Best Claude Prompts for Sales

Why does the AI-detection problem hit sales harder than other roles?

Sales people send more outbound messages than almost anyone. Every cold email, every follow-up, every LinkedIn touch is one more chance to trip a spam filter or a platform throttle. Three changes from 2024-2026 reshape the math.

• Gmail and Outlook spam filters have learned AI-pattern content. Google’s bulk-sender policies (enforced February 2024 and refined since) include AI-content signals as one input to spam classification. The signals are not published, but the practical effect is that a cold email composed of three balanced sentences, an opening with “I hope this finds you well,” and a closing rhetorical question now lands in spam at noticeably higher rates than the same email written by hand.
• LinkedIn now throttles AI-detected posts and comments. Per a public statement from LinkedIn’s VP of Global Editorial in late 2025, the platform is “detecting and limiting the reach of automated and AI-generated comments.” Third-party LinkedIn-marketing blogs estimate ~30% lower distribution for flagged posts. LinkedIn itself has not published a specific number, but the throttling is real and applies to InMail and DMs as well as feed posts.
• Buyers got faster at spotting AI cold outreach. The “Dear [First Name], I hope this email finds you well, I wanted to reach out because…” opener is dead. So is “I noticed you recently…[generic LinkedIn intent signal].” Buyers have seen these patterns enough times that the message gets archived without a reply.

The fix is not to stop using Claude or ChatGPT for sales drafting. The fix is to use them as a first-pass drafter and an editor, then put your own specificity, named research, and voice on top before you hit send. Our full guide on what to strip is at How to Edit AI Out of Your Writing. The full 29-pattern catalog (Wikipedia's “Signs of AI writing” + the open-source Humanizer skill from github.com/blader/humanizer) is in our cornerstone guide. High-volume sales teams should install the Humanizer skill in Claude Code or OpenCode and run it on every outbound batch. The eight prompts below are designed to produce output that already has less of the AI tells in it because the prompts themselves demand specificity.

Which prompt do you need right now?

Prompt 1: Cold outreach with a specific personalization hook

Why this works: The constraint list at the bottom is the entire trick. By naming the patterns to avoid, you force Claude to produce something that does not pattern-match on the AI-cold-email genre. The “one sentence I should personalize further” line acknowledges that AI cannot write the actually personal sentence. You have to do that part. Output is your draft minus 10 minutes.

Prompt 2: Three-touch follow-up sequence

Why this works: Most sales follow-up sequences are obvious bot output because they re-state the same value prop three times. Forcing Claude to reference “something different” each touch produces messages that sound like a person trying different angles. The “one-line reason each was written the way it was” output gives you a learning loop, not just a draft.

Prompt 3: Objection handling script

Why this works: Asking for three responses in different methodologies, plus telling you which buyer each one works on, turns Claude from a script generator into a sales-coaching tool. The “most uncomfortable truthful answer” prompt is the one most people skip writing. It is also the one that closes deals when nothing else works.

Prompt 4: Discovery call agenda

Why this works: The “real problem, not the stated one” framing is what separates a discovery call from an interview. Most sales-training prompts ask Claude for generic discovery questions. Asking for the questions that get past the surface level is the difference between booking a second call and not.

Prompt 5: Proposal first draft

Why this works: Most proposals fail because they are written for the buyer’s champion and not for the CFO who actually has to approve the spend. The 500-word cap and the “CFO has to read it in 90 seconds” line forces Claude to cut the fluff that proposals usually accumulate. The “no synergy / no alignment / all-encompassing” list strips the AI tells.

Prompt 6: Re-engaging dormant leads

Why this works: The “checking in” and “circling back” openers are AI-generated email opening #1 and #2 by frequency. The two-word-answer rule is the conversion trick: a question that costs zero effort to answer gets answered. The “rule for which hook to use” output gives you a system, not a single template.

Prompt 7: LinkedIn DM that does not feel like a pitch

Why this works: The “specific things from their activity, not their bio” rule is what separates a DM that gets read from one that gets ignored. LinkedIn’s algorithm picks up on engagement-versus-bio context. The “no CTA in message 1” rule is counterintuitive but it is what professional sales people who close on LinkedIn actually do. First message earns a reply. Second message asks for the call.

Prompt 8: Loss debrief

Why this works: Asking Claude to be candid and rank hypotheses by likelihood gets you past the polite-AI default. The two confirmation questions to email the buyer are the part most sales reps skip. The “one thing to change in the next similar deal” output turns one loss into one lesson.

How do you make sure these prompts produce sendable output?

Three rules apply to all eight prompts above.

• Add real specifics before you run the prompt. “Has 12 years at Cisco, posted about a recent migration project on LinkedIn last week” beats “experienced engineer.” The prompts are written to demand specifics. If you do not give them, the output will be generic.
• Always edit the AI tells before you send. Run the output through the editing pass in How to Edit AI Out of Your Writing, or paste it back into Claude with the pre-publish prompt from that guide. The single biggest reason cold emails get flagged as AI is the em-dash and “I hope this finds you well” opener.
• Add the one sentence that proves a human wrote this. A specific detail that only a person who actually read their stuff could write. The output of every prompt above includes a placeholder for this. Do not skip it.

What happens when you find yourself running the same prompt every Tuesday?

That is the moment to climb the ladder. A cold-outreach prompt you run by hand for every new prospect is fine. The same prompt run by hand twenty times a week is wasted typing. The upgrade is a Claude skill: same prompt saved as a file Claude calls automatically when you say “draft a cold email.” You stop typing the prompt structure. You start typing the prospect details. Output is the same; setup time is 30 seconds instead of three minutes.

What about the AI-tell editing pass?

Every output from these eight prompts gets edited before it goes out the door. Gmail’s spam filter has learned the patterns. So has LinkedIn. So has the buyer on the other end. Our pre-publish editing prompt and the manual five-pass checklist are at How to Edit AI Out of Your Writing. The single most important pass for sales output: replace ~80% of em-dashes with commas or periods, and cut every comprehensive, rock-solid, and frictionless. Those three words alone account for most of the “this looks AI-written” pattern recognition.

Frequently asked questions about sales prompts

Will my prospect figure out I used AI?

If you send the AI’s first draft, yes, probably. If you edit it well (strip the tells, add a specific personal detail, vary sentence rhythm, name something only a real reader would name), then no. The goal is not to deceive. The goal is to produce a message that sounds like you because the editing made it sound like you. Most buyers do not mind that AI was involved if the final message is good. They mind when the message is obviously generic.

Should I tell prospects I used AI?

You do not need to disclose AI assistance in sales emails any more than you need to disclose that you used spellcheck. What you do need to do is take responsibility for what you send. If a prospect calls you on something Claude got wrong, you cannot blame Claude. The signature on the email is yours.

Which AI is best for sales prompts?

Claude tends to write more naturally for sales contexts than GPT-4o because it is less prone to the “comprehensive solution” framing. ChatGPT is faster at high-volume batch drafting. Gemini integrates with Google Workspace if you live in Gmail. The candid answer is that the prompt structure matters more than the model. The eight prompts above work on all three.

My company has a no-AI policy. Now what?

Most “no AI” policies in sales orgs are aimed at preventing the bad version: sending unedited AI output, exposing customer data to third-party models, or routing pipelines through AI without compliance review. They are rarely aimed at “you cannot use Claude to help you think about an objection script.” Read your policy. Most of them let you use AI for drafting and ideation as long as no customer-confidential data is sent to a third-party model and the output is human-reviewed. If yours is stricter, follow it.

Where do these prompts come from?

They are tested versions of the eight most-used sales prompts in the larger AI Prompt Library. The Library has over 500 prompts across 33+ categories with three difficulty levels, including a Claude Code section with XML-optimized versions for sales teams that want to graduate to skills and plugins. The eight prompts in this post are the preview; the Library is the structured set.

Sources to read next?

• Google: Gmail bulk-sender requirements (2024)
• Social Media Today: LinkedIn limiting AI-content reach (Lorenzetti statement)
• Entrepreneur: LinkedIn’s fight against AI slop
• Salesforce State of Sales (annual industry data)
• Cold email (Grokipedia)
• LinkedIn Sales Solutions Blog

You may also like

• Best AI Prompts for Email Writing — the everyday inbox workhorse
• Best AI Prompts for Customer Support — the queue without burning out the team
• Best AI Prompts for Meeting Notes — decisions that survive the meeting
• How to Edit AI Out of Your Writing — strip AI patterns before you publish
• Prompt to Workflow: The AI Ladder — when to climb from prompt to workflow