Write cold DMs people actually answer

This prompt turns your AI tool into an outbound coach. You give it a few facts about your offer and your buyer, and it hands back a complete outreach system: a research checklist, three short messages, alternate versions, a list of phrases that kill replies, and a scoring sheet you run your drafts through. It works for LinkedIn, email, X or Instagram. Use it when your current messages get opened but ignored, or when you are starting outreach from zero and want a repeatable process instead of a blank page.

What you get

A research checklist of seven things to look up on every prospect, with where to find each one and what to look for. A three-message sequence: a no-pitch opener, a value-only follow-up, and a soft ask. Three full variants for different buyer types in your niche. Ten rules to keep, a list of banned words, and a 1 to 5 rubric covering personalization, value, clarity and next step.

How to use it

1. Fill in the bracketed slots. Swap in your product, your price, a one-line description of your ideal buyer, and the channel you use. Be specific about the buyer. "Marketing leads at 20 to 50 person software firms" beats "marketers."
2. Paste the whole prompt into your AI tool in one go. Do not trim the research section, since that part drives the personalization later.
3. Read the research checklist first and correct it. If it suggests checking a data source you cannot access, say so and ask for a substitute.
4. Run the sequence on one real prospect. Pick a person you know something about and ask the model to apply the framework to them, using the details you supply.
5. Score the output with the rubric it gave you. Anything under 4 on personalization goes back for a rewrite.
6. Save the rules and banned-phrase list somewhere you will see them before every send.

What a good answer looks like

The opener should read like one person who noticed something, not a template. It should be under 50 words, name one specific detail about the prospect, and end with a question a human would actually answer. Message two should give something away with no ask attached. Message three should name a problem you found and offer one clear next step. The variants should feel genuinely different from each other, not the same message with the nouns swapped. If the checklist tells you to find a mutual connection but not where to look, or the banned-phrase list is just "avoid jargon," ask for a sharper pass. Then edit the drafts in your own voice before sending. The prompt gives you structure; the last line still has to sound like you.