How to Use AI to Write Social Media Captions That Don't Sound Like AI

The real problem with AI captions

The AI writing backlash is real, and it's understandable. LinkedIn is full of posts that start with "In today's fast-paced world..." and Instagram captions that say "Elevate your experience with our premium solution."

These captions sound hollow because they are hollow. They're the product of bad prompting, not bad AI.

Here's how to prompt Claude (and other AI tools) so the output sounds genuinely human.

The prompting principles that actually work

1. Give context before the brief

Don't just ask for a caption about your product. Tell the AI who your brand is first.

Bad prompt: *"Write an Instagram caption about our new coffee range."*

Better prompt: *"I run a small specialty coffee roastery in Brunswick, Melbourne. Our voice is knowledgeable but not pretentious — like explaining coffee to a curious friend, not a snob. We use first-person plural (we, our). Write an Instagram caption about our new Ethiopian single-origin, highlighting the floral and citrus notes."*

The second prompt will produce dramatically better output.

2. Include what NOT to do

AI tends toward the clichéd without constraints. Tell it what to avoid.

"Don't use words like 'premium,' 'elevate,' 'unlock,' or 'transform.' No exclamation marks in the first sentence. No corporate language."

3. Ask for specificity

Vague prompts produce vague captions. Specific prompts produce specific, interesting captions.

"Include the specific roasting profile — medium roast, 30 seconds past first crack. Mention we source directly from the Yirgacheffe cooperative."

4. Set a length and structure

"Keep it under 100 words. Start with an observation, not a fact. End with a question."

Structure constraints push AI away from default patterns and toward more interesting openings.

The editing principles that matter

Good AI output isn't good Instagram content until it's been edited. Here's what to look for:

Remove hollow phrases. Ctrl+F for "in today's world," "our journey," "we're excited to," "game-changer," and similar filler. Delete them.

Add specificity. Replace "high-quality ingredients" with "grown at 1,900m altitude in the Yirgacheffe region." Specific beats generic every time.

Add your actual voice. Read the caption aloud. Would you actually say this? If not, rewrite the bits that feel wrong.

Check the first sentence. The first line determines whether someone reads the rest. If AI has opened with a generic statement, rewrite it into something that earns attention.

Verify facts. AI hallucinates. Any specific claims in the caption need to be verified against what's actually true about your product.

The ratio that works

I find the best results come from treating AI as 60% of the work. It:

•Generates the structure

•Provides the core message

•Writes the draft

You:

•Rewrite the opening line

•Add a specific detail or anecdote

•Adjust tone to match your exact voice

•Check for clichés and remove them

With this approach, AI captions take about 3-4 minutes per post rather than 10-15 — still a massive time saving — while producing output that sounds genuinely human.

Testing the output

Before you schedule, read every caption with this question: "Would my target customer believe a person wrote this?"

If the answer is yes: schedule it.

If the answer is maybe: tighten it up.

If the answer is no: rewrite more aggressively or go back to the AI with better prompts.

The goal isn't AI captions that nobody can detect. The goal is captions that serve your audience and grow your business. AI is a tool for getting there faster — not a shortcut that bypasses quality.