AI Prompts for Social Media Managers (2026)

The single biggest unlock is a reusable brand voice block. Write it once — audience, tone, three words you sound like, three you never use, emoji policy, banned phrases, CTA style — and paste it at the top of every prompt. That's what makes AI output sound like you instead of generic AI. Then ask for options, not one answer: generating five hooks and choosing one beats accepting the first.

Treat AI as a drafting partner, not an autopilot. It is excellent at volume, variation, and structure; it is unreliable for facts, figures, and current events, and it does not know your latest performance data unless you paste it. Never let a tool post or reply on its own — keep a human review step, especially for community replies, where tone and timing matter. Build your voice block once with the Social Media Caption Generator or LinkedIn Post Generator.

For the underlying technique, the DAIR.ai Prompt Engineering Guide and our complete guide to prompt engineering explain why giving the model role, context, and constraints produces far better output.

Use these to generate options fast, then cut to the best one. Always paste your brand voice block first.

**1. Five scroll-stopping hooks for one post** — "BRAND VOICE: [PASTE VOICE BLOCK]. Write 5 distinct opening hooks (first line only) for a [PLATFORM] post about [TOPIC], aimed at [AUDIENCE]. Vary the angle: one question, one bold statement, one stat-style (leave the number as [X] for me to verify), one story opener, one contrarian take. Keep each under [12] words. No clickbait that the post can't deliver on."

**2. Full caption in three lengths** — "BRAND VOICE: [PASTE]. Write a caption for [PLATFORM] promoting [POST TOPIC / OFFER]. Give three versions: short (under 50 words), medium (under 120), and long (under 250). Each must have a strong first line, one clear idea, and a single CTA to [ACTION]. Use [BRACKETS] for any claim, price, or stat I must confirm — do not invent them. Suggest line breaks for readability."

**3. On-brand hashtag and CTA set** — "BRAND VOICE: [PASTE]. For a [PLATFORM] post about [TOPIC], suggest: (a) 2 CTA options matching our style, and (b) 8–12 relevant hashtags grouped as broad / niche / branded. Avoid banned-words list: [LIST]. Don't invent a branded hashtag we don't already use — flag if one is needed."

**4. Adapt one idea across platforms** — "BRAND VOICE: [PASTE]. Take this core message: [PASTE]. Rewrite it natively for [INSTAGRAM, LINKEDIN, X, TIKTOK CAPTION] — match each platform's tone, length norms, and CTA style rather than copy-pasting. Keep the same core idea and any [BRACKETED] facts unchanged."

AI is strong at structuring a calendar from a theme; you supply the strategy and verify any dates or claims.

**5. Draft a themed weekly calendar** — "BRAND VOICE: [PASTE]. Build a [7]-day content calendar for [BRAND] on [PLATFORMS] around the theme '[THEME]'. For each day give: format (reel/carousel/static/story), a one-line concept, a hook, and the CTA. Mix educational, social-proof, behind-the-scenes, and promotional posts in a [70/30] value-to-promo ratio. Don't schedule claims or dates I haven't given — use [BRACKETS]. Present it as a table."

**6. Plan a campaign or launch sequence** — "BRAND VOICE: [PASTE]. Outline a [PLATFORM] content sequence for the launch of [PRODUCT/EVENT] over [DURATION]. Map a tease → reveal → proof → urgency → recap arc, with the post concept and goal for each phase. Note where I should insert real assets, dates, and offers as [BRACKETS]. Keep it practical for a small team."

**7. Generate a content-pillar idea bank** — "BRAND VOICE: [PASTE]. For a [BRAND] serving [AUDIENCE], propose [4] content pillars and [5] specific post ideas under each, with a suggested format for every idea. Ideas must be doable without a big budget. Avoid generic filler — make each idea specific to this audience."

Repurpose long content into many posts, and use AI to draft community replies you then send yourself — never auto-reply.

**8. Repurpose long content into a week of posts** — "BRAND VOICE: [PASTE]. From the [blog post / video transcript / webinar notes] below, pull [7] standalone social posts for [PLATFORM]. For each: the key idea, a hook, a short caption, and the format. Stay faithful to the source — do not add claims, stats, or quotes that aren't in it; mark anything I should verify as [CHECK]. Source: [PASTE]"

**9. Draft on-brand community replies (for review)** — "BRAND VOICE: [PASTE]. Below are [comments / DMs] on our posts. For each, draft a short reply in our voice. Categorize first: positive (thank/engage), question (answer or route), complaint (empathize, take it to DM, no defensiveness), spam/troll (ignore or brief). These are drafts for me to review and send — do not promise refunds, dates, or outcomes; use [BRACKETS]. Flag anything that needs a human or a manager. Items: [PASTE]"

**10. Handle a sensitive complaint carefully** — "BRAND VOICE: [PASTE]. A customer posted this complaint publicly: [PASTE]. Draft a calm, empathetic public reply (under 60 words) that acknowledges the issue, avoids admitting legal fault, and moves the conversation to DM or [SUPPORT CHANNEL]. Then draft a short private follow-up. Do not invent what happened or promise a specific resolution — use [BRACKETS] for anything I must confirm."

Don't post AI drafts unedited. Generic, hook-free, emoji-stuffed captions read as AI and erode trust — always cut to one strong idea, tighten the first line, and make it sound like your brand. Don't auto-reply to comments or DMs; community tone and judgment need a human, and a wrong automated reply during a complaint can escalate fast.

Don't let AI invent facts, stats, prices, or current events. Models state plausible-but-wrong numbers confidently and don't know today's news or your latest results unless you paste them. Keep every claim, price, and statistic in [brackets] you verify before publishing. Don't fabricate trends or 'data' for a post.

Don't paste anything sensitive — customer personal data, unreleased launch details you can't risk leaking, or login credentials. For technique and reliable structure, see Learn Prompting and our 12 prompt patterns that convert.

For captions, calendars, repurposing, and replies, any current frontier model works well. The practical differences are speed, cost, how much you can paste at once (handy when repurposing a long transcript), and whether you need real-time awareness.

Fast, low-cost options like Claude Haiku 4.5, Gemini 3.5 Flash, or GPT-5.5 Instant suit high-volume caption and reply drafting. For repurposing long content or planning campaigns, a more capable model such as Claude Opus 4.8, Gemini 3.5 Pro, or GPT-5.5 helps. If you need to reference live conversation on X, Grok 4 has real-time access — but verify anything timely yourself. Check live pricing at OpenAI, Anthropic, Gemini, and xAI, and see how to choose an AI model.