AI Prompts for Social Media Posts
A Reusable Context Block and Eleven Copyable Prompts, From Planning to Replies
Short answer
The best AI prompts for social media start with a context block: your audience, message pillars, voice rules and three real past posts. Then ask for one specific job, such as ten hooks, a caption from a pillar or a reply to a comment, with the platform, length and one idea stated, and edit the output yourself.
1. Why do AI social posts sound generic?
Because the prompt gives the model a topic and nothing else, so it writes the average post on that topic. A point of view, a voice and a purpose have to be supplied; the model cannot infer them from "write a LinkedIn post about onboarding".
A model asked for a post with no context produces the patterns it has seen most often in social content: an opening question, a list of three, a call to engage, a few emojis. None of it is wrong, and none of it is yours. Readers recognize the pattern immediately and scroll.
Every prompt in this guide assumes you paste a context block first. It takes ten minutes to write once and turns every later prompt from "write a post" into "write this specific post for this specific audience in this specific voice". The ai content brand voice guide covers how to turn a voice into rules a model can follow; the block below is the short version.
2. What should I paste before every social media prompt?
A context block with five parts: the brand and audience in two lines, three to five message pillars, voice rules you can check, words to avoid, and three of your own past posts that performed well. Save it and paste it first in every session.
CONTEXT. Brand: [NAME], a [WHAT YOU DO] for [WHO]. Audience on social: [WHO FOLLOWS YOU AND WHAT THEY CARE ABOUT]. Message pillars: 1) [PILLAR AND ONE-LINE MESSAGE] 2) [PILLAR] 3) [PILLAR]. Voice rules: [E.G. SHORT SENTENCES; FIRST PERSON PLURAL; ONE CONCRETE EXAMPLE PER POST; NO EXCLAMATION MARKS; AT MOST ONE EMOJI]. Never use: [WORDS AND PHRASES YOU BAN, E.G. GAME-CHANGER, UNLOCK, SEAMLESS]. Never claim: [CLAIMS YOU CANNOT BACK UP]. Three past posts that sound like us: [PASTE POST 1] [PASTE POST 2] [PASTE POST 3]. Use this context for everything that follows. If a request needs a fact that is not in this context or in my message, ask me instead of inventing it.
The last sentence matters more than it looks. Posts often need a specific: a customer number, a date, a feature name. Telling the model to ask rather than invent turns a hallucinated statistic into a question you can answer in five seconds.
Why paste real past posts instead of describing the voice?
Because descriptions like friendly or bold mean different things to every reader, including a model. Three real posts show sentence length, vocabulary, humor and structure at once. Pick posts that performed well and that you would be happy to publish again, not your most unusual ones.
3. Which prompts help plan social content?
Two: one that turns a pillar into a spread of post ideas with distinct angles, and one that fills a week of calendar slots from those ideas. Both keep the model tied to your pillars instead of trending topics.
Using the context above, give me 12 post ideas for the pillar "[PILLAR]". Spread them across these types: 3 that share a specific customer situation (no names or quotes, I will supply real ones), 3 that explain how something works, 2 that disagree with a common belief in our market, 2 behind-the-scenes, 2 that answer a question customers ask. For each idea give: a working title, the single point the post makes, and which platform it suits best. Do not write the posts yet.
Here is next week's calendar: [LIST SLOTS AS DAY, PLATFORM, PILLAR, FORMAT]. Here are the ideas I have approved: [PASTE CHOSEN IDEAS]. Assign one idea to each slot so that no pillar appears twice on the same day and no two consecutive posts on one platform use the same format. For each slot, give the idea and a one-line brief for the post. Flag any slot where none of the ideas fit.
Asking for ideas without the posts is deliberate. Reviewing twelve one-line ideas takes two minutes; reviewing twelve finished posts takes an hour, and most of that time goes on posts you would have rejected at the idea stage.
4. Which prompts write the posts themselves?
Prompts for hooks, a single post from a brief, platform-specific versions of one idea, a thread and a carousel outline. Each states the platform, the length and the one point the post makes.
Using the context above, write 10 different opening lines for a [PLATFORM] post whose point is: "[THE ONE POINT]". Use a different approach for each: a specific number from my message, a short scene, a plain statement of the point, a mistake people make, a question the audience actually asks, a contrast, a before and after, an admission, a surprising detail, and a direct instruction. Each under [N] words. No questions that can be answered yes or no. No emojis.
Using the context above, write one [PLATFORM] post. Point: [ONE SENTENCE]. Pillar: [PILLAR]. Opening line: [CHOSEN HOOK]. Supporting detail: [FACTS, EXAMPLE OR STORY YOU SUPPLY]. Call to action: [WHAT THE READER SHOULD DO, OR "NONE"]. Length: [TARGET, WITHIN THE PLATFORM LIMIT]. Follow the voice rules. Use only the facts I gave you. Then give me a second version that is 40% shorter.
Here is a post that worked on [ORIGINAL PLATFORM]: [PASTE POST]. Rewrite the same idea for LinkedIn, Instagram and X, using the context above. Keep the point identical. Change the structure to suit each platform: LinkedIn with a short opening and line breaks, Instagram with the key line first and a caption that works alongside an image of [IMAGE DESCRIPTION], X as a single post within the character limit. Do not add facts that are not in the original.
Using the context above, turn this material into an X thread of [5 TO 8] posts: [PASTE MATERIAL]. Post 1 states the point and why it matters, and works on its own if nobody opens the thread. Each following post makes one step or example. The last post sums up in one line and, if relevant, links to [URL]. Every post within the character limit. Number the posts.
Using the context above, outline a [N]-slide carousel for [LINKEDIN OR INSTAGRAM] on: "[TOPIC]". Slide 1: a headline under 8 words that states the payoff. Slides 2 to [N-1]: one idea per slide, a headline under 8 words and at most 25 words of body text. Last slide: a summary line and [CALL TO ACTION]. Also write a caption of 2 to 3 sentences to go with it.
Platform limits change, and the model may not know the current ones. Look them up once, write them into the prompt, and check the output against them. The social media character limits guide keeps the current numbers in one place.
5. Which prompts help with comments and replies?
A reply prompt that drafts two or three options for a comment in your voice, with a rule to hand anything sensitive to a human. Replies are the easiest place for AI to embarrass a brand, so every draft gets read before posting.
Using the context above, draft 3 short reply options to this comment on our [PLATFORM] post. Post: [PASTE POST]. Comment: [PASTE COMMENT]. Options: one that answers directly, one that asks a useful follow-up question, one that thanks them and adds one helpful detail. Under 40 words each. Do not promise refunds, fixes, dates or anything about pricing. If the comment is a complaint, a legal or safety issue, or asks about an individual account, do not draft replies; say "Needs a human" and explain why in one line.
The escalation rule is the guardrail. Complaints, account-specific questions and anything involving money or safety should be answered by a person who can actually resolve them. The respond to negative comments guide covers what that reply should look like.
6. Which prompts improve a draft before publishing?
Two editing prompts: one that removes the patterns that make text read as AI-written, and one that checks the draft against your voice rules and claim list. Run them on your own drafts too.
Edit this draft to remove patterns that make it sound generic: rhetorical questions in the opening, lists of exactly three adjectives, phrases like "in today's fast-paced world", "let's dive in" or "here's the thing", closing questions that ask for engagement, and any sentence that could appear in any company's post. Keep the point and every fact. Show the edited version, then list what you removed. DRAFT: [PASTE]
Check this draft against the context above. List every place it breaks a voice rule, uses a word from the never-use list, or makes a claim that is on the never-claim list or is not supported by facts I supplied. Quote each problem and suggest a fix. Do not rewrite the whole post. DRAFT: [PASTE]
Below are recent posts from three competitors, labeled by company, and our pillars from the context above. List the angles and topics the competitors cover most, with a count for each. Then list which of our pillars they barely touch, and suggest 5 post ideas on those. Only use the posts I pasted. COMPETITOR POSTS: [PASTE, LABELED BY COMPANY]
The last prompt closes the loop between research and content. It works only on posts you paste, since a model asked what competitors are "posting lately" has no reliable way to know.
7. What should I not ask AI to do on social media?
Do not ask it for current trends, for what competitors are posting, for statistics to include, or for customer quotes. It cannot know the first two, and it will invent the last two in a convincing style.
- "What is trending on LinkedIn this week?" A chat model without live data cannot know. Check the platform yourself.
- "Add a statistic to make it stronger." It will produce a plausible number with no source.
- "Write a testimonial from a happy customer." Invented endorsements are deceptive, whatever the format.
- "Post this for me." Publishing should stay a human decision, with a named person approving each post.
- "Make it go viral." There is no instruction that produces reach; it produces a post that imitates viral ones.
Key Takeaways
Context is most of the prompt
Audience, pillars, voice rules, banned words and three real posts turn generic output into drafts that sound like you.
Tell the model to ask, not invent
One line in the context block turns a hallucinated statistic into a question you can answer.
Ideas before posts
Review one-line ideas first. It is faster and cuts most weak posts before anyone writes them.
One post, one point, one platform
State the point, the platform and the length in every writing prompt, and supply the facts yourself.
Replies need an escalation rule
Complaints, account questions and anything about money or safety go to a person, not a draft.
Edit with prompts, approve as a human
Use editing prompts to strip generic patterns and check claims, then have a named person approve every post.
Frequently Asked Questions
Sources
- Reduce hallucinations Anthropic, Claude documentation, September 2026.Recommends explicitly allowing the model to say it does not know, the basis for the "ask me instead of inventing" line.
- Advertising FAQ's: A Guide for Small Business US Federal Trade Commission, September 2026.Why brand posts that make product claims need the same truthfulness and evidence as ads.
Explore More
Related analyses, benchmarks, and industry insights
Related Guides
Glossary Terms
Turn reading into a reaction
Oppira watches your competitors, keeps your playbook current, and drafts the response. See your market clearly by tomorrow.