50 AI Prompts for Social Media in 2026 That Produce Usable Drafts, Not Generic Filler

Good AI prompts for social media share four ingredients: real source material, a clear audience, specific format constraints, and instructions about what to do instead of the usual AI habits. Add 3 to 5 examples of your own posts and most assistants produce drafts that need editing, not rewriting. The 50 prompts below apply that formula from strategy to reporting.

They work with any capable assistant, including Claude, ChatGPT and Gemini. Copy them, replace the bracketed parts with your details, and edit every result before publishing. Prompts marked "connected" assume the assistant is linked to a scheduler such as Postprism through its MCP server, so it can save drafts or read analytics; without that connection the same prompts still work, and you paste the drafts into your scheduler yourself.

Key findings: what makes an AI prompt for social media work in 2026

• Examples are the strongest lever. Anthropic's prompting best practices call examples "one of the most reliable ways to steer Claude's output format, tone, and structure" and recommend 3 to 5. Google's Gemini prompting guide goes further: "We recommend to always include few-shot examples in your prompts."
• Clarity beats cleverness. Anthropic's golden rule: show your prompt to a colleague with minimal context, and "If they'd be confused, Claude will be too."
• Delimiters help. OpenAI's prompt engineering guide says Markdown headers and XML tags help mark distinct sections, such as instructions, examples and source material.
• Edited AI posts can perform. Buffer's study of 1.2 million posts found AI-assisted posts at a 5.87% median engagement rate against 4.82% for human-only posts, among 15,000 people who posted both ways.
• Generic is the default. Originality.ai classified 81.2% of a July 2026 sample of 5,000 long LinkedIn posts as likely AI. A prompt without your material produces one more of those.
• Accuracy is the risk readers care about. Pew Research Center found 66% of US adults highly concerned about people getting inaccurate information from AI, which is why several prompts below ask the model to list its claims.

What has changed in AI prompting for social media in 2026

The model makers converged on the same advice

Anthropic, OpenAI and Google now publish prompting guides that agree on the basics: clear and specific instructions, examples, context, and structure. Google's guide puts it as "provide it with clear and specific instructions" and "Specify any constraints". OpenAI's guide says that when providing examples you should "show a diverse range of possible inputs with the desired outputs". The prompts below follow those rules rather than "prompt hacks".

They still disagree on where long material goes

Anthropic advises putting long documents "near the top of your prompt, above your query, instructions, and examples". OpenAI's guide says context data is "usually best positioned near the end of your prompt". In practice, both work for social posts, which are short; if results are poor with a long transcript, try moving it.

Assistants can act on your calendar

Assistants connected through the Model Context Protocol can call tools such as a scheduler. That turns prompts like "save these as drafts for Tuesday" or "which days next week are empty?" into actions. Our AI agents guides show how to connect Claude, ChatGPT, Cursor and Gemini CLI to Postprism.

Labels target AI media, not AI-assisted writing

YouTube's altered or synthetic content policy lists idea generation, caption creation and help with "a video outline, script, thumbnail" as uses that do not need disclosure, while photorealistic AI content does. Meta applies an "AI info" label to content with industry-standard AI image indicators. The prompts here are mostly about writing, where those labels do not apply; the visual prompts produce briefs and alt text rather than photorealistic images.

The prompt formula

Anthropic recommends telling the model what to do rather than what not to do. "Write in short, plain sentences like the examples" works better than "Don't sound corporate". Where a prompt below bans something, it also says what to do instead.

1. Strategy and planning (6 prompts)

1. Pillars: "Here is what our business does, who we serve and our last 20 posts. Suggest four content pillars that fit our goal of [goal], each with a one-line description and three example posts."
2. Monthly plan: "Using these pillars and key dates [list], draft a posting plan for [month] with [n] posts a week on [platforms]. Give date, pillar, format and a one-line idea per slot."
3. Mix check: "Here is next month's plan. Which share is promotional? Which pillars are under-represented? Suggest specific swaps."
4. Audience sentence: "From these customer reviews and questions, write one sentence describing who we are talking to, what they want, what frustrates them and what they currently believe."
5. Platform fit: "Given this audience and a team of [size] with [hours] a week, which two platforms should we focus on and why? Be direct about the trade-offs."
6. Campaign plan: "We launch [product] on [date]. Draft a two-week tease, launch and follow-up sequence for [platforms], with the goal of each phase."

2. Ideas and angles (7 prompts)

7. Angles: "Give me 15 angles on [topic] for [audience]. Include three contrarian angles and three built on common mistakes."
8. From questions: "Here are 20 questions customers asked this month. Group them into themes and suggest one post per theme."
9. From objections: "These are the reasons prospects give for not buying. Turn each into an educational post idea that addresses it honestly."
10. Series ideas: "Suggest five recurring weekly series we could run for a year on [platform], each easy to produce from things we already do."
11. Behind the scenes: "Here is what my team does in a typical week [describe]. List ten behind-the-scenes post ideas from it."
12. Seasonal: "List the moments in [month] relevant to [audience] in [country], with one idea for each that fits our pillars."
13. Gap finding: "Here are the topics our competitors post about [list]. What useful topics are they missing for [audience]?"

3. Hooks and first lines (6 prompts)

14. Hook options: "Write ten first lines for this post. Each under 12 words, each using a specific detail from the draft."
15. Hook styles: "Give me first lines for this post in five styles: a number, a mistake, a strong opinion, a tiny story and a surprising detail."
16. Video hooks: "Write five opening lines for a 30-second video on [topic] that make the payoff clear in the first two seconds."
17. Carousel covers: "Write six cover slide titles for a carousel about [topic], each under eight words."
18. Critique: "Here is my first line. Why might a busy reader scroll past it? Suggest three stronger versions."
19. Specificity pass: "Rewrite this first line to include a concrete number, place, time or object from my notes."

4. Drafting posts (8 prompts)

20. From notes: "Using only these notes, write a LinkedIn post that teaches one lesson. Short paragraphs, under 1,200 characters, ending with a question that invites experience rather than agreement."
21. From a transcript: "Here is a transcript of me talking about [topic]. Turn it into a post that keeps my phrasing wherever possible."
22. Story post: "Turn this client situation into a short story post: problem, what we tried, what worked, the lesson. Replace all names and identifying details."
23. How-to post: "Write a five-step how-to on [task] for beginners. One or two sentences per step."
24. Opinion post: "I believe [opinion] because [reasons]. Write a post arguing it clearly and fairly, acknowledging the strongest counterargument."
25. Announcement: "Write a launch post for [product] that leads with the customer problem, not the product name. Include the price and one call to action."
26. Key points: "Pull the five most useful points from this article for [audience]. One sentence each, in plain language, keeping any numbers exactly as written."
27. Tighten: "Cut this post by 30 percent and keep every specific detail. Replace filler phrases with nothing."

5. Adapting for each platform (8 prompts)

28. One into four: "Adapt this post into versions for LinkedIn, X (under 280 characters), Threads (under 500 characters) and an Instagram caption. Keep our voice and use only claims from the original."
29. Instagram carousel: "Outline a seven-slide Instagram carousel from this post: cover, five content slides, a final slide with a save prompt. Then write the caption."
30. Short video script: "Write a 30-second script for TikTok from this post: hook, three beats, ending. Include on-screen text for each beat."
31. Bluesky: "Rewrite this as a Bluesky post under 300 characters in a conversational tone."
32. Facebook: "Adapt this for a Facebook Page audience of local customers. Warmer tone, a clear next step, and our opening hours in the last line."
33. YouTube: "Write five YouTube title options under 60 characters and a description with three key points and a call to action for this video."
34. Pinterest: "Write a search-friendly Pin title and a two-sentence description for this blog post, using the phrase [keyword] naturally."
35. Save as drafts (connected): "Create one post in Postprism with these versions for LinkedIn, X and Bluesky, saved as a draft for Tuesday at 9:30 AM."

The Threads limit comes from Meta's Threads documentation, which states "Text posts are limited to 500 characters"; X at 280 and Bluesky at 300 are the limits Postprism enforces. For the norms behind these adaptations, see adapting content for every platform, and check lengths with our character counter.

6. Visuals and accessibility (5 prompts)

36. Image brief: "Write an image brief for a minimal, on-brand visual for this post. Our colours are [colours]. Describe composition, subject and mood. Keep the image free of text."
37. Alt text: "Write alt text for this image for someone using a screen reader. One or two sentences describing what matters for this post."
38. Thumbnail ideas: "Suggest five thumbnail concepts for this video, each with up to four words of overlay text."
39. Template copy: "Write headline text for a quote graphic from this post, under ten words."
40. Caption cleanup: "Here is my video transcript. Fix punctuation and break it into caption lines of no more than 40 characters."

The W3C images tutorial describes good alt text for informative images as "a short description conveying the essential information presented by the image", which is the standard prompt 37 aims for. For generating the visuals themselves, our Prismatic image tool and our guide to AI image generation for social media go further.

7. Engagement and community (5 prompts)

41. Reply drafts: "Here are ten comments on our post. Draft short replies for each in our voice, and flag any that need a personal response from me."
42. Difficult comment: "Draft a calm, helpful public reply to this complaint, and a private message offering to resolve it."
43. FAQ from comments: "Group these comments into the questions people are really asking. Which deserve their own post?"
44. Conversation starters: "Suggest five questions we could ask our audience this month that are easy to answer and relevant to [topic]."
45. Saved replies: "Write saved replies for our five most common direct messages: [list]. Friendly, under 50 words each."

Use these as starting points and send every reply yourself. Buffer's 2026 engagement report estimates that replying to comments lifts engagement by about 42% on Threads, 30% on LinkedIn and 21% on Instagram within the same account, and that effect depends on replies being genuine.

8. Review and reporting (5 prompts)

46. Voice check: "Compare these drafts with our example posts. Where does the voice drift? Quote the line and suggest an edit."
47. Limit check (connected): "List every scheduled post this month and flag any version over its platform's character limit."
48. Gap check (connected): "Look at my scheduled posts for the next two weeks. Which days and platforms are empty?"
49. Monthly summary (connected): "Get analytics for posts published in the last 30 days. Summarise which pillar and platform performed best, suggest three changes to test, and name the numbers behind each suggestion."
50. Report draft: "Turn these numbers and notes into a one-page monthly report: headline, three KPIs, top posts, learnings and next month's changes."

For choosing what to measure, see our social media metrics and KPIs guide.

Anatomy of a complete prompt

Most of the 50 prompts are short because they assume your context block is already saved. When it is not, a complete prompt has five labelled parts. Labels (or XML-style tags, which both Anthropic and OpenAI recommend for separating sections) stop the model confusing your examples with your instructions.

Written out once and saved, this becomes the template every drafting prompt plugs into. The rest of the page assumes you have it, which is why the individual prompts can stay short.

Prompt variations by platform

The same prompt needs different constraints on each network. Paste the relevant line into any drafting or adapting prompt above.

LinkedIn prompts

LinkedIn rewards a point of view and punishes vagueness. Combine prompts 20 and 24 with your own notes, and use prompt 18 to critique the opening. LinkedIn's own Page best practices report that images typically result in a 2x higher comment rate, so prompt 36 is worth running for most LinkedIn posts.

Instagram prompts

On Instagram, carousels and shareable posts carry the most weight. Meta's Instagram Feed ranking explainer lists how likely someone is to share a post in a direct message among its predictions, so prompt 29 (carousel) and prompt 25 (announcement with a clear reason to share) are the most useful starting points.

TikTok and YouTube prompts

Video prompts should produce scripts, hooks and titles, not finished videos. Prompts 16, 30 and 33 cover the essentials. Read every script aloud before filming; anything that sounds written needs rewriting for speech.

Text-first networks: X, Threads and Bluesky

Short networks are where AI is most likely to sound generic, because there is so little room for detail. Run prompt 19 (specificity pass) on every short post, and use prompt 28 to produce versions from a longer post you already wrote rather than from scratch.

Prompt variations by business type

Solo founders and creators

Your best input is your voice. Record a ten-minute voice memo each week and use prompt 21 on the transcript, then prompts 14 and 28. Keep opinions in your own words; use prompt 24 only to structure an argument you have already made.

Local businesses

Your material is daily life: dishes, products, customers, events. Prompts 11 (behind the scenes), 12 (seasonal), 32 (Facebook) and 37 (alt text) do the most work. Photos should be real; ask AI for captions, not images of your food or products.

B2B teams

You have reports, webinars and case studies. Prompts 9 (objections), 22 (story), 26 (key points) and 28 (one into four) turn long assets into a month of posts. Add "Use only claims approved in this document" to every drafting prompt.

Agencies

Agencies need the same prompts per client with a different voice block each time. Keep one saved context per client and use the connected prompts (35, 47, 48, 49) to work inside each client's workspace. With Postprism's MCP server, an assistant can list your workspaces and draft into the right client's calendar.

Prompt variations by assistant

Claude

Put your voice rules, pillars and examples in a Project so every chat starts with them; Anthropic describes Projects as "self-contained workspaces with their own chat histories and knowledge bases". Wrap examples and source material in tags so Claude can tell them apart from instructions. Our guide to using Claude for social media walks through the full setup.

ChatGPT

Use custom instructions or a project for the context block, and separate instructions, examples and source material with Markdown headers or XML tags, as OpenAI's guide suggests. ChatGPT can connect to Postprism as a custom MCP connector on plans that allow developer mode; the ChatGPT setup guide explains how.

Gemini

Google's guide recommends always including few-shot examples and breaking complex tasks into one prompt per instruction. For social posts, that means running the drafting, adapting and checking prompts as separate steps rather than one long request.

Prompt variations by goal

The same stage needs a different prompt depending on what the post is for. Add the goal line to any drafting prompt so the model optimises for the right outcome.

Reach prompts

Recommendation systems show posts to people who do not follow you, so the post has to make sense to a stranger. TikTok says its For You feed considers interactions, video information such as captions, sounds and hashtags, and device settings, and that follower count is not a direct factor. Ask for hooks that need no context.

Engagement prompts

Questions that invite agreement ("Who else agrees?") get likes; questions that invite experience ("What was the first thing you cut when cash got tight?") get comments. Say which kind you want.

Lead prompts

Promotional posts work best when they are rare and specific. Keep them to a small share of your plan (prompt 3 checks this) and ask for the problem, the offer and one next step, nothing more.

Prompt chains: running the stages in order

Single prompts produce single drafts. Chaining them, so each output feeds the next prompt, produces a week of posts with checks built in. Google's guide recommends this approach for complex tasks: "Instead of having many instructions in one prompt, create one prompt per instruction."

Running the chain once a week on three or four ideas fills most calendars. Our guide to batch content creation shows how to fit it into one sitting, and repurposing content across social media applies the same chain to long articles and webinars.

How to judge a prompt's output

Before a draft goes anywhere near your calendar, run it through five questions. If any answer is no, fix the draft or the prompt.

• Is every fact from your material? Check the self-check list against your notes.
• Could a competitor have posted it? If yes, add a name, number, place or opinion only you have.
• Does the first line work alone? Read it without the rest of the post.
• Does it fit the platform? Length, format and tone for that network.
• Would you say it out loud to a customer? If it sounds like a press release, rewrite it.

Teams can make this a formal step: in Postprism, editors' posts go to an approval queue where an owner or admin approves them, so the five questions get asked by someone other than the writer. See team collaboration for how roles work.

The step-by-step method: build your own prompt library

1. Write a context block. One paragraph on who you are and who you serve, three to five pillars, voice rules written as things to do, and a never-do list.
2. Choose 3 to 5 example posts. Your best real posts, ideally from different platforms, wrapped as examples.
3. Save the context. In a Claude Project, ChatGPT project or custom instructions, or a document you paste at the top of each chat.
4. Pick ten prompts from this page. Start with the stages that take you longest; for most people that is drafting (20 to 27) and adapting (28 to 35).
5. Fill the brackets once. Replace placeholders with your defaults so each prompt is ready to paste.
6. Run each prompt on real material. Note what you edit in the output.
7. Add rules from your edits. If you keep cutting the same phrase or adding the same detail, write it into the prompt or the context block.
8. Add the self-check. End drafting prompts with "list every factual claim and where it came from".
9. Connect your scheduler (optional). Link your assistant to Postprism with MCP so drafts land on your calendar as drafts for review.
10. Review monthly. Retire prompts you no longer use and refine the ones that still need heavy editing.

Worked example: one prompt, improved four times

The jump from version 2 to version 3 is the biggest. Real material does more for quality than any amount of clever wording. Version 4 then turns a good draft into one you can trust, because the self-check shows you exactly which lines to verify.

From prompt to published post

The prompts save the most time when their output goes straight onto your calendar. Connect your assistant to Postprism through MCP and it can save drafts in your workspace for you to review, attach media to and schedule; the full tool list is in the MCP documentation. Or use Postprism's optional built-in AI content adaptation to suggest platform versions in your brand voice inside the editor. Either way, Postprism's job is the part after the prompt: putting every version on one content calendar and publishing to six networks at the time you set.

There is a third route for planning-heavy weeks. Postprism's bulk scheduling includes an AI schedule assistant: describe the posts you want in a chat, or drop in a spreadsheet, and it turns the request into rows with dates, times and platforms. You review every row in a preview table before anything is scheduled, which makes prompt 2 (monthly plan) a natural input.

Common prompting mistakes (and fixes)

1. No source material. "Write a post about productivity" produces everyone's post. Fix: paste notes, transcripts or data, and say "using only".
2. Describing your voice instead of showing it. Fix: include 3 to 5 example posts.
3. Only saying what not to do. Fix: say what to do instead, as Anthropic's guidance recommends.
4. No length or format limits. You get padded drafts. Fix: give character limits and structure.
5. One giant prompt for everything. Quality drops as instructions pile up. Fix: split drafting, adapting and checking into separate prompts, as Google's guide suggests.
6. Accepting the first draft. Fix: ask for three options, pick one, edit it.
7. Letting the model invent facts. Fix: add the self-check line and verify every number.
8. Asking for reviews or testimonials. The US FTC's Operation AI Comply targeted an AI tool that generated fake reviews. Fix: use only real customer words, with permission.
9. Same prompt for every platform. Fix: add the platform line from the table above.
10. Prompting for photorealistic images of your product or people. These can be labelled and can mislead. Fix: ask for briefs, illustrations or backgrounds, and use real photos for what customers buy.
11. Publishing straight from the chat. Fix: save output as drafts and approve before scheduling.
12. Never updating prompts. Fix: turn your recurring edits into rules every month.

For the bigger picture on what to delegate, see how to use AI for social media, and for the evidence on results, whether AI content performs.

Frequently Asked Questions

What is the best AI prompt for writing social media posts?

The best prompts include your real source material, the audience, the platform and format, length limits, 3 to 5 examples of your voice and a self-check. "Using only these notes, write a LinkedIn post under 1,200 characters, matching these example posts, then list every claim and its source" beats any generic prompt.

How do I stop AI social media posts sounding generic?

Give the AI specific material, show your voice with real examples, tell it what to do instead of listing bans, ask for several options, and edit the result to add details and opinions only you have. Run a specificity pass on every first line.

Do these prompts work with ChatGPT, Claude and Gemini?

Yes. They follow the advice all three companies publish: clear instructions, examples, context and structure. Prompts marked "connected" assume the assistant is linked to Postprism through its MCP server so it can save drafts and read analytics.

How many examples should I include in a prompt?

Anthropic recommends 3 to 5 examples, and Google recommends always including few-shot examples. For social media, three of your best real posts is a good default. Make them varied so the model learns your voice rather than copying one post.

Can AI write platform-specific versions of the same post?

Yes, and it is one of the most reliable jobs to give it. Ask for each platform's version with its length and format rules, such as 280 characters on X, 500 on Threads and 300 on Bluesky, then check facts, tone and limits before scheduling.

Should I tell my audience I used AI?

Follow each platform's rules on labelling AI media, which mainly apply to realistic images, video and audio. YouTube, for example, does not require disclosure for idea generation, captions or script help. For text you drafted with AI and edited, accuracy and genuine views matter most.

Can an AI assistant schedule posts from these prompts?

Yes, if it is connected to a scheduler. With the Postprism MCP server, Claude, ChatGPT, Cursor or Gemini CLI can create drafts or scheduled posts on your calendar. Ask for drafts first so you review everything before it publishes.

Where should I save my prompts?

Keep your context block in a Claude Project, a ChatGPT project or custom instructions, and keep the prompts themselves in a shared document your team can copy from. Update both monthly with the edits you keep making.