· 7 min read
How to Write Social Media Captions with AI (Without Sounding Like AI)
Most AI captions fail for one reason: the prompt was one line long. Give the model the same information you'd give a freelance copywriter and the output stops sounding generic. Here's the method.
Step 1 — Write a brief, not a prompt
Four lines, every time:
- Audience: "Brides booking hair and makeup in Glasgow, 25–35."
- Goal: "Get them to message us for a trial date."
- Angle: "Trials book out six months ahead — book early."
- Proof: one real detail — a number, a customer line, a result.
The proof line is what separates a caption that sounds like your business from one that sounds like every business.
Step 2 — Set voice rules once
Write five rules and reuse them on every generation: sentence length, whether you use emojis, British vs American spelling, words you never use, and how you refer to customers. Tools that train on your own past posts do this automatically — that's the difference between a generic writer and an AI social media manager that has read everything you've published.
Step 3 — Ask for the hook separately
Generate five first lines before you generate a caption. The first line decides whether anyone reads the rest. Useful shapes:
- A specific number ("Three things we'd fix on your booking page")
- A mild contradiction ("Posting daily is the wrong goal")
- A direct question your customer has actually asked you
- A short story opener ("A client rang us at 11pm last Friday…")
Step 4 — Edit with a fixed checklist
- Cut the first sentence if the second one is stronger.
- Delete every adjective that isn't doing work.
- Replace vague claims with a number or a name.
- One call to action, not three.
- Read it aloud — if you wouldn't say it to a customer, rewrite it.
Phrases to ban
"Elevate", "unlock", "in today's fast-paced world", "game-changer", "dive in", "we're thrilled to announce". Add them to your rules and the model stops reaching for them.
Step 5 — Keep the winners
Save the captions that performed and feed them back as examples. Over a few months this is what makes AI output sound like you rather than like a template — and it's why AI-assisted content automation gets better the longer you run it, instead of staying flat.