How to Write Instagram Captions With AI (Without Sounding Like a Robot)
You've tried it. You open ChatGPT, type "write me an Instagram caption about my new product," and what comes back is fine. Technically a caption. Grammatically correct. Completely forgettable.
It sounds like it was written by a brand, for no one, about nothing. So you rewrite it yourself and wonder why you bothered opening the AI at all.
Here's the part most people miss: the AI isn't the problem. The prompt is.
When you ask for "an Instagram caption," you've handed the AI almost nothing to work with. No audience. No voice. No point of view. So it does the only thing it can — it gives you the average of every caption on the internet. Bland by design.
The fix is a framework. Four parts. Use all four every time, and your captions stop sounding like a robot and start sounding like you.
The four parts of a caption prompt
Every good AI prompt — for captions or anything else — has the same four ingredients.
Role. Who you want the AI to be when it writes. "A thoughtful small-business owner who writes like a friend" produces something very different from "a social media marketer." The role sets the tone before a single word is written.
Task. What you actually want. "Write a caption" is weak. "Write an Instagram caption introducing my new digital planner" is strong. Be specific about the job.
Context. This is the part that does the real work. Who your audience actually is. What your voice sounds like. What's been said too many times in your niche already. The words you're sick of seeing. Context is the difference between a caption that sounds like everyone and one that sounds like you.
Format. The shape of the output. How many paragraphs. Where the hook goes. How it ends. Tell the AI what you want or it'll guess, and it usually guesses long.
Three of those four are mechanical — you can learn them in an afternoon. Context is the one that takes a little thought, and it's the one that changes everything.
Watch the difference
Here's the same product, the same AI, two different prompts.
The weak prompt:
Write an Instagram caption about my new digital planner.
What you get back:
Stay organised and boost your productivity with our brand-new digital planner! Perfect for busy people who want to take control of their day. Get yours now and start crushing your goals! Link in bio.
Nothing's wrong with it. That's the problem. It could be for anyone, about any planner, written by anyone. You've seen this caption a thousand times because it is a thousand captions.
Now the same request with the framework.
The framework prompt:
Role: You're a thoughtful small-business owner who writes like a friend, not a marketer. Task: Write an Instagram caption introducing my new digital planner. Context: My audience has bought a dozen planners that didn't stick. They're tired of "boost your productivity" energy. My planner is deliberately minimal — five priorities a day, with space left for off days. Avoid the words "crush," "boost," and "goals." Format: Four short paragraphs. Open with a one-line hook. End with a question.
What you get back:
I used to think I needed a better system. Turns out I needed a smaller one.
For two years I bought every planner that promised to fix my week. None of them stuck. The problem was never the planner — it was that I kept planning for a person who didn't exist. Someone with no meetings, no kids, no bad days.
This one has room for the bad days built in. Five things a day, not fifty. That's the whole pitch.
What's the one planning habit you've never been able to stick to?
Same product. Same AI. The only thing that changed was the prompt.
The second one sounds like a person because you told the AI who that person is, who they're talking to, and what to leave out. That's context doing its job.
Try it on one caption today
You don't need to memorise anything. Take the next caption you have to write and add the four parts before you hit enter.
Start small. Even two sentences of context — who your audience is and one thing to avoid — will change what comes back. Write the prompt, read the output, and if it's not quite right, tell the AI what to fix and run it again. That back-and-forth is normal. It's faster than staring at a blank page for an hour.
Want the shortcut?
If you'd rather not build the prompt from scratch every time, I put the whole framework on one page in a free guide — plus three ready-to-use prompts you can copy today, for captions and a couple of other formats.
It's five pages. No email course, no upsell hidden in the middle. Just the framework and the prompts, so the next time you sit down to write a caption, you're not starting from zero.
— Em, Side Business Co