, save or post
- FactsIn what you pasted, or checked by you.
- Numbers and datesThey match the notes, the quote or the data, units included.
- NamesReal people, companies and job titles, spelled right.
- Quotes and resultsNo invented quote, testimonial, customer story or result.
- PromisesNo price, date, feature or guarantee you cannot deliver.
- RulesCold email basics, fair housing, fair hiring, no personal data.
Never paste
- Patient, health or other sensitive personal details, home addresses, salaries, passwords or API keys.
- A whole CRM export, or anything your company's AI policy does not allow. Paste only the columns the task needs.
When a check failsFix the draft, then the prompt: add a rule, save a new version and log both runs in the Excel test log.
2. AI prompts for prospecting: research and a first email
For a new lead or a first email. Paste the text from the company's public website yourself, so you know what the AI read.
Prompt 2A: account research from a public websiteI sell your product to who you sell to.
Below is text from the public website of company: paste the text of 2 or 3 pages, such as home, services, locations.
Task:
1. In 5 lines or less: what they do, where, for whom.
2. Up to 3 reasons they might need your product, each with the website sentence it is based on.
3. 3 questions the website does not answer.
Rules: quote the source sentence for every fact. If the text does not say it, write "not on the website". Add nothing from memory: no founding years, staff numbers, owner names or reviews.
Check before you use it
- A source sentence for every fact. Delete any fact without one.
- The reasons are guesses until the buyer confirms them: ask, do not tell.
Prompt 2B: first email from your factsWrite a first email from your name at your company to first name, job title at company.
Why I am writing: one reason and its source: their chat, their website or a referral.
What we do: your offer, in one sentence.
One proof I may share: a customer result shared with permission, or none.
The ask: one next step, with a day.
Rules: under 90 words. Only the facts above. No compliments, no "I noticed" unless the fact is above, no invented customers, results or numbers. A plain subject line that matches the email.
Variant: rewrite in the buyer's wordsRewrite this draft in the buyer's own words from these quotes: paste their words. Keep every fact and number, cut what does not help them, under 90 words. Draft: paste your draft.
Cold emailWriting to someone who never talked to you? Use the cold email templates. US law covers commercial email, B2B included: a matching subject line, your postal address and a clear way to opt out1.
3. Discovery call prompts: prep questions and call notes
Your notes onlyBefore the call: an agenda and the open questions. After it: your own notes as deal fields, with Not said for gaps.
Prompt 3A: discovery call prepI have a 30-minute discovery call with first name, job title at company, on date.
What I know: paste the chat, form answers or emails, and the checked research from prompt 2A.
Task: an agenda with minute marks and up to 8 questions to fill these fields: problem in their words, what it costs them, who decides, timeline, budget, next step. First list the fields I can already fill, with the line each comes from.
Rules: no questions about what they already told me. One question at a time, in plain words. No pitch before minute 15.
Prompt 3B: call notes to CRM fieldsHere are my notes from a call with first name at company on date: paste your own notes.
Turn them into 6 fields: Problem in their words; Numbers they gave; Who decides; Timeline; Budget; Next step, with date and owner.
Rules: use only my notes. Quote their words exactly, in quotation marks. If a field was not discussed, write "Not said". Do not round, add or estimate numbers, and keep the units (per week, per month, per location). Then list anything in my notes that fits no field.
Check before it goes into the CRM
- Each quote is in your notes, word for word, and every number has its unit.
- Use your own notes. Paste a recording or transcript only if your company allows it and everyone on the call agreed.
ThenPaste the checked fields into the deal's Notes. The call questions are in the discovery call script template.
4. Product demo and objection prompts: tailor, then practice
Tailor the demo to the buyer's problem, then let the AI play the buyer so you practice the hard questions first.
Prompt 4A: tailor the demo to the buyerHere is my standard demo outline: paste your outline.
The checked fields from the discovery call: paste the 6 fields.
People joining: names, roles and what each one is responsible for.
Demo length: 45 minutes.
Task: reorder the demo so their main problem comes first. Show at most 3 features, each linked to something they said, quoted. Add one question for each person joining. List what you cut and why.
Rules: only features in my outline. Promise no dates, integrations or results that are not in it.
Check before the demo
- Every feature you show is in your outline and works today.
- Each quote comes from the discovery fields. No promises on integrations, dates or results.
Prompt 4B: objection role-playPlay job title at a type of business who is talking to me about your product. Your concerns, from real calls: paste 3 to 5 objections, in the buyer's words.
Raise one concern at a time and wait for my answer. Push back once if my answer is vague. After 5 rounds, score each answer: did I ask a question first, answer with a fact I can show, and end with one next step?
Rules: stay in the role. Invent no facts about my product: ask me instead.
Keep the best answersSave the answers that scored well. Tested answers to price, timing and trust are in the sales objection responses.
5. Proposal and follow-up prompts: numbers from the quote
Copy the numbersThe AI writes the words around your numbers. Every price, quantity and date comes from the quote.
Prompt 5A: proposal summary from the quoteWrite a short summary for first name at company, to send on top of the quote.
Their problem, in their words: quote from the discovery fields.
What they get: line items from the quote.
Price and terms, copied from the quote: total, billing period, valid until.
Next step: one step, with a day and a time.
Rules: under 150 words. Copy every number and date exactly. Add no discount, deadline, start date or result that is not above. End with the next step.
Check before you send it
- Total, quantities and dates match the quote, line by line.
- No new promises: go-live dates, support hours or results.
Prompt 5B: follow-up after no replyWrite a follow-up email to first name at company. My last email, sent on date: paste it.
One new, useful fact I can add: a new answer, a date or a short resource.
Rules: under 70 words. Do not repeat the last email. No "just checking in", no guilt. Do not mention availability, slots or deadlines. Ask one easy question they can answer in one line.
Two, then stopAfter two follow-ups with no reply, stop. Recap and proposal emails written by people are in the sales email templates.
6. Upselling prompts: find candidates in your usage data
For current customers. Paste account-level usage, not personal data. A person decides whom to contact and what to offer.
Prompt 6A: spot upsell candidates in usage dataOur plans: plan names, limits and prices.
Account usage for the last 3 months, one row per account: paste columns such as account ID, plan, users, usage against the limit. No names, emails or personal data.
Task: list up to 5 accounts close to their plan limits or using features of a higher plan. For each: the numbers that show it, the plan that fits, and one opening question for the account manager.
Rules: use only the data. If a number is missing, write MISSING: do not estimate it. Flag thin evidence. Suggest no discounts.
Check before anyone calls
- Every number is in the data you pasted, and the plan fits the limits you gave.
- An account with an open complaint or billing problem gets that solved first, not an offer.
Prompt 6B: upgrade or renewal emailWrite an email to first name at company, a customer on current plan since date.
What the data shows: 2 facts from prompt 6A.
What they told us they want: their goal, in their words.
The offer: plan, price copied from your price list, from date.
Rules: under 120 words. Start with their goal, then the facts, then the price, stated plainly. No pressure and no deadline. End with one question.
Written by peopleRenewal and upsell emails written by people are in the sales email templates.
7. Real estate AI prompts with a fair housing check
Fair housingThe listing sheet and your showing notes are the only facts. US rules make it unlawful to publish a housing ad or statement that shows a preference or limitation based on race, color, religion, sex, disability, familial status or national origin2. Describe the home, not who should live there.
Prompt 7A: listing description from the listing sheetWrite a listing description for address or area, using only this listing sheet: paste beds, baths, size, lot, year built, features, fees and what is included.
Rules: under 150 words. Copy every number exactly. Describe the home and its features, not the people who might live there. No claims about schools, safety, neighbors or future value. If something is not on the sheet, write MISSING.
Prompt 7B: buyer follow-up after a showingWrite a short text to buyer first name after the showing of address on date.
What they said during the showing: paste your notes, in their words.
Next step: second showing, disclosures or a call, with a day.
Rules: under 60 words. Mention only what they said about the home. No comments about the people in the area. End with one question.
Prompt 7C: fair housing language checkCheck this draft for words that describe people instead of the property: paste the listing, ad or email.
List every word or phrase that could show a preference or limitation based on race, color, religion, sex, disability, familial status or national origin. For each, suggest wording that describes the property. If you find nothing, say so. Change nothing else.
Never in listing copy
- Who the home is for: 'adults only', 'no kids', 'Christian neighborhood', 'English speakers only'. No claims about safety, schools or neighbors.
- The check helps you look. Your broker, your MLS rules and the law decide.
8. ChatGPT prompts for recruiters: clients and candidates
Fair hiringStaffing firms sell twice: to the hiring company and to the candidate. Use only job facts the client approved. US law bans job ads and recruiting that show a preference based on race, color, religion, sex, national origin, age (40 or older), disability or genetic information3.
Prompt 8A: business development email to a hiring managerWrite a first email from your name at your firm to first name, job title at company.
Why now: one public fact, such as an open role on their careers page.
What we do, in one sentence: roles you fill, area, how you work.
One proof I may share: a placement result the client agreed to share, or none.
The ask: a 15-minute call, with a day.
Rules: under 90 words. No invented clients, fill rates or time-to-hire numbers. End with the ask.
Prompt 8B: candidate outreachWrite a short message to candidate first name about a job title role.
The job, as the client approved it: location, pay range, schedule, must-haves.
Why them: one fact from their public profile that matches a must-have.
Rules: under 80 words. State the pay range. Mention nothing about age, family, health, religion, origin or anything not related to the job. No praise you cannot back with the fact above. Ask for a 15-minute call.
Prompt 8C: job ad language checkCheck this job ad for words that could discourage people from applying because of race, color, religion, sex, national origin, age, disability or genetic information: paste the ad.
For each phrase, say why and suggest neutral wording. Also flag requirements the job does not need. Change nothing else.
Check before it goes out
- An ad that asks for 'recent college graduates' may discourage people over 40, and one that asks for 'females' may discourage men3.
- Pay range and must-haves come from the client, in writing. A person reviews every candidate, and resumes go into an AI tool only if your firm allows it.
For small sales and support teams
Run the prompt anywhere. Keep the result with the deal.
In CRMsoftware.pro, the website AI agent answers leads 24/7 from your Instructions and Knowledge Base. Convert a lead to a deal in one click, paste the checked call fields into Notes and turn each next step into a task with a Deadline.
- AI chat 24/7 that captures leads from your website
- Deals with stages and Notes; tasks with an Assignee and a Deadline
- Free for 3 users. Quotes from Launch, $19 a month billed yearly
Free plan: 3 users, no credit card.
Filled example
One clinic deal, six prompts, every output checked
How these AI prompts for sales work on one deal. Fieldfare Software, a fictional company that sells online booking software to clinics, uses them with Maya Lindqvist, practice manager at Halden Physical Therapy (3 locations). Rep Elliot Navarro runs the prompts, checks every output and keeps the results in CRMsoftware.pro.
Prompt 3B, call notes to CRM fields, Thu 8 Oct
| Field | Output | Check |
|---|---|---|
| Problem, her words | "Our front desks spend every morning on the phone moving appointments." | Word for word |
| Numbers she gave | About 60 reschedule calls a week, 3 clinics | Match |
| Who decides | Maya chooses; the owner signs anything that runs for a year | Match |
| Timeline | All 3 clinics live before Mon 4 January 2027 | Match |
| Budget | Not said | Fixed: version 1 invented 'about $500 a month' |
| Next step | Demo Tue 13 Oct, 14:00, Maya and 3 front desk leads | Match; task set |
In CRMsoftware.pro on 14 October
| Deal | Online booking · Halden Physical Therapy, $5,364, stage Proposal |
| Notes | The 6 fields from prompt 3B, checked against Elliot's notes |
| Quote | Q-1048: 3 locations at $1,788 a year, Sent, Valid Until 30 Oct |
| Checks caught | An invented budget (3B) and a go-live promise (5A), both fixed before sending |
| Task | Call with Maya and the owner, Deadline Fri 16 Oct, 9:00 |
| Never pasted into the AI | Maya's email address and any patient data |
Why check
Why every sales AI prompt needs a check
AI assistants write fast and can be wrong with confidence. OpenAI's own help center says ChatGPT may sound confident even when it is wrong, lists fabricated quotes, studies and citations among the errors, and tells users to verify quotes, data and references4.
Speed does not fix this. In a preregistered field experiment with 758 knowledge workers, people using AI finished 12.2% more tasks and worked 25.1% faster on tasks the AI handled well. On a task outside its strengths, a recommendation built from spreadsheet data and interview notes, the two AI groups were correct 60% and 70.6% of the time, against about 84.5% without AI: 19 percentage points lower5. Sales work often looks like that task.
The rules stay the same when a machine writes the draft. The FTC has said there is no AI exemption from the laws on the books6. You need a reasonable basis for a claim before you make it7. Since 2024, an FTC rule bans fake testimonials, including AI-generated ones8. And CAN-SPAM makes no exception for business-to-business email, with penalties of up to $53,088 per email that breaks it1. So every prompt here ends with rules, and every output gets the same 6 checks.
19 points
fewer correct answers with AI on a task outside its strengths (field experiment)
$53,088
maximum penalty per email that breaks CAN-SPAM (FTC guide)
6 checks
before any AI draft is sent, saved or posted (this template's rule)
By deal stage
Sales AI prompts by deal stage: what to paste, what to check
One prompt per job: what to run at each stage, what to paste in and what to check before the result goes anywhere.
| Stage | Prompt | You paste | Check first |
|---|---|---|---|
| Lead comes in | 2A Account research | Text from their public website | A source sentence for every fact |
| First touch | 2B First email | One true reason, your offer | No invented praise, customers or numbers |
| Before the call | 3A Call prep | Chat, form answers, research | No questions they already answered |
| After the call | 3B Call notes to CRM fields | Your own call notes | Quotes word for word; Not said for gaps |
| Demo | 4A Tailored demo | Demo outline and the 6 fields | Only features that work today |
| Objections | 4B Role-play | Objections from real calls | A question, a fact, one next step |
| Proposal | 5A Proposal summary | Quote lines, total, dates | Every number matches the quote |
| No reply | 5B Follow-up | Your last email, one new fact | No fake deadline; stop after two |
| Customer | 6A and 6B Upsell | Account-level usage, plan limits | Numbers from the data only |
Real estate and recruiting prompts are in tabs 7 and 8. Versions without AI: the cold email templates, sales email templates, discovery call script template and sales objection responses. Support prompts: the customer service AI prompt templates.
What never to paste
What never to paste into AI, and never to send
Six mistakes that cost trust or break the rules, in any assistant.
Personal data you may not share
Business facts, yes. Private details about people, no: patient or health information, home addresses, salaries, passwords. Your company's AI policy comes first.
Chats that may train the model
In ChatGPT, turn off 'Improve the model for everyone' under Settings > Data controls; business plans are not used for training by default9. Google asks Gemini users not to enter confidential information they would not want a reviewer to see10. Claude Free, Pro and Max users choose in Privacy Settings11; Copilot users can opt out12.
Invented personalization
'Loved your recent post', a mutual contact or a problem they never mentioned. If it is not in what you pasted, the AI made it up.
Fake testimonials or results
The FTC rule bans fake testimonials: by people who do not exist, such as AI-generated ones, or by people with no real experience of your business8. Quote real customers exactly, with permission.
Claims you cannot prove
Savings, results, 'most clinics' or 'number 1'. US law expects a reasonable basis before you make a claim7, also when AI wrote the sentence6.
Commercial email without the basics
CAN-SPAM covers commercial email, B2B included: a matching subject line, your postal address and a clear opt-out that you honor within 10 business days1.
This page is not legal advice. Rules differ by country, state and industry: check the official rules where you sell, and your broker's, MLS or firm's rules.
Excel log
The Excel prompt library and test log
Four sheets: Prompt library with the 16 prompts, Test log with 19 sample tests, Summary and Lists. Auto columns fill in by themselves.
| Column | What it does |
|---|---|
| Prompt library: ID, Stage, Prompt, Prompt text, Check first | The 16 prompts by deal stage, ready to copy, with the main check for each. |
| Current version, Owner | The version in use and who keeps it up to date. |
| Tests on this version, Passed, Last test, Status Auto | From the Test log, current version only. Status: Not tested, Keep testing, Ready (3 or more tests, all passed) or Fix and retest (any fail). |
| Test log: Date, Prompt ID, Version, Assistant, Test input | One row per run: prompt, version, assistant and what you pasted. |
| Prompt, Stage, Counts for status Auto | Filled in from the library. Rows on an older version show Old version. |
| Check result, Problem found, Used in deal, Outcome | Pass or Fail after the 6 checks, what went wrong, the deal it went into and what happened. |
| Summary Auto | Library status, pass rate, problems by type, results by stage and outcomes. |
The sample is Fieldfare Software, 29 September to 14 October 2026: 19 tests, 14 passed (74%). All 5 failures were invented facts, numbers or promises, and 4 led to a new prompt version. 2 prompts are Ready, 6 need more tests, 1 needs a fix, 7 are not tested yet. Works in Excel, LibreOffice and Google Sheets.
How to use
How to use these sales AI prompt templates
- 1
Pick the prompt for the stage
One prompt per job, not one for the whole deal.
- 2
Paste facts, not guesses
Fill in every bracket or delete the line.
- 3
Run the 6 checks
Fix the draft, never the facts.
- 4
Log the test
Pass or Fail, and the problem, in the Excel test log.
- 5
Fix the prompt, not only the output
Add a rule, save a new version, test again.
Set it up in CRMsoftware.pro
- Open My Bots, choose your bot and add your offer, prices and limits to Instructions and the Knowledge Base on the Basics tab. On the Behavior tab, turn on Collect User Data. Click Save Bot and test with Try Bot.
- Website chats land in Leads. Paste the checked research from prompt 2A into the lead's Notes. When a call is booked, click Convert to Deal.
- After each call, paste the checked fields from prompt 3B into the deal's Notes. In Tasks, click New Task: Task title, Assignee, Deadline, Linked deal.
- On Launch, use New Quote in Quotes, set Valid Until and paste the checked summary from prompt 5A into the quote's Notes.
You run the prompts in your own AI assistant and keep the checked results on the deal. Leads, 1 pipeline, Notes and basic tasks are free for 3 users. Quotes come with Launch, $19 a month billed yearly; Growth, $49, adds unlimited pipelines and the CRM AI Co-Pilot.
Weekly prompt review, 15 minutes
- Which check failed most often? Add a rule to that prompt.
- Every prompt with Fix and retest: a new version and 3 new tests.
- Spot-check 3 deal Notes: every quote is in the call notes.
FAQ
Sales AI prompts: common questions
What are the best AI prompts for sales?
The ones that give the AI your real facts and tell it what not to invent. This page has 16 by deal stage, plus real estate and recruiting, each with what to paste and what to check.
How do you write a good ChatGPT prompt for sales?
Three parts: the facts you paste, the task with a format and a word limit, and rules: use only these facts, write MISSING or Not said instead of guessing, invent no names, quotes or results. It works the same in Claude, Gemini and Microsoft Copilot.
Can AI write my sales emails?
It can draft them; you check and send them. AI assistants can be wrong with confidence and invent quotes and sources4. The email is still yours under the law: CAN-SPAM makes no exception for business-to-business email1.
Is it safe to paste customer data into ChatGPT, Claude or Gemini?
Paste only what your company allows and the task needs, never patient or other sensitive details. Check the training setting: ChatGPT under Settings > Data controls9, Claude under Privacy Settings11. Google asks Gemini users not to enter confidential information10.
How do I stop AI from making things up in sales emails?
Paste the facts, forbid everything else and check. Tell it to write MISSING or Not said instead of guessing and to quote only your notes. OpenAI itself tells users to verify quotes, data and references4.
What are good ChatGPT prompts for real estate agents?
Prompts built on the listing sheet and your showing notes: a listing description, a buyer follow-up and a fair housing language check (tab 7). Describe the home, not who should live there2.
What are good ChatGPT prompts for recruiters?
Tab 8 has a business development email to a hiring manager, candidate outreach with the pay range and a job ad language check. The EEOC's example: an ad asking for 'recent college graduates' may discourage people over 403.
Can AI summarize sales calls?
It can turn your own notes into deal fields with prompt 3B, with Not said for anything nobody said. Check the fields, then paste them into the deal's Notes.
Where do I keep AI-drafted emails and call notes in CRMsoftware.pro?
Send each email from your own mailbox after you check it, then paste the final text or the checked call fields into the deal's Notes and add a task with a Deadline for the next step. The website AI agent answers chats from your Instructions and Knowledge Base and leaves each lead in Leads, free for 3 users13.
Related templates
Free tools
Sources
- US commercial email rules, read 14 October 2026: Federal Trade Commission, CAN-SPAM Act: A Compliance Guide for Business
- eCFR, up to date as of 9 October 2026: 24 CFR 100.75, Discriminatory advertisements, statements and notices
- US job ad rules, read 14 October 2026: U.S. Equal Employment Opportunity Commission, Prohibited Employment Policies/Practices
- OpenAI Help Center, read 14 October 2026: Does ChatGPT tell the truth?
- Field experiment, 758 knowledge workers, Organization Science: Dell'Acqua et al., Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality
- FTC press release, 25 September 2024: FTC Announces Crackdown on Deceptive AI Claims and Schemes
- FTC, read 14 October 2026: Federal Trade Commission, Advertising FAQ's: A Guide for Small Business
- FTC press release, 14 August 2024: Federal Trade Commission Announces Final Rule Banning Fake Reviews and Testimonials
- ChatGPT data settings, read 14 October 2026: Data Controls FAQ
- Gemini data settings, read 14 October 2026: Gemini Apps Privacy Hub
- Claude data settings, Anthropic, 28 August 2025: Updates to Consumer Terms and Privacy Policy
- Copilot data settings, read 14 October 2026: Protecting Your AI Security, Privacy, and Data | Microsoft Copilot
- Plans and prices: CRMsoftware.pro pricing
Paste the facts. Check the draft. Keep it with the deal.
Start free with 3 users, AI chat, leads, 1 pipeline, Notes and basic tasks. Launch, $19 a month billed yearly, adds quotes and human handover; Growth, $49, unlimited pipelines.