25 ChatGPT Prompts for Social Media Managers (2026)

Most lists of ChatGPT prompts for social media managers are written for someone who treats Instagram as a side hobby. They give you "write a caption about my product" prompts and pretend that is the job.

The job is bigger than that. You manage a calendar across four platforms, draft 15 captions a week without sounding like a robot, answer DMs that range from "where is my order" to "I hate your brand," and explain to a director why reach dropped 22% in March. This list covers prompts for all five of those slices, written the way you would actually paste them into a chat.

You can copy any prompt straight into ChatGPT. Each one includes the role, the task, the constraints, and the output shape the model needs to do something useful. The bracketed inputs are what you fill in with your actual brand, audience, and goals.

A few things to get out of the way before you start.

ChatGPT is good at the writing work: variant captions, hook variants, reply drafts, post calendar scaffolding, analytics commentary you can hand to a director. It is bad at anything that requires live data: current platform algorithm rules, ad costs, follower counts of named accounts, or what is trending today. The free tier will confidently invent any of these if you ask. Use a real social-listening tool, the platform itself, or your own analytics for those.

Do not paste anything into the free tier you would not paste into a competitor's group chat. That means no unreleased campaign briefs, no client roster, no private DM transcripts, no internal channel screenshots. The free tier of ChatGPT may use your inputs for training. For work that matters, use a paid plan with chat history disabled or an enterprise-licensed tool.

Replace the bracketed inputs with real brand voice notes, real audience descriptions, and real campaign goals. Vague inputs produce vague captions, and you can already write vague captions on your own.

Caption writing and copy variants

Five prompts for the daily caption work.

1. Write a caption from a product brief

Prompt: "You are a senior social media manager for [brand, e.g., 'a DTC mid-priced skincare brand for women 28-45']. Brand voice: [3-5 adjectives + 1 banned word]. Platform: [Instagram / TikTok / LinkedIn / X]. Write a caption for this post: [describe the product or moment, e.g., 'launch of a new vitamin C serum, 15% concentration, $42, in stock now']. Constraints: under [character count], lead with a hook in the first 7 words, one clear CTA at the end, zero buzzwords. No question marks unless the question is genuine. Output the caption only, no commentary."

The "one banned word" line is load-bearing. Most brand voice notes are abstract until you pin one word the brand explicitly does not use ("never say 'magical'"). That word does more to anchor the voice than three positive adjectives.

2. Generate 5 caption variants for A/B testing

Prompt: "Below is an approved caption. Write 5 variants for testing, each varying one element: (1) hook style only, (2) CTA only, (3) length only (cut by 30%), (4) tone shift to more direct, (5) tone shift to more playful. Constraint: do not change the substantive offer or any of the facts in the caption. Brand voice tag: [1-line voice description]. Approved caption: [paste]. Output as a numbered list with a 1-line label for each variant."

Use this when you have a winning caption and want to push it further. The "do not change facts" instruction prevents the model from inventing a discount or feature that does not exist.

3. Rewrite a caption to land under a character cap

Prompt: "Rewrite the following caption to fit under [X] characters including spaces. Keep the substantive offer and the CTA identical. Cut every word that does not earn its place. No em-dashes. No buzzwords. Caption: [paste]. Output: the trimmed caption only, then the new character count."

Hand this prompt to ChatGPT when the social tool flags "caption too long for first comment" or when an X (Twitter) post needs to land under 280. It will outperform a manual edit at 9 PM on a Friday.

4. Adapt one caption across five platforms

Prompt: "I have one caption that performed well on [source platform]: [paste]. Adapt it for each of: Instagram (max 220 chars), LinkedIn (max 700 chars, more professional register), X/Twitter (max 280 chars), TikTok (max 150 chars, casual), Threads (max 500 chars, conversational). Preserve the core offer. Change only voice, length, and platform conventions. Output as five labeled blocks."

This is the prompt that turns one good caption into five posts. Build it into your repurposing workflow.

5. Add accessibility text for an image-heavy post

Prompt: "Write image alt text and an in-caption description for this post: image of [describe the image factually, e.g., 'three women laughing on a beach holding the new product, golden hour lighting'], primary message of the caption is [the main point]. Alt text constraints: under 125 characters, lead with the visible subject, no marketing language. In-caption description constraints: 1-2 lines, written for someone who cannot see the image but is reading the caption. Output: Alt text, then In-caption description, each in its own labeled block."

Most social posts ship without proper alt text. This prompt removes the friction.

Content ideation and calendar planning

Five prompts for the planning work that fills the calendar.

6. Generate 30 post ideas for a single product

Prompt: "I need 30 social media post ideas for [product or service], audience [describe], brand voice [1-line]. Mix the formats: 6 educational, 6 behind-the-scenes, 6 customer-led (UGC, testimonials, replies), 6 product-led, 6 entertainment or culture moments. For each: a 1-line concept, the platform best fit, and the role it plays in the funnel (top / middle / bottom). Constraints: each idea must be specific enough to brief a creative team in 30 seconds. No 'share a quote' or 'ask a question' filler. Output as a markdown table."

The format distribution forces variety. Without it, the model defaults to 25 educational variants and 5 product posts.

7. Build a monthly content calendar from a campaign brief

Prompt: "Build a 4-week social content calendar for [campaign goal, e.g., 'driving newsletter signups for a new B2B SaaS product']. Platforms: [list]. Weekly cadence: [X posts per platform]. Audience: [describe]. Required moments: [list, e.g., 'product launch Tuesday Week 2, founder Q&A Wednesday Week 3']. For each post slot: day, platform, format, hook angle, KPI it serves. Output as a 4-column table grouped by week."

This is the prompt that turns a campaign brief into a real calendar you can hand to a designer.

8. Repurpose a long-form piece into a content series

Prompt: "I have a [long-form asset, e.g., 'a 2,400-word blog post on our pricing teardown of three competitors']. Asset summary: [2-3 sentences]. Generate a 7-post series that repurposes the key claims into platform-native content. Distribution: 3 LinkedIn posts (longer, professional tone), 2 X threads (each 5-7 posts), 1 Instagram carousel (8 slides), 1 short-form video script (under 60 seconds). For each: format, hook, the specific data point or argument from the source it draws on, and the CTA. Constraint: do not invent statistics not in the source asset."

Stops you from rewriting the same blog post manually 7 times.

9. Pitch 5 reactive post angles to a news moment

Prompt: "A news event just happened: [1-2 sentence description]. My brand is [brand and category]. My brand voice is [1-line]. Generate 5 post angles a brand in my category could ship within 24 hours without looking opportunistic. For each: the angle in one line, the platform fit, why this brand has standing to comment, the highest risk of the angle, and a 1-line check we should do before posting. If none of the five clear that bar, say so."

The "if none clear the bar, say so" line stops the model from being a yes-man. Most news moments are not opportunities, and an honest answer matters more than five forced ones.

10. Plan a launch week thread on X

Prompt: "Plan an 8-post launch thread for X for [product/launch]. Audience: [describe]. The arc: post 1 hook (under 240 chars, no link), posts 2-7 build the case (mix of context, problem, solution, proof, feature, social proof), post 8 the soft CTA + link. For each post: the post text, character count, image/video suggestion in brackets. Constraints: no em-dashes, no exclamation marks outside post 1, no questions in posts 2-7. Output as 8 labeled blocks."

Threads on X are a craft. Pre-planning the arc beats writing as you go.

Hooks, openers, and shortform video

Five prompts for the work that determines whether anyone reads anything else.

11. Generate 10 hook candidates for one post

Prompt: "Write 10 hook candidates for a social post about [the topic]. Audience: [describe]. Platform: [Instagram / TikTok / LinkedIn / X]. Mix the angles: number-led, contrarian, result-led, curiosity-gap, problem-statement, observation. Constraints: each hook is under 12 words. No questions. No 'You won't believe.' No 'In today's world.' Pick your top choice and explain in 2 sentences. Output as a numbered list, then the top pick."

Most posts die because the hook does not work. Generate ten, pick one.

12. Tighten an opening line that is not working

Prompt: "Here is the opening line of a post: [paste]. The audience is [describe]. The platform is [X]. The opening is supposed to do [describe, e.g., 'make a busy founder stop scrolling']. Diagnose in 2 sentences why it might not be doing that. Then write 5 stronger candidates that fix the specific weakness. No clickbait. No questions. Output: Diagnosis, then 5 numbered candidates."

The diagnose-before-rewrite step is what makes the new versions actually better instead of just different.

13. Write a 60-second short-form video script

Prompt: "Write a 60-second script for [platform: TikTok / Reels / Shorts] about [topic]. Brand: [brand]. Audience: [describe]. Hook in the first 3 seconds, payoff by second 50, CTA in the last 10. Format as labeled timecodes (0:00, 0:03, 0:08, 0:15, 0:30, 0:45, 0:55) with the spoken line and a 1-line note on the visual. Constraint: under 130 words total. No buzzwords. No 'today we are going to.'"

The timecode format forces the model to think about pacing instead of dumping a wall of text.

14. Salvage a post that flopped

Prompt: "Here is a post that underperformed on [platform]: [paste post]. Engagement: [paste numbers if you have them, or describe how it flopped]. Brand voice: [1-line]. Diagnose the most likely reason for underperformance, ranked: hook, payoff, format, timing, audience-fit, topic-fit. For each ranked reason, name the specific change that would help. Then rewrite the post with the top-ranked change applied. Do not propose changes that require new content or new creative."

This is the prompt that turns a postmortem into a v2 instead of a Slack thread of theories.

15. Generate caption-hook pairings for a content batch

Prompt: "I have 5 captions written for [platform]. For each, write 3 alternative first-line hooks I could swap in: one number-led, one observation, one contrarian. Constraint: each hook is under 10 words and works as the visible preview before the 'more' fold on [platform]. Captions: [paste 5 captions, separated by ---]. Output as 5 labeled blocks, each with 3 numbered hooks."

Use this for batch days. Three hook options for five captions is fifteen variants in a single prompt cycle.

Community management and DM/reply drafts

Five prompts for the work that lives in the inbox.

16. Draft 5 reply variants for a recurring DM type

Prompt: "I receive this type of DM repeatedly: [paste an example DM, anonymized]. Brand voice: [1-line]. Draft 5 reply variants for it, ordered from most formal to most casual. Constraint: each reply is under 40 words, includes one clear next step (link, answer, expectation), and never asks for information the customer already gave. No starting with 'Thanks for reaching out.' Output as a numbered list with a 1-line label for each variant."

Banning the boilerplate "Thanks for reaching out" opening pushes the model to write replies that sound like a person.

17. Handle a complaint in public comments

Prompt: "A customer left this complaint as a public comment on our [platform] post: [paste comment, anonymized]. The complaint is [valid / partly valid / not valid]. Our position is: [your 1-line position]. Draft a public reply that: (1) acknowledges the specific issue, not 'feedback,' (2) does not over-promise, (3) moves the rest of the conversation to DM if more detail is needed, (4) reads in our brand voice. Constraint: under 50 words. No 'we hear you' template language. Output the reply only."

The "not 'we hear you'" guardrail is critical. That phrase has been hated by real customers since approximately 2018.

18. Write a comment-section moderation note

Prompt: "We need to post a pinned moderation note on our [platform] page after [describe situation, e.g., 'a launch post that attracted a lot of off-topic political comments']. The note should: (1) explain the rule we are enforcing, (2) not lecture, (3) leave room for genuine engagement on the original topic. Brand voice: [1-line]. Constraint: under 60 words. No threats. No 'we reserve the right.' Output the note only."

This is the post that prevents the comment section from spiraling into the next 48 hours.

19. Triage a backlog of DMs by intent

Prompt: "Below are 15 recent DMs we received on [platform]. Cluster them into 4 buckets: (1) Needs a real reply from a human, (2) FAQ-answerable (I will send a template), (3) Sales lead worth flagging, (4) Spam/troll/ignore. For each DM: which bucket and a 1-sentence reason. DMs: [paste, numbered]. Output as a table."

Saves you 20 minutes of triage every morning.

20. Generate an influencer outreach DM

Prompt: "Draft an outreach DM to a small creator I want to gift product to. Creator: [name, niche, follower count, why they fit our brand]. Brand: [1-line]. The ask: gift them [product], no posting required, with an invitation to post if they want. Constraints: under 80 words, opens with a specific reference to one piece of their content (I will paste it in), zero 'we love your content' filler, ends with a clear next step. Reference content: [paste 1-line description or link]. Output the DM only."

Generic outreach DMs get ignored. The specific-reference instruction is what makes the difference between a 5% and a 40% response rate.

Analytics commentary and reporting

Five prompts for the work that justifies your role to the director.

21. Translate a metrics table into a stakeholder summary

Prompt: "Below is our monthly social performance: [paste a table of metrics including followers, reach, engagement rate, top-post performance, click-throughs, by platform]. Audience for the summary: [role, e.g., 'a Head of Marketing who has 5 minutes']. Write a 1-page summary with: (1) the single most important finding in 25 words, (2) the 3-4 supporting data points with numbers, (3) what changed vs last month and the most likely reason, (4) one thing we recommend doing differently next month. No buzzwords. No hedging like 'engagement was solid.' Output as 4 labeled paragraphs."

The "single finding in 25 words" cap is enforceable and prevents the model from leading with three competing findings.

22. Diagnose a sudden reach drop

Prompt: "Our [platform] reach dropped from [X] to [Y] between [time period]. Other context: [what we noticed, e.g., 'engagement rate held steady, posting cadence was the same, no algorithm announcement that week, top post was a Reels in both periods']. List 6 plausible hypotheses for the drop, ranked by how testable they are with the data I already have. For each: the hypothesis, the specific number or chart that would confirm or rule it out, and what we should do if it is true. Do not assume the drop is permanent. Do not reassure."

The "do not reassure" line stops the model from defaulting to "this is normal variance, do not worry."

23. Write a 2-sentence top-post commentary

Prompt: "Below is our top-performing post from last month: [paste post and metrics]. Below are 4 nearby posts that underperformed: [paste with metrics]. Write a 2-sentence commentary explaining why the top post worked: one sentence on what it did differently from the underperformers, one sentence on what is repeatable for next month. No buzzwords. No 'engagement was strong.'"

Two sentences forces precision. Most "top post commentary" in real reports is three vague paragraphs that say nothing.

24. Draft a competitor-content scan summary

Prompt: "I monitored 5 competitor accounts on [platform] over the last 2 weeks. Below are their highest-engagement posts: [paste post titles, formats, and metrics for each]. Write a summary covering: (1) the dominant content format across competitors, (2) any topic angle they are all working that we are not, (3) any topic angle we own that none of them touched, (4) one specific opportunity for us this month. No comparative headcount or follower-count claims. Output as 4 labeled paragraphs."

This is the prompt that turns 90 minutes of competitor monitoring into a 5-minute briefing.

25. Build a quarterly performance narrative

Prompt: "Below are 12 weeks of social performance data: [paste]. Goal we set at the start of the quarter: [paste]. Write a quarterly narrative for a board-level reader: (1) did we hit the goal (yes/no/partial, with the number), (2) the 3 most important wins with numbers, (3) the 2 biggest misses with the honest reason, (4) what we are doing differently in the next quarter. Constraint: under 350 words. No 'in summary.' No 'overall.' Tone: direct, no hedging."

Use this in the second-to-last week of the quarter so you have time to fix the misses before the deck is final.

What Prompt 2 actually returns (tested 2026-05-21 for the 2026-05-20 slot)

We ran Prompt 2 on free ChatGPT with a concrete brief baked in: [approved caption: 'New Vitamin C 15% Serum is here. We tested it for 90 days against the leading drugstore brand. Brighter skin, no irritation, 32% less hyperpigmentation. $42. In stock now. Tap to shop.'; brand voice tag: 'direct, results-led, no fluff, never use the word "magical"']. Reproduced verbatim:

Here are 5 controlled variants, each changing only one element from the approved caption.

Hook Style Only "90 days. One serum. Measurable results. New Vitamin C 15% Serum is here. We tested it against the leading drugstore brand. Brighter skin, no irritation, 32% less hyperpigmentation. $42. In stock now. Tap to shop."
CTA Only "New Vitamin C 15% Serum is here. We tested it for 90 days against the leading drugstore brand. Brighter skin, no irritation, 32% less hyperpigmentation. $42. In stock now. Get yours today."
Length Only (30% Shorter) "New Vitamin C 15% Serum is here. Tested for 90 days against the leading drugstore brand. Brighter skin, no irritation, 32% less hyperpigmentation. $42. Shop now."
Tone Shift: More Direct "New Vitamin C 15% Serum delivers results. In a 90-day test against the leading drugstore brand, users saw brighter skin, no irritation, and 32% less hyperpigmentation. $42. In stock now. Tap to shop."
Tone Shift: More Playful "Your glow-up just showed up. New Vitamin C 15% Serum is here. After 90 days testing against the leading drugstore brand: brighter skin, no irritation, 32% less hyperpigmentation. $42. In stock now. Tap to shop."

Editor's notes. The model held the load-bearing constraint cleanly: the 90-day test, the 32% number, the $42 price, and the in-stock status are preserved without invention across all five variants. The model also held the banned-word check (no "magical" anywhere) and avoided em-dashes throughout, which is rare for the default register. The four things to fix before shipping: (1) variant 3 was supposed to cut length by 30% and cut only about 19% (original 37 words, variant 30 words), so a follow-up "tighten by another 6 words" is still needed; (2) variants 2 and 4 sit so close to the original caption that they are not meaningful A/B test variants, with the CTA-only swap moving from "Tap to shop" to "Get yours today" not the dramatic CTA experiment a real test would want, so push the model to "give me a CTA that does not include a verb-to-shop pattern" for a sharper variant; (3) the playful variant ("Your glow-up just showed up") is the boldest of the five but uses the word "glow-up" which has crossed from fresh to overused in skincare copy in 2025, so a senior copywriter would want one more pass; (4) the model added an unrequested preamble "Here are 5 controlled variants, each changing only one element from the approved caption" which is a minor format violation against the prompt that asked for the numbered list only, easily ignored but worth noting because it shows up in every batch run if not banned explicitly. Net: usable as a starting set, but the variants are conservative and a second cycle ("push variants 2 and 4 further from the original") is where the testable A/B options come from.

Tips for getting better results

Three things that matter more than any specific prompt.

Use specific brand voice anchors. The single biggest determinant of whether ChatGPT writes a usable caption is whether your voice description names one banned word and one signature move. "Direct, casual, smart" produces beige captions. "Direct, never uses 'journey,' always leads with a number" produces captions you can actually ship.

Ask for 10, pick 1. ChatGPT is faster at generating variants than picking the best one. Use it for the volume, use your taste for the selection. The reverse does not work.

Treat the first answer as a draft. Even with a strong prompt, the first response is the safer, more generic version. Push back: "Make variant 5 more contrarian." "Cut the introduction." "The hook is too soft for our audience." That second cycle is where the useful version comes from.

FAQ

Is ChatGPT good enough to write social media captions on its own?

Not quite. It is reliable enough to draft variants you can pick from, and excellent at length-tightening, platform adaptation, and hook brainstorming. It is not reliable at writing a caption that lands the specific tone of voice your audience knows you by, and it cannot invent the specific detail or anecdote that turns a generic post into a memorable one. Treat it as a senior copy assistant who needs an editor, not as a solo writer.

Should I paste my analytics into ChatGPT?

Not into the free tier. The free tier may use your inputs for model training, and platform analytics tied to a named brand or client are competitive intelligence. For real analytics work, use a paid plan with chat history disabled, an enterprise-licensed tool your company has approved, or anonymise the numbers before pasting (rename platforms, scrub follower counts, strip dates).

Can ChatGPT keep up with platform algorithm changes?

No. ChatGPT does not have live access to current platform algorithms or recent announcements. It will give you historical best-practice advice that may be 6-18 months out of date. For current algorithm guidance, use platform-published creator resources (Meta Creator, X Creator, TikTok Creator Portal), social-listening tools, and recent industry posts. Use ChatGPT for the writing work, not for the strategy.

What is the single most useful prompt for a junior social media manager?

Prompt 7: build a monthly content calendar from a campaign brief. Most junior SMMs are handed a campaign goal and expected to invent the calendar. The bottleneck is structure, not creativity. Use Prompt 7 to get the scaffold, then bring your own taste to the specific posts.

How do I tell whether a ChatGPT caption is good?

Read the first 7 words out loud. If they sound like every other brand on the platform, the caption is not good, regardless of what comes after. The hook is 80% of the post on a feed-based platform. Captions that earn the scroll-stop are the captions that lead with something specific, surprising, or numbered.

What to do next

If you only try one prompt from this list, try Prompt 11 (generate 10 hook candidates). The hook is where most social posts live or die, and Prompt 11 fixes the biggest weak point in most brand accounts in a single chat session.

If you run a team, bookmark Prompts 7 and 21 (monthly calendar and metrics-to-summary). Together they turn the two highest-friction recurring tasks into 10-minute jobs.

Share one prompt with a teammate today. The team that uses a shared prompt library is the team that ships faster.

How to use these ChatGPT prompts for social media managers