PromptZone - Leading AI Community for Prompt Engineering and AI Enthusiasts

Promptzone - Community
Promptzone - Community

Posted on

15 Sales Prompts for ChatGPT & Claude (Each One Actually Run)

Most sales prompt lists are strings nobody ran. Every prompt below was run in a sandbox
against its target model before publishing, and each one passed review by three independent
editorial AI agents. The run evidence — the input fixture, the unedited model output, and
the reviewer verdicts — is on each prompt's card in our
Sales prompt directory — this page is the tour; the directory is the source.

Who this is for: anyone doing outbound or working deals — SDRs, AEs, founders selling
their own product — who wants prompts that demand real inputs and refuse to invent case
studies, urgency, or familiarity.

How to use them: paste the prompt, replace the {{placeholders}} with your real
material (the research, the thread, the deal notes). These prompts deliberately demand
inputs — outreach without your data is spam with extra steps.

1. Cold email built from real prospect research

The difference between outreach that gets replies and outreach that gets deleted is whether the research did any work. This prompt refuses to write a generic email — if your research is too thin to personalize honestly, it says so and tells you what to find.

Write a cold email to this prospect using only the research provided. Open with the single most specific, recent fact from the research and connect it to one problem we plausibly solve — no flattery, no 'I hope this finds you well'. State what we do in one plain sentence, then make one small ask (a reply or a look, not a meeting). Use only facts from the research and product description; if the research is too thin to personalize honestly, say so and list what to find instead of writing a generic email. Begin your response with the email itself: subject line first, then body. Research: {{research}} Our product: {{product}} Sender: {{sender}}
Enter fullscreen mode Exit fullscreen mode

Heavier model recommended — this one was sandbox-run on a stronger model.

See this prompt's run evidence →

2. Cold email subject lines that survive the inbox

Five honest candidates with the psychological angle and the risk each one carries — no fake 'Re:', no clickbait, no open-rate astrology.

Write five subject line candidates for this cold email. Constraints: each must be honest about what the email contains (no fake 'Re:', no fake urgency, no clickbait), short enough for a phone inbox, and specific to this prospect or their problem — a subject that could top any email to anyone fails. For each candidate, one line on the psychological angle it plays (curiosity, relevance, peer proof, directness) and the risk it carries. Then recommend one and say why. Do not predict open rates. The email: {{email}} The prospect: {{prospect}}
Enter fullscreen mode Exit fullscreen mode

See this prompt's run evidence →

3. Follow-up that adds something instead of 'bumping'

The rule that fixes most sequences: every follow-up must give the prospect something new. This one reads your thread, picks the angle the silence calls for, and tells you which angle it chose and why.

Write a follow-up email for this thread. Rule: it must give the prospect something new — a relevant resource, a sharper framing of their problem, or an easier version of the ask — never 'just bumping this' or guilt. Read the thread and pick the follow-up angle the silence most likely calls for, then state in one line (before the email, labeled 'Angle:') which angle you chose and why. Then the email. Use only facts from the thread and the material provided; do not invent new offers or urgency. Thread so far: {{thread}} Material we could share: {{material}}
Enter fullscreen mode Exit fullscreen mode

See this prompt's run evidence →

4. The breakup email that keeps the door open

The gracious exit that gets replies precisely because it isn't a guilt trip — closes the loop, leaves a usable artifact, keeps the door open.

Write the final email in this dead thread. Tone: gracious exit, zero guilt, zero passive aggression — assume they were busy, not rude. Close the loop in a way that makes replying easy if timing was the issue: state you'll stop following up, leave one concrete artifact they can use without talking to us, and one sentence on when it makes sense to reconnect. Do not manufacture scarcity or a fake deadline. Begin your response with the email, subject line first. Thread summary: {{thread}} Artifact we can leave: {{artifact}}
Enter fullscreen mode Exit fullscreen mode

See this prompt's run evidence →

5. LinkedIn connection note that isn't a pitch

No pitch, no link, no meeting ask. One specific reference to something they actually did. The only goal is an accepted connection from someone who'd recognize why you asked.

Write a LinkedIn connection request note to this person. It must fit LinkedIn's short note field, reference one specific thing they said or did (from the research), and contain no pitch, no link, and no meeting ask — the only goal is an accepted connection from a person who'd recognize why you asked. If the research contains nothing specific enough to reference honestly, say so and list two things to look for instead of writing a generic note. Begin your response with the note itself. Research: {{research}} Who I am: {{sender}}
Enter fullscreen mode Exit fullscreen mode

See this prompt's run evidence →

6. Turn raw company signals into an outreach angle

Turns raw signals — job posts, funding, tech stack, exec posts — into one committed outreach angle, ranks the evidence, and flags the signals that would feel creepy to reference.

From these raw signals about a company, extract the strongest outreach angle for what we sell. Under a heading 'Signals ranked', list each signal with a one-line read of what it implies about their current pain and priorities, ranked by how directly it connects to our product. Under 'The angle', commit to one angle: the specific tension between what they're visibly trying to do and what our product changes. Under 'Do not use', flag any signal that would feel invasive or creepy to reference in an email. Base everything strictly on the signals given. Signals: {{signals}} Our product: {{product}}
Enter fullscreen mode Exit fullscreen mode

Heavier model recommended — this one was sandbox-run on a stronger model.

See this prompt's run evidence →

7. Reply to a warm 'tell me more' without overplaying it

A warm 'tell me more' dies fastest under a feature dump and a calendar link. This response answers their actual question first and sizes the next step to their energy level.

A prospect replied with mild interest. Write the response that converts interest into a concrete next step without overwhelming them: answer their actual question first in plain language, then offer exactly one next step sized to their energy level (a curious one-liner gets a resource; a detailed question gets a short call offer). Read their reply carefully for what they did NOT ask — do not dump features they showed no interest in. Use only facts from the material provided. Begin your response with the email. Their reply: {{reply}} Original thread: {{thread}} Product facts: {{facts}}
Enter fullscreen mode Exit fullscreen mode

See this prompt's run evidence →

8. Plan a multi-touch sequence that doesn't harass

A full sequence plan — channel, day offset, and the job of every touch — with kill criteria for the reply signals that should end it immediately.

Design an outreach sequence for this prospect segment: for each touch give the channel, the day offset, the job that touch does (labeled 'Job:'), and a two-line sketch of its content — not full copy. Constraints: every touch must carry something new (no 'bumping'); mix channels only where the segment plausibly lives; the sequence ends with a clean breakup, not a fade-out. Then, under 'Kill criteria', state the reply signals that should immediately end or redirect the sequence. Ground everything in the segment description — where it's too vague to choose a channel or angle, say so explicitly. Segment: {{segment}} Product: {{product}} Assets we have: {{assets}}
Enter fullscreen mode Exit fullscreen mode

Heavier model recommended — this one was sandbox-run on a stronger model.

See this prompt's run evidence →

9. Diagnose a sequence's reply rate from the copy

Feed it your underperforming sequence and it names the reply-killer in each touch with the exact quoted phrase as evidence — and tells you when the problem is your list, not your copy.

This sequence underperforms. Diagnose why from the copy alone: for each touch, name the single biggest reply-killer (seller-centric opening, oversized ask, fake personalization, no new value, guilt) with the exact phrase as evidence, quoted. Then, under 'Rewrite priorities', rank the three changes that would most plausibly move replies, each tied to the evidence — not generic best practices. Judge only what's in the copy; where the problem could be targeting or deliverability instead of copy, say so explicitly rather than inventing a copy fix. Sequence: {{sequence}} Audience: {{audience}}
Enter fullscreen mode Exit fullscreen mode

Heavier model recommended — this one was sandbox-run on a stronger model.

See this prompt's run evidence →

10. Discovery questions that surface the real problem

Discovery questions built from what you actually know going in — never a generic BANT list, never a question whose answer is already in your notes.

Build a discovery question plan for this call from the context given. Structure: under 'Must learn', the three unknowns that most determine whether this deal is real (from the context — not a generic BANT list); under 'Questions', an ordered sequence where each question earns the next, moving from their world to the problem to consequences — no question answerable by yes/no unless it's a deliberate commitment check, and no question whose answer is already in the context (asking those burns credibility); under 'Listen for', the specific answers that would qualify or disqualify. Context: {{context}} What we sell: {{product}}
Enter fullscreen mode Exit fullscreen mode

Heavier model recommended — this one was sandbox-run on a stronger model.

See this prompt's run evidence →

11. Answer an objection with only what's true

Acknowledge the legitimate core, answer with only verified facts, and end with a step the prospect can verify themselves. If your facts don't answer the objection, it says so instead of papering over it.

Craft a response to this objection using only the facts provided. Structure: acknowledge the legitimate core of the objection in one sentence (find it — every objection has one); answer with specific evidence from the facts, clearly separating what we know from what varies by customer; end with a verification step the prospect can run themselves (a pilot, a reference call, a document) rather than asking them to take our word. If the facts don't actually answer the objection, say that plainly and state what we'd need to provide instead — never paper over a gap. Objection: {{objection}} Facts: {{facts}}
Enter fullscreen mode Exit fullscreen mode

See this prompt's run evidence →

12. Decode what 'too expensive' actually means

'Too expensive' means four different things. This prompt commits to a diagnosis from your deal context — and each diagnosis gets a different move. It never opens with a discount.

The prospect said it's too expensive. From the deal context, diagnose which kind of price objection this most likely is — no budget exists, budget exists but value unproven, cheaper competitor anchoring, or a negotiation opener — and commit to one diagnosis with the evidence from the context (labeled 'Diagnosis:'). Then write the response for THAT diagnosis: each type gets a different move (qualify out honestly, build a value case on their numbers, reframe the comparison, or hold with a concession path). Never respond with an unprompted discount. If the context can't distinguish between two diagnoses, give the one question that would. Context: {{context}}
Enter fullscreen mode Exit fullscreen mode

Heavier model recommended — this one was sandbox-run on a stronger model.

See this prompt's run evidence →

13. Handle 'we're also looking at X' without trashing X

Handles 'we're also looking at X' the way that survives being forwarded to X: respects what they're good at, draws the real dividing line, and gives your prospect one evaluation question you win on merit.

The prospect is comparing us to a named competitor. Write the response: respect the competitor honestly (name one thing they're genuinely good at, from the facts), then draw the real dividing line — the difference in approach or fit that decides which buyer should pick which product — and land on which side THIS prospect's situation falls, using their context. Give them one sharp evaluation question to ask both vendors that we win on merit. Use only the provided facts about both products; where competitor facts are missing, say 'verify with them' rather than guessing. Their statement: {{statement}} Our facts: {{ours}} Competitor facts: {{theirs}} Prospect context: {{context}}
Enter fullscreen mode Exit fullscreen mode

Heavier model recommended — this one was sandbox-run on a stronger model.

See this prompt's run evidence →

14. Arm your champion for the internal fight

Your champion sells this deal in rooms you'll never enter. This brief arms them: the hallway one-liner, arguments matched per stakeholder, and honest answers to the pushback they'll hit.

Your champion must sell this internally. From the deal context, write their ammunition: under 'The one-liner', how they should describe the purchase to an exec in a hallway; under 'The case', three arguments matched to what each named stakeholder cares about (from the context — finance argues differently than ops); under 'The objections they'll hit', the two most likely internal pushbacks with honest answers the champion can own; under 'What to ask for', the specific decision or meeting the champion should request. Everything grounded in the context — never invent ROI numbers; where a number is needed but missing, mark it '[their number needed]'. Context: {{context}}
Enter fullscreen mode Exit fullscreen mode

Heavier model recommended — this one was sandbox-run on a stronger model.

See this prompt's run evidence →

15. Diagnose why a deal went quiet

The deal went quiet. This commits to a diagnosis from the timeline evidence, names the alternative, and prescribes one re-engagement move that produces information whether or not they reply.

This deal stalled. From the timeline, diagnose the most likely cause — champion lost power, priority shifted, a stakeholder objected off-stage, budget cycle, or we were the free education for a decision already made — and commit to one primary diagnosis with the timeline evidence (labeled 'Diagnosis:'), plus the strongest alternative and what would distinguish them. Then prescribe the single re-engagement move that tests your diagnosis without begging — a move that produces information whether or not they reply. If the timeline honestly supports 'this deal is dead', say so and prescribe the graceful close instead. Timeline: {{timeline}}
Enter fullscreen mode Exit fullscreen mode

Heavier model recommended — this one was sandbox-run on a stronger model.

See this prompt's run evidence →

FAQ

Are these prompts free to use?
Yes. Copy them, adapt them, ship them. Attribution is appreciated, never required.

What does "actually run" mean?
Before publishing, each prompt was executed in a sandbox against its target model with a
realistic input fixture, and the output was checked by three independent editorial AI agents
— accuracy, safety, usefulness — from three different model families. Prompts that failed
did not get published.

Will these get my emails marked as spam?
The prompts can't fix deliverability — but they're built on the one principle that helps:
every email must be specific enough that it could only have been written to that one
person. Generic AI outreach is what filters (and humans) delete.

Which model should I use?
Any current frontier model handles most of these. The ones marked heavier model recommended
involve multi-document judgment. Not sure which model fits?
Our which-LLM picker takes two minutes.


All 30 Sales prompts, with run evidence and review verdicts, live in the
PromptZone prompt directory. New evidence-gated categories ship every few
weeks.

Top comments (0)