AI Social Media Prompts Library (2026): ChatGPT, Claude, Gemini, Copilot & Perplexity
AI Social Media Prompts Library (2026): ChatGPT, Claude, Gemini, Copilot & Perplexity
If your AI captions still sound like a press release wrote them, the problem is almost never the model — it’s the brief. Vague asks like “write a LinkedIn post about our product” produce vague output. Specific briefs with audience, offer, voice, platform rules, and a clear output format produce posts you can actually schedule.
Start here: paste the universal context block below into ChatGPT, Claude, Gemini, Copilot, or Perplexity, then run one task prompt at a time (hooks → captions → calendar → channel adaptations). Edit for voice, then queue the week in one place instead of bouncing between chat tabs and native apps. For model-specific packs, see ChatGPT prompts and Claude prompts. When you’re ready to stop copy-pasting into five apps, Agent Mode and Calendar Mode keep drafting and publishing in one workflow.
Key Takeaways
- You get usable drafts when every prompt includes brand, audience, offer, voice, platform, and output format — not when you ask for 'a caption'
- You protect reach by adapting one core idea per channel instead of identical cross-posts to TikTok, LinkedIn, and Instagram
- You cut revision time with a dedicated 'remove AI smell' pass: shorter sentences, specific verbs, one real example
- You ship faster when drafting lives next to scheduling — Agent Mode + Calendar Mode beat tab-hopping across chat tools and native apps
How to brief AI like a creative director
Treat the model like a junior strategist who has never met your brand. That person needs the same inputs you’d put in a creative brief — not a one-line request.
Name the job. “Write 10 hooks for TikTok” beats “help with content.” Narrow jobs produce tighter first drafts and fewer rewrite rounds.
Lock the audience and the offer. Who is scrolling, what pain they feel, and what (if anything) you’re asking them to do next. Soft CTAs (comment, save, follow) usually outperform hard pitches in feed posts.
Constrain voice and forbidden territory. Three adjectives for tone, plus a short “never say” list (competitors, unverifiable claims, banned slang). This is how agencies keep five clients from sounding like the same chatbot.
Demand a format. Word count, structure (hook + proof + CTA), number of variants, and what to leave out (hashtag spam, emoji walls, “Are you tired of…” openers).
Iterate on purpose. First pass for volume, second pass to humanize, third pass to adapt per platform. One mega-prompt that tries to do everything usually fails at all of it.
Model-specific nuance lives in the companion guides — ChatGPT for fast iteration, Claude for long-form and voice control, Gemini for research-heavy posts, Copilot for Microsoft-centric teams, Perplexity when you need fresh trend context before you write. This library is the shared briefing system that works across all of them.
Universal context block (paste above any prompt)
Copy this once per brand or client. Keep it pinned at the top of the chat (or in a custom GPT / Claude Project). Every task prompt below assumes this block already exists.
Brand: [name] Audience: [who they are + the pain they feel while scrolling] Offer: [product / lead magnet / none — be explicit] Voice: [3 adjectives, e.g. direct, witty, expert] Forbidden: [words, claims, competitors, overpromises] Proof we can use: [1–3 real metrics, customer lines, or case facts] Platforms this week: [TikTok, Instagram, LinkedIn, X, YouTube Shorts, etc.] Goal: [awareness | engagement | clicks | demo bookings | community replies] Compliance notes: [industry rules, required disclaimers, approval workflow] Do not invent: testimonials, numbers, partnerships, or certifications we did not provide
Hook prompts
Hooks decide whether anyone reads the rest. Ask for many options, then pick one that sounds like your brand — not the “cleverest” line in the batch.
Write 12 scroll-stopping opening lines for [platform] about [topic]. Constraints:
- Max 12 words each
- No openers that start with "Are you" or "In today's world"
- Include one concrete number, named object, or specific outcome in every line
- Audience: [who]
- Tone: [voice from context block] Return as a numbered list only — no explanations.
Write 10 LinkedIn post openers about [insight or story]. Each opener: 1–2 sentences, max 40 words. Mix: 3 story-led, 3 contrarian, 2 data-led, 2 question-led (but not "Are you…"). Audience: [role + industry]. Goal: [comments | follows | profile visits]. Flag any opener that feels clickbait without a real payoff.
Here are 8 hooks I drafted: [paste]. Score each 1–5 for: specificity, brand fit, scroll-stop power, and honesty. Rewrite the bottom 3. Keep the top 2 unchanged. Explain failures in one short note each.
Caption prompts
Once you have a hook, generate caption packs with a fixed structure so editing is mechanical instead of existential.
Using the context block, write 5 captions for [platform]. Core idea: [1–2 sentences] For each caption include:
- Hook line
- 2–4 sentences of value or story
- One proof point (use only facts from the context block)
- One soft CTA: [follow | comment with X | save | link in bio | book a demo] Length: [short | medium] Rules: zero hashtag spam (max 3 relevant tags if the platform benefits), no buzzword stacks, no "game-changer" / "leverage" / "synergy".
Write 5 Instagram captions for a Reel about [topic]. Assume the first line is on-screen text; the caption can expand the story. Include: 1 line of social proof, 1 question that invites comments, 1 soft CTA. Keep each caption under 150 words. Suggest 5 niche hashtags max (not the same #marketing #business set every time).
Write one LinkedIn post (180–220 words) teaching [insight]. Structure: hook → short story or observation → 3 concrete bullets → CTA to comment with a specific prompt. No external link in the body (put the link suggestion in a "first comment" line at the end). Tone: peer talking to peers, not vendor pitching buyers.
Short-form script prompts
Scripts need visual direction, not just spoken words. Ask for both or you’ll get a radio read.
Write a 30–40 second script for [platform] on [topic]. Return as a table with columns: seconds, visual, spoken line, on-screen text. Include a pattern interrupt in the first 3 seconds and a clear CTA in the last 5. Native feel — not overproduced. Audience: [who]. Voice: [voice]. Do not invent stats; if a number is needed and missing, mark it as [NEED METRIC].
Outline an 8-slide carousel for [platform] about [topic]. For each slide: headline (max 8 words) + body (max 25 words) + visual note. Slide 1 must stop the scroll. Slide 8 is the CTA. Audience: [who]. Offer: [offer or none].
Content calendar prompts
Calendars fail when every day is a product pitch. Force a mix, then schedule with platform-specific times — not one timestamp everywhere. Channel timing defaults and scheduling tips live on Postevery’s channel guides for TikTok, Instagram, LinkedIn, and YouTube.
Build a 7-day content calendar for [brand] on [platforms]. Mix: 40% educational, 30% social proof / stories, 20% product or offer, 10% personality / behind-the-scenes. For each day return: day, platform, format (post / Reel / carousel / Short / thread), hook, caption outline (3 bullets), CTA, asset needed, suggested local time window. Assume we batch-create on Monday and schedule from a multi-platform calendar. Do not schedule the same caption on two platforms the same day without an adaptation note.
Create a 30-day theme map for [brand] serving [audience]. Group days into weekly themes (4 themes). For each theme list: 3 post angles, 1 proof angle, 1 personality angle. Flag any theme that requires assets we probably don't have yet. Goal for the month: [metric]. Platforms: [list].
Here are last week's top 3 posts with performance notes: [paste]. Propose 6 follow-ups that extend the winning angles without repeating the same hook. Include: platform, format, new hook, why it should work, risk if we overdo it.
Multi-platform adaptation prompts
Identical cross-posts underperform because feeds, norms, and attention spans differ. Keep the insight; change the packaging.
Adapt this core post for TikTok, Instagram Reels, LinkedIn, X, and YouTube Shorts. Keep the same insight; change hook, length, CTA, and native format notes. Core: [paste] Then list 5 things NOT to copy verbatim across platforms (hashtags, link placement, emoji density, formality, length).
Take this LinkedIn post and produce:
- A 35-second Reel/TikTok script
- An Instagram caption under 80 words
- An X thread of 5 posts (each under 260 characters) Preserve the core insight. Cut jargon. Add a visual direction for the short-form version. Original: [paste]
Expand this short-form script into a LinkedIn post (180–220 words) with a story beat and 3 bullets. Do not pad with fluff. Add one concrete example from the context block. Script: [paste]
Agency & multi-client prompts
Agencies need volume without voice bleed. Put client name, industry, and compliance in every prompt — never rely on chat memory alone across clients.
For client [name] in [industry], produce:
- 3 educational posts
- 1 social-proof / case post
- 1 personality / behind-the-scenes post Platforms: [list]. Voice: [voice]. Forbidden claims: [list]. Output as a table ready for calendar review: day suggestion, platform, hook, caption draft, asset needed, approval risk (low/med/high). If a claim needs legal review, mark it — do not soft-pedal.
Here are 5 posts this client loved and 3 they rejected: [paste]. Infer voice rules (do / don't) in 8 bullets. Rewrite this new draft to match the loved set: [paste new draft]. Call out any remaining lines that still sound generic.
I manage three clients. For each, generate 3 hooks + 1 caption outline only (not full captions yet). Client A: [brief] Client B: [brief] Client C: [brief] Keep voices visibly different. Label every line with the client name. Stop after outlines — I will request full captions after human selection.
Remove AI smell
AI smell is predictable: stacked adjectives, empty intensity (“unlock,” “elevate,” “revolutionize”), perfect parallel lists, and zero lived detail. Fix it on purpose before anyone hits publish.
Rewrite the text below to sound human and brand-true. Rules:
- Shorter sentences
- Specific verbs (cut "leverage," "utilize," "delve," "landscape," "robust," "seamless")
- One real example or sensory detail from the context block
- Keep facts unchanged — do not invent proof
- Preserve the CTA Text: [paste] Return: rewritten post + a 3-bullet list of what you removed and why.
Audit this draft for AI tells. List every phrase that sounds generated, explain why, and propose a replacement that a real [role] would say out loud. Draft: [paste] Score overall naturalness 1–10 and tell me what would raise it by 2 points.
In today's fast-paced digital landscape, brands must leverage cutting-edge strategies to unlock unprecedented engagement and drive meaningful results across the entire customer journey.
Most brands post more and get ignored. We cut our weekly plan to five posts, wrote one real customer story, and replies doubled in two weeks.
Compliance & brand-safety prompts
Regulated industries and careful brands need the model to refuse fabrication — not “sound confident.” Put the rules in the prompt every time.
Rewrite this post for [industry: finance | health | legal | alcohol | other]. Required: [disclaimers / "not advice" language / age gates / required phrases]. Forbidden: [guarantees, medical claims, income claims, competitor comparisons]. Keep the hook strong without crossing the line. If the original cannot be saved without inventing claims, say so and propose a safer angle. Draft: [paste]
List every claim in this draft that would need a source, screenshot, or legal approval before publish. For each: quote the claim, risk level (low/med/high), and a safer alternative wording. Draft: [paste] Context block facts we can stand behind: [paste allowed facts]
Do this / skip that
Do
- Paste the universal context block before every task prompt
- Ask for 5–12 options, then pick — never publish the first draft untouched
- Adapt hooks, length, and CTAs per platform even when the insight is the same
- Run a dedicated remove-AI-smell pass with your real examples and metrics
- Batch on Monday, schedule per-channel times, leave mid-week for replies and DMs
Don't
- Ask 'write a caption about our brand' with no audience, offer, or format
- Identical-cross-post the same caption to TikTok and LinkedIn
- Stuff 20 hashtags or lead with 'Are you tired of…'
- Let the model invent testimonials, stats, or certifications
- Treat scheduling as a substitute for engagement — replies still matter
From prompt to published
Load the context block
Generate in layers
Human-edit voice
Adapt per channel
Queue and ship
FAQ
Frequently Asked Questions
Which AI model is best for social captions in 2026?
ChatGPT and Claude lead for most teams on voice and fast iteration. Gemini is strong when you need research-backed posts. Copilot fits Microsoft-centric workflows. Perplexity helps when you want fresh trend context before you draft. Pick the model you already open daily, then use a tight brief — the brief matters more than switching tools.
Should I post the same caption on every platform?
No. Keep one insight, then change the hook, length, CTA, and format. TikTok rewards fast pattern interrupts; LinkedIn rewards peer-level depth; Instagram often needs a strong first line plus a comment prompt. Identical cross-posts usually underperform.
How do I stop AI captions from sounding robotic?
Add a dedicated rewrite pass: shorter sentences, specific verbs, one real example, and a ban list for words like leverage, unlock, and seamless. Compare the draft to posts your audience already engages with. If you would not say it out loud, do not publish it.
Do these prompts replace a content calendar?
No. Prompts accelerate ideation and drafting. You still need a calendar for cadence, approvals, asset readiness, and platform-specific publish times. Draft with AI, then schedule the week in one queue.
How should agencies use this library without voice bleed?
Give every client a separate context block and a loved/rejected example set. Label outputs with the client name. Run compliance and claim-scrubber prompts before client review. Never reuse one chat thread for multiple brands.
Where should I start if I only have one hour this week?
Paste the universal context block, run the 7-day calendar prompt, fully write and humanize three posts for your top two platforms, adapt hooks per channel, and schedule them. Leave the rest of the calendar as outlines for next batch day.
Can I use these prompts with Postevery Agent Mode?
Yes. Agent Mode lets you connect Claude, ChatGPT, Cursor, OpenClaw, and other agents so drafting and scheduling live in one workflow instead of browser tabs. Use these prompts inside the agent, review drafts, then place approved posts on the calendar.
What is the difference between this library and the model-specific guides?
This library is the shared briefing system and cross-model prompt set. The ChatGPT and Claude guides go deeper on model quirks, output styles, and copy-paste packs tuned to each tool. Start here for the system; open a model guide when you want that tool’s best defaults.
One calendar. 27+ platforms. AI built in.
Schedule with Calendar Mode or automate with Agent Mode. Start a 7-day free trial on any monthly plan.
Start free trial