---
title: "25 ChatGPT Prompts for Instagram DMs That Book Calls"
slug: chatgpt-prompts-for-instagram-dms
author: "Leonardo Maldonado"
category: "AI Setter Operations"
articleType: listicle
tags: ["Instagram DM prompts","AI DM scripts for coaches"]
publishedAt: 2026-10-08T16:00:00.000Z
updatedAt: 2026-10-08T16:00:00.000Z
canonical: https://setluca.com/blog/chatgpt-prompts-for-instagram-dms
---# 25 ChatGPT Prompts for Instagram DMs That Book Calls

> ChatGPT prompts for Instagram DMs work as a drafting tool, not a sending tool. You paste the thread and your offer, it gives you a message, you edit it and send it yourself. The 25 prompts below are named and grouped by the job they do: research the lead, open, qualify, handle an objection, re-engage a ghost, and ask for the call.

## Key takeaways

- ChatGPT drafts, you send. There is no path from a chat window to a sent Instagram DM without going through Instagram's Messaging API and an approved app.
- Every prompt needs a context block above it. Without your offer, your fit criteria and your real phrasing, the output reads like a template because it is one.
- Feed it their words, not your summary. That single change produces the biggest quality jump of anything here.
- The library breaks down at volume. Nothing in a chat window knows who is owed a follow-up on Thursday, and nothing stops you sending at 2am.
- Five prompts do most of the work: the thread recap, the next-question picker, the objection namer, the day-three nudge, and the two-slot ask.

---

## What can ChatGPT prompts for Instagram DMs actually do?

ChatGPT prompts for Instagram DMs write the message. That is a real job, because most coaches lose leads to a blank reply box rather than to a bad reply. Paste a thread, get a draft, edit it in ten seconds, send it yourself.

What they cannot do is touch your inbox. Sending a DM programmatically runs through Instagram's Messaging API, which needs a professional account and an app holding messaging permissions, per Meta's platform docs. A chat window has none of that, so what you have is a writing assistant with your thumbs attached. That's a different animal from [AI DM automation that replies on its own](/blog/does-ai-dm-automation-actually-work).

The clock is unchanged either way. Meta's messaging policy runs on a 24-hour window that opens when someone messages your business, with a Human Agent option that extends manual replies to seven days. Drafting help doesn't buy you time. It helps you use the day you had.

## How do you set ChatGPT up so the prompts work?

Load your context once at the account level so you stop re-explaining your business in every chat. Custom instructions hold up to 1,500 characters on Free and Go, and 5,000 on Plus and above, per OpenAI's documentation. Put your offer, price, fit criteria and most-used phrases there. Nothing else.

Then make a Project for DM work. Projects carry their own instructions and files across every chat inside them, per OpenAI's Projects documentation, so upload your sales page, your fit criteria and twenty of your own real DMs, because the model needs something of yours to imitate. [Training the AI to reply in your voice](/blog/how-to-train-ai-to-reply-in-your-voice) is worth doing before you paste a single prompt.

` CONTEXT -- paste above any prompt. I coach [who] to [outcome] in [timeframe]. Offer: [name], [length], [price]. Fit: [3 criteria]. Not a fit: [2 disqualifiers]. How I write: short sentences, lowercase, no exclamation marks. Phrases I use: [3]. Never say: [your banned words]. Booking link: [link]. Timezone: [tz]. Rules: my voice, one question per message, under 40 words unless I say otherwise. `

## What does one of these prompts look like end to end?

One paste, one prompt, one draft, two edits. That is the whole loop, and seeing it once is worth more than reading twenty prompts you've never run. The thread below is an illustrative composite, assembled for this post from the shape these conversations usually take. It is not a real lead's messages, and yours shouldn't leave your inbox either: a lead's DMs are their words, not content.

### The thread you paste in

` LEAD (Tue 9:14pm): saw your story about deload weeks. do you work with people who lift but keep stalling ME (Tue 9:31pm): I do. Is it the same weights for a few weeks, or up and down? LEAD (Wed 7:02am): same weights since june. i train 4x and my food is honestly all over the place. i've done two programs off instagram before and stopped following both after about 3 weeks `

Under your context block, that goes in with prompt 10, the next-question picker. Paste the thread oldest message first, both sides, and your fit criteria underneath it. Nothing else. The model has no access to the profile, the story they replied to, or anything you didn't paste.

### The draft it hands back

` Best next question: "Totally hear you on the plateau! Before we go further, how consistent would you say your nutrition is across a typical week, and what's your current protein target?"

Why: consistency is the variable they named themselves, and the answer decides whether this is a programming problem or an adherence problem.

Don't ask yet: whether they're ready to invest in coaching. They've told you two past programs didn't stick, so a money question here reads as a pitch. `

## What would you change before sending it?

Two things change before a draft goes out, and they're the same two most weeks. The reasoning and the "don't ask yet" line are usually right. The message itself always needs a trim, because a model writing a DM writes an email.

### Cut the greeting, then put their words back

**Edit one: cut the greeting and the second question.** "Totally hear you on the plateau!" is not how anyone types at 7am. The draft also asks two things at once, so you'd get half an answer back. Protein target is a coaching question, not a qualifying one. Keep the question that decides something.

**Edit two: put their words back.** They wrote "all over the place." The draft upgraded that to "consistent" and "nutrition," which is your vocabulary, not theirs. Swapping it back is what makes a reply sound like it was written by someone who read the message. That's prompt 4 doing its job inside a different prompt.

### The message that actually goes out

What gets sent looks like this, in lowercase because that's what the context block says:

> is the food all over the place most days, or mainly weekends?

Twelve words, one question, their language. The reasoning line did the other half of the work, because it told you what not to ask yet. Use that whole loop as your calibration. If a first draft needs more than two edits, the gap is in your context block rather than the prompt. Something about your voice, your offer or your fit criteria never made it into the window. Fix the input and the same prompt starts landing.

## Prompts 1-4: research the lead before you type

### 1. The Profile Read

` Read this Instagram bio and these three captions. Give me three facts about this person, the problem they'd admit to out loud, and one thing I should not assume. `

**When:** before any opener to someone you don't know. **Paste in:** the bio, three recent captions, follower count. **Good output:** three specifics you could quote back, plus one flagged guess.

### 2. The Thread Recap

` Here is a DM thread. In five lines: what they want, what they've told me, what I've asked, what is still unanswered, and the single next move. `

**When:** any thread you left sitting more than two days. **Paste in:** the full thread, oldest message first, both sides. **Good output:** it names the unanswered question, not a polite summary.

### 3. The Signal Sort

` Rank these Instagram handles by how close each is to a sales conversation. Use only my notes. For each, say what would move it up one rank. `

**When:** Monday triage, when forty story replies are waiting. **Paste in:** one line of notes per handle. **Good output:** an order you'd defend, with a reason on each. **Fails when:** your notes all say the same thing. If every line reads "asked about price," the ranking is guesswork dressed as a ranking. Go back to the threads, write one specific detail per handle, then run it again.

### 4. The Vocabulary Mirror

` Pull the exact phrases this person used for their problem and their goal. List them verbatim. Do not paraphrase, tidy, or upgrade the language. `

**When:** before you write anything they will read. **Paste in:** their messages only, never yours. **Good output:** their words, typos included, ready to reuse.

## Prompts 5-9: open the conversation

### 5. The Story-Reply Opener

` They replied to my story about [topic] with the message below. Write three one-line replies that ask about them, mention no offer, and can be answered in four words. `

**When:** inside the hour, while the story is still live. **Paste in:** the story topic and their exact reply. **Good output:** short, curious. If one opens with "Love that!", regenerate.

### 6. The Comment Carry-Over

` This person commented "[comment]" on my post about [topic]. Write a DM continuing that specific thought, ending in a question. Under 30 words. No greeting stack. `

**When:** moving a public comment into a private thread. **Paste in:** the comment, the post topic, their handle. **Good output:** a message that could only be sent to this one person.

### 7. The Warm Re-Intro

` This person followed me [time] ago and never messaged. Write an opener that guesses why they followed and asks one question. Don't mention that time has passed. `

**When:** mining existing followers instead of buying reach. **Paste in:** their bio and what you were posting back then. **Good output:** no apology, no "I know this is random."

### 8. The Freebie Follow-On

` They downloaded [lead magnet]. Write a message asking which part they'll try first. No pitch, no link, one question, under 25 words. `

**When:** a day or two after the download, not the same minute. **Paste in:** the lead magnet name and its three main sections. **Good output:** a question about them, not a survey about your PDF.

### 9. The Three-Register Split

` Write the same opener three ways: plain, warm, blunt. Same information in all three. Number them. Do not tell me which you prefer. `

**When:** you can't tell whether your instinct is right. **Paste in:** the situation in one sentence. **Good output:** three different messages, not three synonyms. Judge them against [Instagram DM openers that get replies](/blog/instagram-dm-openers-that-get-replies). **Fails when:** all three are polite. Blunt should read a little cold to you and still be sendable. If the third one is just the first with the pleasantries shaved off, add "the blunt version should risk sounding curt" and run it again.

## Prompts 10-13: qualify without interrogating

### 10. The Next-Question Picker

` Given this thread, what is the single best next question? Give me the question, one sentence on why, and the question I should not ask yet. `

**When:** every time you're tempted to send two questions at once. **Paste in:** the thread and your fit criteria. **Good output:** one question under fifteen words, plus a clear "not yet."

### 11. The Qualifier Ladder

` Turn my fit criteria into four qualifying questions, ordered easiest to hardest to answer. Each under 15 words, each answerable one-handed on a phone. `

**When:** once, when you build your process. Then you reuse the ladder. **Paste in:** your fit criteria and your disqualifiers. **Good output:** questions you'd actually ask out loud.

### 12. The Answer Grader

` Score this lead against my fit criteria: fit, not a fit, or unclear. Quote the line that decided it. If unclear, give me the one question that resolves it. `

**When:** before you spend another twenty minutes on a thread. **Paste in:** the thread and the criteria, verbatim. **Good output:** a quoted line. No quote means it guessed.

### 13. The Budget Bridge

` Write a message that surfaces whether [price] is realistic for this person, without naming a number and without asking what they can afford. `

**When:** the thread is warm and you don't want to waste a call. **Paste in:** the thread and your price. **Good output:** an indirect question about what they've invested in before.

## Prompts 14-18: handle the objection

### 14. The Objection Namer

` Which objection is this actually: price, timing, trust, or authority? Quote the words that decided it, then name the second most likely one. `

**When:** first, always, before you write a reply. **Paste in:** their objection and the two messages before it. **Good output:** a named objection with evidence. Most bad replies answer the wrong one, which is why [objection handling in the DMs](/blog/objection-handling-in-dms) starts here.

### 15. The Price Reply

` They pushed back on price with the message below. Write a reply that asks one clarifying question before defending anything. No discount, under 40 words. `

**When:** the number landed badly. **Paste in:** their exact wording and your price. **Good output:** a question, not a paragraph about value. **Fails when:** the draft explains what's included before it asks anything. A reply that defends the price has already answered an objection nobody identified, and price is rarely the one underneath. Regenerate with "the first sentence must be a question."

### 16. The Timing Reply

` They said they're too busy right now. Write a reply that treats that as true, asks what would have to change, and offers no solution in this message. `

**When:** "after the holidays," "once work calms down," "next quarter." **Paste in:** their message and their stated goal. **Good output:** no urgency tricks, no invented deadline.

### 17. The Burned-Before Reply

` They've tried programs like mine and it didn't work. Write a reply asking what specifically broke last time. Do not defend or describe my program here. `

**When:** skepticism arrives as a story about the last coach. **Paste in:** their message. **Good output:** curiosity about their history, zero rebuttal.

### 18. The Decision-Maker Reply

` They need to check with a partner. Write a reply that asks what their partner's first question will be. Nothing else in the message. `

**When:** a second person enters the thread. **Paste in:** their message and your price. **Good output:** one question. The longer play is in [the spouse objection](/blog/i-need-to-talk-to-my-spouse-objection).

## Prompts 19-22: re-engage a ghost

### 19. The Ghost Diagnosis

` This thread went quiet after my message below. Give the most likely reason from what is on screen, and say whether a follow-up helps or hurts. "Leave it" is a valid answer. `

**When:** day three of silence, before you write anything. **Paste in:** the last four messages and how long it's been. **Good output:** it sometimes tells you to stop. Believe it.

### 20. The Day-Three Nudge

` Write a follow-up that gives them something they don't have yet and asks nothing. Under 25 words. No "just checking in." No question mark. `

**When:** three days after your last message went unanswered. **Paste in:** their goal and one resource you actually have. **Good output:** a message that costs them nothing to ignore.

### 21. The Reopen

` Ten days of silence. Write one message that reopens on their original goal, references something specific they said, and gives them an easy way back in. `

**When:** day ten to fourteen, once only. **Paste in:** the whole thread, so it can find the line. **Good output:** it quotes them. If not, run prompt 4 first, then reread [the re-engagement DM for cold leads](/blog/re-engagement-dm-for-cold-leads).

### 22. The Clean Close

` Write a final message that gives them permission to say no, closes the thread without guilt, and leaves the door open. Under 30 words. `

**When:** after two unanswered follow-ups. Then you stop. **Paste in:** the thread. **Good output:** no guilt, no "last chance," no sad emoji. **Fails when:** it reads as one more nudge wearing a goodbye. If the message ends in a question mark or points at a future check-in, it isn't a close. Ask for a version with no question and no future tense, then send that.

## Prompts 23-25: write the booking message

### 23. The Two-Slot Ask

` Write the message that asks for the call. Offer two specific times in [their timezone], name what we'll cover in one clause, under 35 words. `

**When:** they've answered your qualifying questions and nothing is blocking. **Paste in:** your two real openings and their timezone. **Good output:** two times, one clause, no calendar link yet.

### 24. The Link Handoff

` They said yes. Write the message that sends my booking link, says how long the call is, and what to have ready. One link. No numbered instructions. `

**When:** immediately after the yes, in the same minute. **Paste in:** your link and the call length. **Good output:** one sentence and one link, so the friction stays low.

### 25. The Show-Up Insurance

` Write a confirmation for a call booked [n] days out. Confirm the time in their timezone, ask one question they'll answer before the call, invite a reschedule. `

**When:** right after booking, and again the day before. **Paste in:** the booked time, their timezone, your pre-call question. **Good output:** a reply-worthy message. A lead who replies to a confirmation shows up.

## Where does copy-pasting prompts break down?

Copy-pasting prompts breaks down at volume, and sooner than most coaches expect. The prompts themselves are fine. What fails is the workflow around them, in four ways, and each one gets worse as the inbox fills.

### What the drafting loop costs at volume

Context switching is the first cost. Every draft means an app switch, a paste, a read, an edit, and a switch back. Call it ninety seconds on a simple thread, repeated on every message, all day.

Tone drift is the second. A fresh chat has no memory of the last message you sent that lead, and saved memories hold facts you asked it to keep rather than the rhythm of one thread (OpenAI's memory FAQ), so by message four you sound like three different people.

The third is pacing, or the absence of it. A chat window will hand you nine openers at 1am, and Instagram treats a burst of new conversations very differently from replies inside a thread someone opened. [Instagram's DM automation rules](/blog/instagram-dm-automation-rules-2026) draw that line. No prompt knows about them.

### Who reviews the draft before it sends?

You do, and that is the fourth gap. Being the review step yourself works at ten threads and quietly stops working at eighty.

The review gap is why Luca is built the way it is: replies only, at human pace, inside the window a lead opened, with review as an optional switch on any channel if you want to approve or edit before a reply goes out. That option exists because the failure mode in this category is a well-written message landing on the wrong person at the wrong hour with nobody reading it first. [The complete guide to AI setters for coaches](/blog/what-is-an-ai-dm-setter) covers that arrangement.

## Where does a good human setter still beat both?

A good human setter still wins on the thread that changes shape halfway through. A lead spends four messages asking about your program, then mentions they're moving house in three weeks. A good human hears the real timeline and quietly changes the goal from "book this week" to "book in a month, and remember why." Neither a prompt nor a setter catches that reliably, because it was never stated as a constraint.

Unstated constraints are also why the rescue cadence in Luca stops instead of escalating. Software that can't hear reluctance shouldn't be trusted to push through it, which is why the cadence ends rather than getting cleverer about the push. That ceiling is deliberate. [AI lead qualification in the DMs](/blog/how-ai-qualifies-leads-in-dms) is that boundary drawn in practice.

## A worked example: Priya's Tuesday inbox

Priya coaches strength athletes online, and her offer runs $1,800 for twelve weeks. She is a composite built to show the arithmetic, not a customer of ours, and the week below is illustrative: 42 inbound DMs, nearly all story replies, arriving between 6am and midnight.

She used the library on 18 threads, three drafts each. That's 54 drafts. At roughly 90 seconds per draft including the app switch, she spent 81 minutes on writing help that week, on top of reading and sending. The other 24 threads were one-liners she answered herself.

Then she changed one thing. On Sunday she ran prompts 5, 9 and 11 once, generated six openers and her qualifier ladder, and saved them as phone text replacements. The next week she opened a chat window eleven times instead of 54: about 17 minutes, same quality.

What didn't move was reply speed on the 6am and 11pm messages. Those still waited for Priya, which is the whole argument in [DM response time and speed to lead](/blog/dm-response-time-speed-to-lead).

## Five mistakes that make ChatGPT-written DMs flop

**Pasting your summary instead of their words.** "She wants to lose weight" produces a generic message, while her own sentence, typos and all, produces one she recognises as being about her. Prompt 4 exists for this.

**Sending the first draft.** The first output runs long and one register too warm. Cut the greeting, cut the closing line, send the middle.

**Asking two questions in one message.** Models pad. Cap the question count in the prompt or you'll get a paragraph ending in a double question, and half an answer.

**Batch-generating openers for cold strangers.** This is where a drafting habit turns into a volume habit, and volume outbound is what gets accounts limited.

**Letting it invent details.** A model will happily claim your program includes weekly calls when it doesn't. Read any draft mentioning a feature, a price or a date twice.

## Why doesn't the output sound like you?

| Symptom | Likely cause | Fix |
| --- | --- | --- |
| Every message opens with "Love that!" | No banned-words line in your context | Add "Never say:" with your five worst offenders |
| Drafts twice as long as your real DMs | No word cap in the prompt | Put the cap in the prompt itself |
| It keeps pitching in the opener | Your offer is in context, nothing forbids using it | Add "mention no offer" to opener prompts |
| Replies contradict what you said two messages ago | Fresh chat with no thread history | Paste the full thread, oldest first, every time |
| Output is fine but generic | You described the lead instead of quoting them | Run prompt 4, then rerun the original prompt |
| It invents a feature you don't offer | Uploaded sales page is stale or missing | Re-upload it, add "use only this document for offer details" |

## Which edge cases break a prompt library?

**The 2am reply.** Someone answers your qualifying question at 2am their time. The prompt writes a fine reply, and sending it at 2:04am is what tells them a machine is on the other end. Draft it, send at 8am, and treat the 24-hour window as a deadline rather than a target.

**The lead who already bought.** A current client messages from a new angle and lands in the same inbox. A prompt that doesn't know will qualify them from scratch. Add a line to your context block: "if the thread suggests they're already a client, say so and stop."

**The wrong-timezone booking.** Prompt 23 will confidently offer "Tuesday 2pm" without knowing whose Tuesday. Paste the lead's timezone, and ask for their city first if you don't have it.

**The thread with a seven-month gap.** A conversation that ran hot in March and revived in October needs both halves pasted and a note that time passed, or it replies as though yesterday.

## Should you use ChatGPT, Instagram's native tools, or an AI setter?

Choose by volume first. Then ask which part of the job actually hurts: writing the message, or being there when it arrives.

|  | ChatGPT prompts | Instagram's native tools | AI setter | Human setter |
| --- | --- | --- | --- | --- |
| Who sends | You | Instagram, on triggers | Software, automatically (review optional) | The setter |
| Best at | Wording | Instant replies, FAQs, away messages | Consistent qualifying and follow-up | Hard threads |
| Timing control | None | Trigger-based only | Reply-only, at human pace | Human |
| Thread memory | Only what you paste | None | Full thread | Full thread |
| Typical cost | $0, or $20/mo for Plus | Included | $99-$499/mo with Luca | Salary or commission |
| Breaks down at | Roughly 30+ threads a week | Anything conversational | Threads needing real judgment | Cost, at volume |

Pick the prompts if your problem is not knowing what to say, your volume sits under about thirty threads a week, and you're online enough to answer inside the day. Pick Instagram's native replies if the same three questions arrive constantly and the answers never change. Pick an AI setter if messages sit unanswered while you're coaching and you want qualifying done the same way every time. Pick a human when the threads go somewhere unpredictable and the offer is big enough to pay for one. Running the first two together is the cheapest sensible start. [How to set up an AI setter](/blog/how-to-set-up-an-ai-setter) is the step after that.

## The verdict

ChatGPT prompts for Instagram DMs solve a writing problem, and they solve it well. Give the model your offer, your fit criteria and the lead's own words, and you'll get a better message than the one you'd have typed while distracted between clients. Name them, keep five within reach, and the blank reply box stops costing you leads.

What they don't solve is presence. Nobody drafts at 6am on a Sunday, nothing in a chat window remembers that three leads are owed a follow-up on Thursday, and no prompt stops you sending a burst of openers on a night your account can't afford one.


## FAQ

### Can ChatGPT reply to my Instagram DMs automatically?

No. ChatGPT writes drafts you copy and send yourself. Automated sending runs through Instagram's Messaging API, which needs an Instagram professional account and an app approved for messaging permissions, per Meta's Instagram Platform overview. A standard chat window has no connection to your inbox and cannot see or send your messages.

### Is it against Instagram's rules to use ChatGPT to write my DMs?

Writing a message with AI help and sending it yourself is not what Meta's messaging policy restricts. The policy governs automated experiences and the 24-hour window that opens when someone messages you. Unsolicited outbound volume is the real risk, whoever or whatever wrote the words.

### Do I have to tell people an AI helped write my DM?

If you write and send it yourself, you're the sender. If an automated experience is replying, Meta's messaging policy asks you to disclose it where law requires, including for California and German users, and recommends disclosure as good practice anyway. Never deny AI involvement when someone asks you directly.

### Which ChatGPT plan do I need for these prompts?

The free plan runs every prompt here. Plus, at $20 a month, raises custom instructions from 1,500 to 5,000 characters and project files from 5 to 25, per OpenAI's help center, which matters if you want your sales page, fit criteria and voice samples all loaded at once. Start free.

### How do I stop the drafts from sounding like a bot?

Paste the lead's exact words rather than your summary, cap the length inside the prompt, and keep a "never say" list in your context block. Then cut the greeting and the closing line from every draft. The middle of the message is usually the only part worth sending.

### How many of these 25 prompts will I actually use?

Five, most weeks. The thread recap, the next-question picker, the objection namer, the day-three nudge and the two-slot ask cover the daily work. The other twenty earn their place in specific situations, like a ten-day ghost or a lead who mentions a partner.

### Can ChatGPT remember my offer between conversations?

Partly. Saved memories keep facts you ask it to keep, and a Project holds files and its own instructions across every chat inside it. Neither remembers the rhythm of one particular DM thread, so paste the thread each time rather than trusting recall.

### What's the fastest way to use this library day to day?

Run the batch prompts once on a Sunday. Generate your openers, your qualifier ladder and your follow-up variants, then save them as phone text replacements. Open a chat window only for genuinely new situations. This cuts most of the app switching that makes prompt libraries feel slow.


## Sources

1. [Meta for Developers. "Messenger Platform and IG Messaging API Policy" (24-hour standard messaging window, Human Agent 7-day window, bot disclosure guidance).](https://developers.facebook.com/documentation/business-messaging/messenger-platform/policy)
2. [Meta for Developers. "Instagram Platform Overview" (professional account requirement, messaging permissions, messaging rate limits).](https://developers.facebook.com/docs/instagram-platform/overview/)
3. [OpenAI Help Center. "ChatGPT Custom Instructions" (1,500- and 5,000-character limits).](https://help.openai.com/en/articles/8096356-chatgpt-custom-instructions)
4. [OpenAI Help Center. "Projects in ChatGPT" (project instructions and per-plan file limits).](https://help.openai.com/en/articles/10169521-projects-in-chatgpt)
5. [OpenAI Help Center. "Memory FAQ: How does reference saved memories work?"](https://help.openai.com/en/articles/11146739-how-does-reference-saved-memories-work)
6. [OpenAI Help Center. "What is ChatGPT Plus?" ($20/month).](https://help.openai.com/en/articles/6950777-what-is-chatgpt-plus)
7. [OpenAI. "Usage Policies" (prohibition on deceit, fraud, scams, spam and impersonation).](https://openai.com/policies/usage-policies/)

---

Published by SetLuca, the company behind Luca.