Most “100 ChatGPT prompts for social media” lists look the same: “Write an engaging Instagram post about [product].” You paste it in, get “In today’s fast-paced world, everyone dreams of…” with three rocket emojis, and close the tab. The prompt is only half to blame. The model knows nothing about your business, your customers or how you usually write.

What works has two parts. Brand context gets written once and lives where the model sees it in every chat. A short prompt then sets the task, the constraints and the output format. With that setup, even a simple request returns a draft that needs five minutes of editing instead of a rewrite.

Below: a context template and 30 prompts by task — planning, copy, platform versions, comments and DMs, analytics, editing. Swap whatever is in square brackets for your own details. The prompts work in ChatGPT, Claude and Gemini.

Context first, then the prompt

In ChatGPT, the easiest place for context is a project or your custom instructions; other chatbots have equivalents. A minimal template:

Brand: [who you are, city, what you sell, average order value]
Customers: [who buys, 3 common questions, 2 main objections]
Voice: [formal or casual, 3 "we write like this / never like this" rules]
Samples: [your 3 best posts, in full]
Don'ts: [no promised results, no "game-changer", no 🚀]
Facts: [pricing, timelines, locations]. If a fact is missing, write [check].

That last line saves the most time: instead of an invented price, the model leaves a flag you’ll spot immediately. How to describe your voice briefly, without “friendly yet professional,” is covered in our post on tone of voice.

Planning and ideas (1–5)

  1. Content pillars. “Suggest 5 content pillars for the month. For each: its job (reach, trust or sales), a sample topic, and how many times a week.”
  2. Topics from DMs. “Here are 20 customer questions from this month: [list]. Group them and give two post topics per group.”
  3. Objections. “List 10 reasons [customer] doesn’t buy [product]. For each, a post topic that addresses the doubt without pressure.”
  4. Monthly hooks. “List content hooks for [month] for a [niche] in [city]: holidays, seasonal customer problems, local events. Flag what needs two weeks of prep.” Double-check holiday dates; models mix them up.
  5. Weekly grid. “Lay out these 8 topics across one week: 3 posts, 2 Reels, Stories daily. Explain the order.”

Copy and formats (6–15)

Posts

  1. Hooks. “Give me 10 opening lines under 100 characters for a post about [topic]: a number, a mistake, a question, a scene, a surprising fact. Mark the three strongest.”
  2. Post from bullet points. “Turn my notes into an Instagram caption: [notes]. Under 900 characters, one call to action at the end: [CTA]. Don’t add facts that aren’t in my notes.”
  3. Sales post. “Structure: the customer’s situation → what it’s costing them → how we fix it → the offer. Product: [description, price]. No ‘act now’ or ‘amazing deal.’”
  4. Case study. “Project notes: [before, what we did, results with numbers]. Open with the result, then the path, and end with who this approach suits.”
  5. Carousel. “Break this text into an 8-slide carousel: a headline under 8 words and up to 30 words per slide. The second-to-last slide is a cheat sheet.” What makes carousels worth saving is in our Instagram carousel guide.

Other platforms

  1. LinkedIn. “Rewrite this as a LinkedIn post: the first two lines must work on their own, short paragraphs, no hashtag wall, end with a question.”
  2. Threads and X. “Compress the idea to 500 characters for Threads and 280 for X. No hashtags, end with a question.”
  3. Reels. “A 30-second script as a table: seconds, on screen, on-screen text, voiceover. Voiceover under 75 words, shot on a phone by one person.” Six more video prompts are in our post on AI video scripts.
  4. YouTube. “A title under 70 characters starting with [search phrase], plus a description: two paragraphs, then chapters from this outline: [outline].”
  5. Spanish version. “Adapt this post for Spanish-speaking customers in [country]: local examples, prices in local currency, tú or usted per our voice. Don’t translate word for word.” Leave the final proofread to a native speaker.

Comments and DMs (16–20)

  1. Negative review. “A customer wrote: [review]. Give three replies: acknowledge the problem, explain without excuses, offer a fix. Under 400 characters, no ‘we apologize for any inconvenience.’”
  2. “How much?” “A DM reply: price range [prices], what drives the final cost, one clarifying question. Under 300 characters.”
  3. Saved replies. “Here are 10 common questions and our answers: [list]. Turn them into DM templates in plain language, under 250 characters each.”
  4. Comment triage. “Here are 50 comments from this month: [list, names removed]. Sort them into questions, complaints, praise and spam. Which questions repeat?” Repeats are ready-made post topics.
  5. Cool down first. “Rewrite my reply calmer and shorter, same meaning: [draft].” Handy when you’re typing angry.

Analytics (21–25)

  1. Patterns. “A table of posts from the last two months: date, format, topic, reach, saves, shares. Find 3 patterns and 3 hypotheses to test. Don’t draw conclusions from groups under 5 posts.”
  2. Formats. “From this file, calculate median saves per 1,000 reach for Reels vs. carousels. Show the formula so I can check it.” Models are bad at mental math, so have them calculate from an uploaded file.
  3. Report. “Turn these numbers into a half-page report for my manager: what went up, what went down, why, and what we’ll do next. No jargon.”
  4. Two hooks. “Two posts on the same topic reached [X] and [Y]: [texts]. How do the opening lines differ? Suggest a third version to test.”
  5. Post-mortem. “This post reached half our usual audience: [text, time, format]. Give 5 possible reasons and how to check each.”

Editing (26–30)

  1. Clichés. “Find filler and phrases any brand could use. Suggest specific replacements.”
  2. Cut a third. “Shorten by 30% without losing facts or examples. Cut throat-clearing and repetition first.”
  3. Customer’s eyes. “You’re [customer description], scrolling on the train. Where would you stop reading, and why?”
  4. Fact list. “List every claim with a number, date or promise in this text.”
  5. Voice check. “Compare this post with the samples in the context. Show three places where the voice slips and fix only those.”

Rules for checking any output: verify numbers, dates and prices against your own documents, because models are confidently wrong. Two bad answers in a row mean the context is missing a sample or a fact; don’t rephrase the prompt a fifth time. Ask for three versions instead of one “perfect” one. Keep your best prompts in a doc next to an example of the output: in a month you’ll have 15–20 of your own, and they’ll beat any list, this one included.

When prompts won’t help

  • You have nothing to say. Without facts, cases and a point of view, the model produces polished emptiness at any volume.
  • Trends and real-time moments. The model can’t see today’s feed; the “trending audio” it suggests may have died months ago.
  • Personal stories. Only you know the details, and fakes read as fakes.
  • Health, finance, legal. A specialist checks every claim, not a “double-check yourself” prompt.
  • Personal data. Don’t paste customer phone numbers, names or chat logs into a general-purpose chatbot; anonymize before analysis.

A chatbot has one weak spot: the context lives apart from your price list and goes stale the moment prices change. In Repost21, the brand knowledge base plays that role: Google Docs, Sheets or Notion connect live, Luna drafts posts from current data in your voice, and finished posts go straight onto the content calendar. Try it free.