The best AI prompts for social media are specific, not clever. A vague request gives you a vague post that sounds like every other AI-written caption on the timeline. A precise prompt, with your voice, your constraints, and a real example attached, gives you copy you can publish today.
This is a working library, not a theory piece. Every section below is organized by the job you are actually trying to do, from hooks to captions to a full content calendar. Each one includes real prompts you can copy, paste, and adjust for your own brand. Swap the bracketed placeholders for your own details and the output improves immediately, because the model has something concrete to build on instead of guessing.
A quick note on how to use these. Treat every prompt as a starting point, run it, then tell the model what to fix. The second and third rounds are usually where the good copy shows up, so do not settle for the first draft. Good social media prompts are a conversation, not a vending machine.
We have grouped these AI prompts for social media by use case so you can jump straight to what you need. At the end, we cover when to reach for ChatGPT versus Claude versus Gemini, how to stop AI from sounding generic, and the one thing no prompt can do for you.
Hooks and Opening Lines
The first line decides whether anyone reads the second. Most posts die in the first three seconds because the opening is a throat-clear instead of a hook. A line like "I wanted to share some thoughts on productivity" tells the reader nothing and gives them every reason to keep scrolling. Ask AI for many options, then pick the sharpest one yourself.
The reason you ask for twelve and not one is simple. A model handed a single "give me a hook" instruction will reach for the safest, most average phrasing it knows, because average is what it was trained to produce on demand. Force it to generate a dozen and the range widens. Somewhere in that list is usually one angle you would never have written yourself, and that is the one worth keeping.
Give the model a topic, an audience, and the emotion you want to trigger. Then ask for variety, not a single "best" answer.
You are writing scroll-stopping opening lines for short-form social posts.
Topic: [your topic]
Audience: [who they are and what they struggle with]
Platform: [Instagram / TikTok / LinkedIn / X]
Goal emotion: [curiosity / surprise / relief / mild controversy]
Give me 12 opening lines. Rules:
- Under 12 words each.
- No questions that can be answered "no" and scrolled past.
- Mix formats: bold claim, specific number, contrarian take, tiny story.
- No emojis, no hashtags, no exclamation marks.
Label each with the format used.
When you have a winner, expand it:
Take this opening line: "[paste your favorite hook]".
Write three different second lines that keep the tension going and make the reader want line three. Keep the same tone and reading level.
Captions by Platform
A caption that works on LinkedIn will flop on TikTok. The tone, length, and formatting all shift by platform. The trick is telling the model the platform norms explicitly instead of assuming it knows your version of "good."
Instagram. Warm, visual, a little personal. Line breaks matter.
Write an Instagram caption for a post about [topic].
Voice: friendly, first person, quietly confident. Not salesy.
Structure: a 1-line hook, a short story or insight in 3 to 4 short lines with line breaks, then one clear call to action (save, comment, or share).
Length: 80 to 130 words.
End with 5 relevant hashtags, mixing broad and niche.
Do not use the word "unlock" or "elevate".
TikTok. Short, casual, built to be read fast under a moving video.
Write 3 TikTok caption options for a video about [topic].
Each under 150 characters, casual and conversational, one soft hook that adds context the video does not say out loud. Suggest 3 hashtags per option, no more.
LinkedIn. Professional but human. Lead with a specific result or lesson, not a platitude.
Write a LinkedIn post about [lesson or result].
Open with a concrete, specific first line (a number, a moment, or a blunt statement). Use short paragraphs of 1 to 2 sentences with white space between them. Share the actual insight, including what did not work. Close with a genuine question to the reader.
Length: 150 to 220 words. No corporate buzzwords, no "thrilled to announce".
X (Twitter). Tight, punchy, one idea per post.
Turn this idea into 5 standalone X posts: [paste idea].
Each under 240 characters, one clear thought each, no threads. Vary the angle: a claim, a tip, a contrarian take, a short observation, and a question.
The reason you ask for five angles on one idea, rather than five different ideas, is that it teaches you which framing lands. Post the claim on Monday and the contrarian take on Thursday, then watch which one earns replies. Over a few weeks you learn the shape of the post your audience actually responds to, and you can tell the model to lean into it next time.
Content Calendars and Batch Planning
Planning one post at a time is where consistency goes to die. The stronger move is to batch a full week or month from a few content pillars, then fill in the details later. AI is genuinely good at this scaffolding work because it is structured and repetitive.
Start by defining your pillars, then ask for a calendar built from them.
My content pillars are:
1. [pillar one, e.g. beginner tutorials]
2. [pillar two, e.g. behind the scenes]
3. [pillar three, e.g. client results]
4. [pillar four, e.g. industry opinions]
Build a 4-week posting calendar for [platform], 5 posts per week.
For each post give: the day, the pillar, the format (Reel, carousel, single image, text), a working title, and a one-line angle.
Rotate the pillars so no two similar posts land back to back. Return it as a table.
To batch a single week in depth, go one level deeper:
Take week 1 of the calendar above. For each of the 5 posts, write the hook, a 3-bullet outline of the content, and the call to action. Keep my voice casual and specific. Do not write the full captions yet.
Batch planning saves the most time when you treat the AI output as a first draft of the plan, not the finished schedule. You still decide what actually earns a slot.
Repurposing One Piece Into Many
The highest-leverage AI prompts for social media are the ones that turn a single asset into a week of content. One blog post, one video, or one podcast episode holds enough ideas for ten posts if you break it apart deliberately.
Paste your source and ask for a spread of formats:
Here is a [blog post / video transcript / podcast transcript]:
"""
[paste the full text]
"""
Pull out the 8 strongest standalone ideas. For each one, tell me:
- The single point it makes.
- The best platform for it (Instagram, TikTok, LinkedIn, or X).
- A one-line hook.
Do not water anything down into generic advice. Keep the specific examples and numbers from the source.
Then convert the best ideas into finished posts:
Take ideas 1, 3, and 5 from above. Write each as a finished [platform] post in my voice, using the platform rules I gave you earlier. Keep the specific details from the source. Flag anything you had to invent so I can check it.
That last instruction matters. AI will happily fill gaps with plausible-sounding facts, so asking it to flag inventions saves you from publishing something that is not true.
Carousel and Short-Video Scripts
Carousels and short videos live and die on structure. A good script has a hook slide, a build, and a payoff. Prompt for that shape directly rather than asking for "a carousel."
For an Instagram or LinkedIn carousel:
Write a 7-slide carousel about [topic] for [audience].
Slide 1: a hook that names the problem or a surprising claim.
Slides 2 to 6: one idea each, with a 3 to 6 word headline and one supporting sentence.
Slide 7: a summary and a call to action.
Keep language plain and skimmable. Return it slide by slide.
For a short-form video (Reel, TikTok, or Short):
Write a 30 to 40 second video script about [topic].
Line 1: a hook for the first 2 seconds that creates an open loop.
Then 4 to 6 short spoken lines that deliver on the hook.
End with a payoff and a soft call to action.
Add a one-line on-screen text suggestion for each spoken line. Conversational tone, no jargon.
Ask for on-screen text and pacing notes and you will spend far less time reformatting the output into something you can film. If the hook feels weak, do not rewrite the whole script. Ask only for ten new first lines and keep the body that already works. Fixing one part at a time gives you more control than regenerating everything and hoping the next version is better.
One honest limit here. AI is good at structure and pacing, but it does not know what actually performs for your audience. The script is a strong skeleton. The performance, the delivery, and the specific detail that makes people stop are still yours to add.
Community, Replies, and Engagement
Replies are where relationships get built, and they are also where creators burn out. AI can help you draft warm, on-brand responses at volume, as long as you keep the human check. If you have used SocialBooster's free tools to spot which posts are pulling the most comments, you can prioritize where those replies actually matter.
Here are 10 comments from my latest post: "[paste comments]".
Draft a short reply to each. Rules:
- Sound like a real person, not a brand account.
- Match the energy of the comment.
- Ask a follow-up question where it fits naturally.
- Never repeat the same phrasing twice across replies.
- Keep each under 30 words.
Flag any comment that looks like it needs a human decision (complaint, refund, sensitive topic).
For starting conversations rather than answering them:
Write 5 genuine, specific comments I could leave on posts from other creators in [niche]. Each should add a real thought or ask a real question, not "great post". Under 25 words each.
Ad and Promo Copy
Promotional copy is where generic AI writing does the most damage, because hype reads as noise. The fix is to feed the model the specific offer, the specific objection, and the specific outcome, then ask for restraint.
Write ad copy for [product or offer].
Audience: [who] who currently [their situation/frustration].
Main benefit: [the concrete outcome].
Biggest objection: [what makes them hesitate].
Offer: [price, bonus, guarantee, or deadline].
Give me 3 variations:
1. Problem-first (lead with the pain).
2. Result-first (lead with the outcome).
3. Story-first (a short before-and-after).
Each under 90 words. No hype words, no fake urgency. Address the objection directly in each one.
For a quick round of promo hooks to test:
Write 8 one-line promo hooks for [offer], each under 15 words. Vary the angle: price, speed, guarantee, social proof, and the cost of doing nothing. No exclamation marks.
When to Use ChatGPT, Claude, or Gemini
There is no single best model. Each one has a shape it fits, and knowing that saves you from fighting the wrong tool. The same AI prompts for social media will produce noticeably different results depending on which model reads them, so it pays to match the job to the strength.
ChatGPT for fast general drafting. When you need volume quickly, a batch of captions, twenty hook variations, a rough calendar, ChatGPT is quick and reliable. It is the default workhorse for high-throughput drafting and iterating in a conversation.
Claude for long context and brand voice. When you paste a full podcast transcript, a long blog post, or three pages of your brand guidelines and ask for nuanced writing that holds a consistent voice, Claude tends to keep more of the detail and follow tone instructions more closely across a long document. It is the one to reach for when the writing needs to sound genuinely like you across a lot of material.
Gemini for real-time search and trend grounding. When you need current information, what is trending this week, a recent event to react to, live context from search, Gemini's grounding in real-time results is the advantage. Use it to find the angle, then draft the post wherever you write best.
In practice, many people mix them. Find the trend in Gemini, draft the volume in ChatGPT, and polish the long brand pieces in Claude. The models are cheap enough that using the right one for each job costs you almost nothing.
How to Avoid Generic AI Output
Most "AI-sounding" content is a prompting failure, not a model failure. The model returns beige copy because you gave it a beige prompt. Four inputs fix the majority of it.
- Give it your voice. Paste two or three of your own best posts and tell the model to match the rhythm, vocabulary, and sentence length. "Write in this voice" plus real examples beats any adjective you could pick.
- Give it constraints. Word counts, banned words, required structure, reading level. Constraints force specificity. A post capped at 90 words with no buzzwords cannot hide in fluff.
- Give it specifics. Real numbers, real names, real moments from your own experience. AI cannot invent your Tuesday. Feed it the details and it will build around them instead of reaching for cliches.
- Ban the tells. List the words you never want to see. Common offenders include "unlock", "elevate", "supercharge", "game-changer", "in today's landscape", and "dive in". Telling the model to avoid them removes the biggest giveaways in one line.
Here is a reusable voice-training prompt you can save:
Here are 3 posts I wrote: """[paste posts]""".
Study the voice: sentence length, word choice, how formal or casual, how much I use humor, and how I open and close. Describe my voice in 5 bullet points. Then write a new post about [topic] in that exact voice. Avoid these words: [your banned list].
The honest tradeoff is time. Setting this up takes ten minutes the first time. After that you reuse it forever, and the output stops sounding like a robot wrote it in a hurry.
The One Thing Prompts Cannot Do
Here is the limit worth being honest about. AI writes the words. It does not create the reach. A perfect caption still lands in front of almost no one if the post has no early engagement to signal the algorithm that people care.
Every major platform weighs the first hour heavily. Posts that pick up likes, comments, saves, and shares early get pushed to more feeds. Posts that sit silent get buried, no matter how good the writing is. This is the part no prompt can solve, because it is about distribution, not copy.
That is where the work stops being about words. You need real people engaging with real content, and you need enough of that early signal to give a good post a fair chance. This is the gap SocialBooster fills, with real engagement across 21+ platforms so your best AI-assisted content is not left talking to an empty room. The prompts get you the post. The engagement gets it seen.
Use both together and the math changes. Sharper copy earns more from every viewer, and stronger early signals put that copy in front of more viewers to begin with.
The Bottom Line
The best AI prompts for social media are the specific ones: a clear job, your real voice, hard constraints, and concrete details. Grab the prompts above, adapt them to your brand, and pick the model that fits each task, ChatGPT for speed, Claude for long-context brand writing, and Gemini for live trends.
But remember the split. AI produces the words, and it does that faster and better than most of us can alone. Reach is a different problem. Great copy still needs early engagement to travel, and that is the piece worth investing in once your content is genuinely good. Write with AI, distribute with real engagement, and give every post you make an honest shot at being seen.