Claude Prompts for Social Media That Return Posts You Would Actually Publish
Type "write me an Instagram caption about our new feature" into Claude and you get something tidy and forgettable: an exclamation in the first line, three emojis, a question at the end and hashtags nobody searches. That is not a weakness of the model. Claude has no idea who you are, who follows you, what you shipped or how you talk, so it writes the average caption for the average brand.
The library below fixes that the way Anthropic's own guidance suggests: give Claude your context once, show it examples of your real writing, wrap each part of the prompt in a labeled tag, and say exactly what shape the answer should take. There are 30 prompts, numbered and grouped by job: brand voice set up in a Claude Project, post ideas, captions for each network, Instagram and LinkedIn, carousels, hooks, repurposing, replies to comments, analysis of a post export, a weekly review and an audit. Words in square brackets are what you replace. Every prompt is plain text you can paste into a Claude chat on any plan, including Free.
The examples use Streakly, an invented habit tracking app, and Alex, an invented founder; none of the outputs are real posts or results. If you plan a whole month, the ChatGPT prompts for a social media content calendar go deeper on calendars and work in Claude unchanged. For slide by slide carousel scripts, use the ChatGPT prompts for Instagram carousels. For what Claude can and cannot do on social media at all, plans and connectors included, read Claude for social media. This page sticks to the prompts and to why they work.
What Makes Claude Prompts for Social Media Work
Anthropic keeps one living reference for prompting its current models. Its prompt engineering overview sends readers to Prompting best practices, which covers clarity, examples, XML structure, roles and long documents. Five pieces of that guide matter most for social media work, and every prompt on this page is built on them.
Be clear and direct. The guide asks you to treat Claude like a brilliant new employee who does not know your norms yet, and offers a golden rule: show the prompt to a colleague with no context, and if they would be confused, Claude will be too. For a caption that means naming the network, the length, the reader, the one thing the post should make them do, and what to leave out.
Say why. The guide shows that a reason behind an instruction works better than a bare rule, because Claude generalizes from the explanation. "No exclamation marks" works once. "Our readers are tired developers who distrust hype, so keep the tone calm and plain" shapes every sentence that follows, including the cases your rule did not cover.
Show examples. Anthropic calls examples one of the most reliable ways to steer format, tone and structure, recommends three to five of them, asks that they be relevant and varied, and suggests wrapping them in example tags so Claude can tell them apart from instructions. Your own best posts are the best examples you have. Three of them teach Claude more about your voice than three paragraphs of adjectives.
Use tags, and put long material first. The guide says XML tags help Claude parse prompts that mix instructions, context, examples and inputs, recommends consistent descriptive tag names, and suggests nesting when content has a hierarchy. For long inputs such as a post export, it advises putting the data at the top and the question at the end, noting that queries at the end improved answers by up to 30 percent in Anthropic's tests on complex inputs, and asking Claude to quote the relevant parts first. One more formatting tip from the same page applies to captions: tell Claude what to do instead of what not to do, and match the style of your prompt to the style you want back. A prompt written in plain sentences gets fewer bullet points than one full of them.
Claude Prompt Formatting: The Tags Used in This Library
Tag names are not magic words; Claude reads any clear, consistent label. These are the ones the prompts below reuse, so a voice sheet saved once slots into every later prompt.
| Tag | What goes inside | Used in prompts |
|---|---|---|
| <brand> | What you make, for whom, and the one thing that sets it apart, in three or four sentences | 1, 2, 4, 26 to 30 |
| <voice> | The voice sheet from prompt 1: reader, tone, words you use and avoid, sentence length, emoji and hashtag habits | Every writing prompt |
| <examples> | Three to five of your own posts, each inside its own <example> tag | 1, 7 to 15 |
| <news> | Release notes, a changelog entry, a review, a support question or a number that moved | 4, 5, 8, 18, 19 |
| <post> | A finished post or draft you want rewritten, repurposed or checked | 16, 18, 19, 22 |
| <comments> | Comments or messages to answer, pasted with the commenter's first name removed if you prefer | 20, 21 |
| <data> | A post export or a pasted table of results, placed at the top of the prompt | 23 to 27 |
| <rules> | Hard limits: length, banned claims, links, hashtags, words to avoid | 2, 3 and any prompt that needs them |
| <output_format> | The exact shape of the answer: a table with named columns, a numbered list, or one block per network | Most prompts |
Tag style follows the XML guidance in Anthropic's prompting best practices. A Claude Project's instructions can hold the brand, voice and rules blocks so you paste only the news.
How to Set Up Claude as a Social Media Manager Before the First Prompt
Ten minutes of setup saves pasting the same context into every chat. Each step uses a Claude feature that exists on the plan named, checked on Anthropic's help pages today.
- Step 1
Create a Project Named After Your Brand
Anthropic's page on Projects describes them as self contained workspaces with their own chat histories and knowledge bases, available to every user including Free, where the cap is five projects. One project per brand or app keeps voices from bleeding into each other.
- Step 2
Paste the Voice Sheet Into the Project Instructions
Run prompt 1 in a normal chat, correct it, then paste the result and the rules from prompt 2 into the project's instructions. From then on every chat in the project starts with your voice loaded, and the writing prompts below only need the news.
- Step 3
Upload Project Knowledge That Changes Slowly
Good files for project knowledge: 20 of your best posts, your App Store description or product page, a changelog, a list of customer questions and last month's results. The same help page says paid plans switch to a retrieval mode when knowledge nears the context limit, expanding capacity by up to ten times. A plain chat holds a lot too: Anthropic's context window page puts most models at 200,000 tokens, about 500 pages, and several current models at 1 million tokens on paid plans.
- Step 4
Keep Fast Moving Material in the Chat
This week's release notes, a new review and the comments you want to answer belong in the message, inside a news or comments tag, not in project knowledge. Knowledge is for what stays true for months.
- Step 5
Turn Your Most Used Prompt Into a Skill
Once a prompt earns a place in your week, such as the caption pack in prompt 7, save it as a Skill so Claude loads it whenever you ask for captions, in any project. The steps are further down this page, and ready made ones are in our list of Claude skills for social media.
- Step 6
Connect a Tool Only When You Want Finished Posts
Everything on this page works without a connector. Add one when you want Claude to design the picture, save the draft or read your real numbers instead of a paste. The last section shows how that works with Spreadcast.
Brand Voice Prompts to Put in a Claude Project
Run these three once. Their results go into the project instructions and every later prompt leans on them.
1. Voice Sheet From Your Own Posts
Prompt: "<examples><example>[post 1]</example><example>[post 2]</example><example>[post 3]</example><example>[post 4]</example><example>[post 5]</example></examples> These are posts I wrote and liked. <brand>[what you make, for whom]</brand> Write a voice sheet I can reuse: who I am talking to, three adjectives each with a sentence quoted from my examples as proof, words and phrases I use, words I never use, typical sentence length, how I open and close a post, and my habits with questions, emojis, hashtags and the word we. Quote my examples for every trait and do not invent traits the examples do not show. Then write one short post about [topic] in this voice so I can check it." What it returns: a one page voice sheet built from evidence, plus a test post you can judge in seconds.
2. Project Instructions for Every Chat
Prompt: "Turn this voice sheet into project instructions for a Claude Project: [paste the corrected result of prompt 1]. Include: <brand>[brand, product, audience]</brand>, the networks I post on [list], how often [number] a week, and these rules: write in plain sentences, use the reader's words rather than ours, keep every claim to what is in my notes, write [check] where a fact is missing, and never invent statistics, quotes, customers or dates. Explain the reason for each rule in one line, because the reasons help you apply them. Keep the whole thing under 400 words." What it returns: instructions you paste once into the project settings.
3. Banned Claims and Words List
Prompt: "<brand>[brand]</brand> <rules>[anything legal, App Store or industry rules you must follow, such as no health claims]</rules> List the claims I must never make in a social post, the words that make posts in my category sound like ads, and five safer ways to say the things people in my category usually exaggerate. Format: three short lists. If a rule depends on a law or a platform policy, say which one and tell me to confirm it rather than stating it as fact." What it returns: a list to add to your project instructions, with the parts that need checking marked.
Claude Prompts for Social Media Post Ideas
Ideas fail when they are generic. These prompts force every idea to come from something real: a release, a question, a result.
4. Twenty Post Ideas From What Actually Happened
Prompt: "<news>[release notes, reviews, support questions, numbers and anything else from the last two weeks]</news> Give me 20 post ideas for [networks], each built on one item from the news. For each: a working title, the item it came from, the format it suits best (carousel, single image, screen recording, text post), and whether it needs a screenshot, a photo or only text. Skip any idea that could be posted by any app in my category. Output as a table." What it returns: ideas you can trace back to a source, so none of them is filler.
5. Ideas From Customer Questions
Prompt: "<news>[paste 10 to 30 questions from support email, reviews, comments or sales calls]</news> Group these questions by the worry behind them. For the five biggest groups, suggest one post that answers the worry directly, with the question quoted as the hook. Mark which answers need a screenshot of the app." What it returns: posts that answer what people already ask, which tend to be saved and shared because they are useful.
6. Content Pillars for a Small Account
Prompt: "<brand>[brand]</brand> <data>[your last 20 posts with one number each, such as saves or comments]</data> Suggest three to five recurring topics, called pillars, based on what this account already does well. Name each in two to four words, give the share of posts it should get, and list three ideas per pillar that only my brand could post. Tell me which current habit to stop." What it returns: pillars grounded in your own results. For dated monthly plans built on them, use the content calendar prompts.
Network Limits to Put in Your Caption Prompts
Claude knows roughly how each network works, but a stated limit beats a remembered one. These are the published limits for posting through each platform's own API, read today.
| Network | Limit to state in the prompt | Source |
|---|---|---|
| A caption of up to 2,200 characters, with at most 30 hashtags and 20 @ tags; a carousel holds up to 10 images or videos, and slides carry no captions of their own | Instagram Platform media reference | |
| Threads | Text posts of up to 500 characters, with emojis counted by their byte length, at most 5 links, and topic tags of 1 to 50 characters | Threads API posts guide |
| Write to the reader rather than to a limit: a long post is folded behind a see more link, so the point has to land in the opening lines | Your own feed; check how your last post displays | |
| A title and a description that describe what the Pin shows in search words, because people find Pins by searching | Pinterest's own pin builder shows the fields |
Paste the relevant row into the rules tag of prompts 7 to 11. A limit stated in the prompt is one Claude does not have to guess.
Best Claude Prompts for Social Media Posts: Captions for Each Network
Run these inside the project from prompt 2, so the voice is already loaded. Each asks for one network's native shape rather than one caption trimmed five ways.
7. One Update, a Caption for Every Network
Prompt: "<news>[what changed, who it helps and one detail only we could know]</news> Write one post per network for [Instagram, LinkedIn, Threads, Facebook Page, Bluesky]. <rules>Instagram: up to 2,200 characters, first line works alone, three to five hashtags at the end. Threads: under 500 characters, no hashtags in the body. LinkedIn: point lands in the first two lines, no links in the body. Facebook Page: two short paragraphs. Bluesky: one or two sentences.</rules> <output_format>One block per network, headed by its name, then a line saying which detail from the news you led with and why.</output_format>" What it returns: five native posts from one piece of news, with the reasoning visible so you can push back.
8. Instagram Caption in Three Lengths
Prompt: "<news>[what the post shows]</news> Write three Instagram captions for this picture: a short one of one line, a medium one of three short paragraphs, and a long one that teaches something in five numbered points. Each opens with a line that makes sense without the picture. End with a question only if a reader would genuinely want to answer it. Add alt text of one sentence describing the image for someone who cannot see it." What it returns: three options to choose from by mood, and alt text you would otherwise skip.
9. Threads Post That Starts a Conversation
Prompt: "<news>[topic]</news> Write five Threads posts under 500 characters each. Each makes one honest observation or asks one specific question that people in [audience] would argue about kindly. No links, no hashtags, no calls to action. Sound like a person, not a brand account." What it returns: short posts built to get replies, which is what Threads rewards.
10. Pinterest Pin Title and Description
Prompt: "The Pin shows [describe the image]. It links to [page]. Write five short Pin titles and a description of two or three sentences for each, using the words someone would type into Pinterest search when looking for this, such as [example searches]. No hashtags. Then say which title you would test first and why." What it returns: search worded Pins, which suit a network where people search rather than scroll a feed of friends.
11. Facebook Page and Bluesky Versions of a Long Post
Prompt: "<post>[a long LinkedIn or blog post]</post> Rewrite this for a Facebook Page as two short paragraphs a customer, not a peer, would read, and for Bluesky as one or two sentences with the single most useful point. Keep my voice. Do not add facts that are not in the original." What it returns: two shorter versions that keep the substance without copying the LinkedIn tone into places it does not fit.
Claude Prompts for Instagram Content Creation
Carousels, Reels scripts and the bio. For deeper carousel work, the carousel prompt library has 25 prompts that run in Claude as written.
12. Carousel Outline, Slide by Slide
Prompt: "<news>[topic or feature]</news> Outline an Instagram carousel of [6] slides at 1080 x 1350. Slide 1 is a cover headline of under eight words that promises one result. Slides 2 to [5] each make one point with a headline of under eight words and a body line of under twenty. The last slide says where to get [app] without shouting. For each slide, note the screenshot or visual it needs. Then write the caption, which carries the context the slides leave out." What it returns: a script a designer or a design tool can build, with every visual named.
13. Reel or Short Video Script
Prompt: "Write a script for a [20] second screen recording of [feature] for Instagram Reels. Give the on screen text for each beat, in lowercase, under six words per beat, with the second it appears. The first beat must show the result, not a logo. Add a caption of two lines and three hashtags. I will film the screen myself, so describe only what to tap." What it returns: a shot list and on screen text you can film in one take.
14. Instagram Bio in Five Versions
Prompt: "<brand>[brand]</brand> Write five Instagram bios short enough for the bio field. Each says who it is for and what they get, in the words those people use. One plain, one with a light joke, one leading with a number from my notes [number, if any], one aimed at [secondary audience], one under 80 characters. No emojis unless my voice sheet uses them." What it returns: bios to test for a month each, rather than one you second guess.
15. Story Sequence for a Release Week
Prompt: "Plan five Instagram Stories for the week [feature] ships: a teaser, the reveal, a how to in two frames, a poll and a thank you. For each give the text on the frame, the sticker to use, and what the picture shows. Keep each frame under 15 words." What it returns: a week of Stories that support the feed post instead of repeating it.
Claude Prompts for LinkedIn Posts and Profiles
LinkedIn rewards a clear point from a named person. These prompts write as you, not as a company page.
16. LinkedIn Post From a Lesson You Learned
Prompt: "<news>[a decision, mistake or result from your work this month, with the numbers you are happy to share]</news> Write a LinkedIn post in my voice. The first two lines state the lesson plainly. Then tell what happened in short paragraphs, with one concrete detail per paragraph. End with what I would do differently, not a question to the audience. No hashtags in the body, at most three at the end. Under 1,300 characters." What it returns: a post built around one lesson, the shape that suits a personal profile.
17. LinkedIn About Section Rewrite
Prompt: "Here is my current LinkedIn About section: [paste]. Here is what I want people to contact me about: [goal]. Rewrite it in first person in my voice, with the first two lines saying who I help and how, since the opening lines are what shows before the see more link. Then three short paragraphs: what I have done, how I work, how to reach me. Keep every claim from my original and invent nothing." What it returns: a sharper About section with no made up credentials.
Hook Prompts for the First Line
The first line decides whether the rest is read. Ask for many, then pick.
18. Fifteen Hooks, Five Kinds
Prompt: "<news>[the point of the post]</news> Write 15 opening lines for a post on [network], three of each kind: a plain statement of the result, a specific number from my notes, a question the reader is already asking, a small surprise, and a before and after. Under 12 words each. Mark the three you think are strongest and say why in one line each. Do not use stock phrases such as here is why or let that sink in." What it returns: a spread of hooks so you choose by ear, not by the first one Claude wrote.
19. Fix a Weak Hook
Prompt: "<post>[your draft]</post> The first line is weak. Explain in one sentence why a reader would scroll past it, then give five replacements that keep the same facts. Each must make sense to someone who has never heard of [brand]." What it returns: a diagnosis and options, which teaches you to spot the problem yourself next time.
Repurposing Prompts: One Piece, Many Posts
The cheapest post is one you already wrote. These turn long material into a week of short pieces without repeating yourself.
20. Blog Post or Newsletter Into Six Posts
Prompt: "<post>[your article, newsletter or long post]</post> Pull six separate points from this, each strong enough to stand alone. For each, write a post for the network it suits best from [networks], quoting or closely paraphrasing my own sentences. Spread them so no two consecutive posts make a similar point. List the source paragraph for each." What it returns: six posts with their origin marked, so nothing drifts from what you actually said.
21. Release Notes Into a Week of Posts
Prompt: "<news>[full release notes for version X]</news> Pick the three changes a user would notice and care about, in order of how much they change someone's day. Write one post per change for [networks], leading with what the user can now do rather than what we built. Drop internal fixes users would not notice. Say which change you dropped and why." What it returns: release posts that speak to users, which is where most changelog posts go wrong. The app update announcement examples show the pattern in real posts.
22. A Good Old Post, Rewritten for Today
Prompt: "<post>[a post that did well six months or more ago]</post> <data>[its result, such as saves or comments]</data> Write a new version that keeps what worked and changes the angle, the example and the first line, so it reads as new to someone who saw the old one. Update any detail that may have changed since, and mark it [check]." What it returns: a second life for a proven idea, with stale facts flagged.
Claude Prompts for Replies to Comments and Messages
Replies are where a voice is tested hardest, because they are short and personal. Keep a human on send for every one of these.
23. Reply Drafts for a Batch of Comments
Prompt: "<comments>[paste comments, one per line]</comments> Draft a reply to each in my voice. Sort them first into thanks, questions, feature requests, complaints and spam. Thanks get one warm sentence. Questions get the answer, or [check] if you do not know it. Feature requests get an honest note on whether it is planned, using only this list: [planned features]. Complaints get prompt 24. Spam gets no reply. Output as a table: comment, type, reply." What it returns: a sorted batch with drafts, so twenty comments take five minutes to review.
24. A Calm Reply to an Angry Comment
Prompt: "<comments>[the comment]</comments> <news>[what actually happened, from our side]</news> Write three replies of under 60 words: one that apologizes for the specific thing and says what happens next, one that asks for details privately, and one that corrects a factual error politely if there is one. No corporate phrases such as we value your feedback. Tell me which you would send and whether this needs a public reply at all." What it returns: options that lower the temperature, with a judgment on whether to reply in public.
Example Output: What Prompt 7 Returned for Streakly
Streakly is invented. The news pasted was: rest days shipped, a streak now survives a planned day off, users asked for it in reviews for months. The rows below were written by us to show the shape prompt 7 asks for; they are not real posts or measured results.
| Network | Length | First line | Why it led there |
|---|---|---|---|
| 4 short paragraphs and 4 hashtags | Your streak can now take a day off. | The user's result in seven words, readable without the picture | |
| 6 short paragraphs | We said no to rest days for two years. Our reviews changed our mind. | A founder's decision, which suits a personal profile | |
| Threads | 2 sentences | Shipped rest days. Turns out a streak that punishes a sick day is a streak people quit. | An honest observation that invites replies |
| Facebook Page | 2 paragraphs | Planning a day off? Your Streakly streak will wait for you now. | Plain customer language for a broader audience |
| Bluesky | 1 sentence | Streakly streaks now survive planned rest days, because habits should not break on a Sunday. | One point, nothing else |
Every post leads with what the user gets. None of them says excited to announce, which is the first thing a voice sheet built from real posts tends to remove.
Claude Prompt for Social Media Analysis: Getting the Numbers Into the Chat
Claude cannot see your analytics unless you give them to it, either as a file, a pasted table or through a connected tool. The quality of any analysis depends on what you paste: one row per post, with the date, network, format, the first line and the numbers you actually care about. Followers gained, comments and saves usually say more about a small account than likes.
Two official export routes are worth knowing. LinkedIn's help page on creator analytics says every member has post and audience analytics and can export them to an XLSX file with the Export button in the upper right; the date range defaults to the past 7 days, and a custom range is only available in the mobile app. On Instagram, Meta's help page on Meta One says exporting insights to a spreadsheet, as a CSV, is part of the Meta One Advanced, Expert and Max plans, for eligible public accounts, and that those plans also look back further than 90 days. Without that, copy the numbers from the professional dashboard into a sheet by hand; ten posts take a few minutes.
Upload the file to the chat or paste the table, wrap it in a data tag, and put the question after it, as Anthropic's long context advice suggests. Ask Claude to quote the rows it relies on. That one line turns a confident paragraph into an answer you can check against the sheet.
Claude Prompts for Analysis of a Post Export
Paste the data first, the question last. Each prompt asks Claude to show the rows behind every claim.
25. What Worked and What Did Not
Prompt: "<data>[paste your export or table: date, network, format, first line, and your numbers]</data> Using only this data, find which formats, topics and first line styles beat my median and which fell below it. For every finding, quote the rows it rests on. If a pattern rests on fewer than three posts, call it a hunch, not a finding. End with three things to do more of and one to stop, each tied to the rows." What it returns: an analysis you can verify, with small samples labeled as such.
26. Best Days and Times From My Own Data
Prompt: "<data>[posts with date, time, time zone and one number each]</data> Group these by weekday and by three hour block. Show the median result per group and how many posts each median rests on. Tell me whether any difference is large enough to act on, given the sample size, or whether I should keep my current slots." What it returns: an honest answer about timing, which for most small accounts is that consistency matters more than the hour.
27. Compare Two Networks
Prompt: "<data>[the same posts' results on network A and network B]</data> For posts published on both, compare how each did relative to that network's own median. Which kinds of posts travel well between the two, and which only work on one? Suggest how I should adapt posts for the weaker network, quoting the rows." What it returns: a reason to stop posting identical copy everywhere, backed by your own numbers.
Weekly Review and Instagram Audit Prompts
A short review every week beats a long audit every year. Run prompt 28 on Fridays and prompts 29 and 30 once a quarter.
28. The Friday Weekly Review
Prompt: "<data>[this week's posts and results]</data> <news>[what happened in the business this week]</news> Write my weekly social review in under 200 words: the best post and the likely reason, the weakest and the likely reason, one comment worth answering properly, and next week's three posts built on this week's news. Format: four short headed parts." What it returns: a habit that keeps the next week fed by real material. Save it as a Skill once the format feels right.
29. Claude Prompt for an Instagram Audit
Prompt: "<brand>[brand and audience]</brand> <data>[bio, the first lines of the last 15 posts, their formats and results, plus a description of the grid]</data> Audit this Instagram account as a new visitor would see it in ten seconds. Does the bio say who it is for? Do the last nine posts show a clear topic? Which posts would make someone follow, and which are only announcements? Give a score out of ten for clarity, consistency and usefulness, each with the posts that justify it, and five fixes ordered by effort." What it returns: an outside view of your profile, with every score tied to a post.
30. Full Social Media Audit Across Networks
Prompt: "<brand>[brand]</brand> <data>[for each network: followers, posts in the last 90 days, the three best and three worst posts with numbers]</data> Audit my social media as a whole. Where am I posting without results, where am I under posting with good results, and where does my voice drift from my voice sheet? Recommend which network to drop or pause, if any, and give a 30 day plan of no more than three changes. Quote the data for each recommendation." What it returns: a short list of decisions rather than a long list of tips.
Turn Your Best Claude Prompt Into a Skill
A Project holds context; a Skill holds a procedure. Anthropic's page on Skills describes them as bundles of instructions, scripts and resources that help Claude handle specialized tasks consistently, available on Free, Pro, Max, Team and Enterprise plans, with code execution turned on in settings. Claude loads a skill only when a request calls for it, and skills work across Claude in general, while project knowledge is always included in that one project's chats. So the voice belongs in the project, and the repeatable job, such as writing the caption pack from prompt 7 or the Friday review from prompt 28, belongs in a skill.
Anthropic's guide to creating custom skills says a skill is a folder with at least a skill.md file that begins with two required fields: a name of up to 64 characters and a description of up to 200 characters that tells Claude what the skill does and when to use it. Claude reads that description to decide when to load the skill, so write it the way you would ask: "Writes a caption for each social network from a product update, in the voice of the current project. Use for captions or social posts." Under the fields, paste the body of prompt 7 with its rules and output format. Zip the folder so the folder itself sits at the root of the archive, upload it, and turn it on in Customize, Skills.
For worked examples of the format, Anthropic's public skills repository holds sample skills, many under the Apache 2.0 license, which the README says are for demonstration and should be tested before you rely on them. The Claude for social media page covers which community skills are worth copying and what to check before installing one.
Video: What Skills Are in Claude
The Claude team's short explainer, useful before you save your first prompt as a skill.
Video: Prompting 101 From Anthropic
Anthropic's applied AI team builds a prompt live, using the same moves this library relies on.
Claude vs ChatGPT for Social Media Prompts
Every prompt on this page also runs in ChatGPT. The differences are in where context lives, what each can make, and how far each can reach into other tools. Facts from each vendor's own pages, read today. If you use Google's assistant instead, see Gemini for social media.
| Question | Claude | ChatGPT |
|---|---|---|
| Do tags help? | Yes. Anthropic's best practices recommend XML tags for prompts that mix instructions, context, examples and inputs | Also yes. OpenAI's prompt engineering guide suggests Markdown headers and lists to mark sections and XML tags to show where a document begins and ends |
| Where does the brand voice live? | A Project's instructions and knowledge; Free accounts can create up to five projects | A Project's instructions, which OpenAI's Projects page says override your global custom instructions; Free projects hold up to 5 files, Plus and Go 25, Pro 40 |
| Reusable procedures | Skills, on every plan with code execution on, loaded only when relevant | Project instructions and saved chats; this page does not compare custom GPTs |
| Can it make the picture? | No photos or illustrations, per Anthropic's help page on images; it can draw diagrams and charts with HTML and SVG and read images you upload | Yes. OpenAI's Images page says ChatGPT Images creates and edits images on all tiers |
| How much can one chat hold? | 200,000 tokens on most models, 1 million on several current models on paid plans | Depends on plan and model; check OpenAI's plan page |
| Can it save posts in another tool? | Custom connectors on Free, Pro, Max, Team and Enterprise, with Free limited to one, per Anthropic's connector guide | Full MCP support with write actions on Business and Enterprise or Edu on the web, per OpenAI's MCP help page; on Business only admins and owners add them |
| Price to start | Free, then Pro at $20 a month or $17 a month billed yearly, per Claude's pricing page | Free, with paid plans above it |
In practice: Claude is the stronger choice when the job is long brand documents, careful voice matching and reading exports; ChatGPT when you also want it to generate the image in the same chat. Both write better with the prompts above than with a one line request.
Best Claude Prompts for Social Media Marketing: Mistakes That Make Output Generic
Most disappointing results trace back to the same handful of habits. The first is asking for a post without news. A prompt with nothing real inside it can only return something general; prompts 4, 5 and 21 exist so there is always a fact at the center. The second is describing a voice with adjectives instead of showing it. Friendly, witty and professional describe half the internet. Three of your own posts inside example tags describe you.
The third is asking for one caption and trimming it for every network. A LinkedIn post cut down to Threads length reads like a summary of a LinkedIn post. Prompt 7 asks for each network's own shape and the reason for each opening, which is where the quality comes from. The fourth is trusting numbers Claude did not get from you. Every prompt here asks it to write [check] where a fact is missing; keep that rule in your project instructions and search the draft for it before posting.
The last is treating the first answer as the answer. Anthropic's guide describes a simple loop, draft, review against criteria, refine, as the most common way to chain prompts. In a chat that means following any prompt with "Check this against my voice sheet and rules, list every line that breaks them, then rewrite only those lines." It costs one message and catches most of what you would have fixed by hand.
Run These Prompts Through the Spreadcast MCP Server So Claude Saves Designed Drafts
A prompt library ends at text. Someone still has to make the picture at the right size, paste each caption into each network and pick a time. For people who make App Store apps, Spreadcast takes that last step from inside the same chat. It runs a remote MCP server at spreadcast.app/api/mcp. Add it in Claude under Customize, Connectors as a custom connector, sign in with Spreadcast and choose Make Drafts or Make and Publish on the consent screen. Drafts are the default. Claude Code and Cursor can use a key from Settings, AI Assistants instead, which stays draft only unless Can Publish Posts is turned on.
Then the prompts on this page do more. Prompt 7 with Spreadcast connected writes a caption for each network you connected and saves the post as a draft on your calendar. Prompt 12 comes back as a designed carousel of 3 to 8 slides at 1080 x 1350 in your app's brand, with real screenshots in an iPhone frame, shown in the chat. Spreadcast designs posts in 11 kinds, makes no AI photos and no video, and reads App Store moments such as a new version or a rating milestone, so prompt 21 can start from the release itself. Prompts 25 to 28 can read likes, comments and follower counts from your own posts instead of a paste; Instagram reach is not among them. The server also ships two ready prompts, Draft Posts From My App's News and Plan A Week Of Posts, listed wherever your assistant shows a connector's prompts.
Networks in Spreadcast are Instagram, Facebook Pages, LinkedIn personal profiles, Threads, Pinterest, TikTok, YouTube, Bluesky, Telegram channels, Discord channels and Mastodon. TikTok posts stay private until TikTok finishes auditing the app. Drafts and design are on the Free plan; Pro at $29 a month adds scheduling and publishing, and Studio at $79 a month covers 5 apps and 3 seats. Claude's custom connectors work on every Claude plan, with Free limited to one. Setup is on the MCP page, plans on the pricing page, and a full walk through is in schedule social media posts with Claude. If you do not make an app, every prompt above still works as text, and the social media caption generator page lists other ways to finish posts.
Frequently Asked Questions
What are the best Claude prompts for social media?
The ones that carry context: your brand in a tag, three to five of your own posts as examples, the news the post is about, the network's limits and the exact output format. Prompt 1 builds a voice sheet from your posts and prompt 7 turns one update into a caption for each network; those two cover most weeks.
Do Claude prompts need XML tags?
No, but they help when a prompt mixes instructions, examples and pasted material. Anthropic's prompting best practices recommend consistent, descriptive tag names such as instructions, context and input, and wrapping examples in example tags so Claude can tell them apart from instructions.
Can Claude analyze my social media data?
Yes, if you give it the data. Upload an export or paste a table with one row per post, put it at the top of the prompt, ask your question at the end and ask Claude to quote the rows it uses. LinkedIn exports creator analytics to XLSX; Instagram's CSV export is part of Meta One's paid plans.
Is Claude better than ChatGPT for social media?
For writing in a consistent voice from long brand documents and for reading exports, Claude is a strong choice, and Projects and Skills keep the setup reusable. ChatGPT can also generate images, which Claude does not. The prompts on this page run in both.
Can I use these prompts on the free Claude plan?
Yes. Every prompt works in a free chat, and Free accounts can create up to five Projects and use Skills, per Anthropic's Projects page and Skills page. Paid plans add an expanded project knowledge mode and larger usage.
Can Claude post to Instagram or LinkedIn for me?
Not on its own. Claude writes and analyzes; posting needs a tool connected through a custom connector. With Spreadcast connected, Claude can design the post, write captions per network and save drafts, and on Pro it can schedule and publish once you allow it.
How do I keep Claude in my brand voice every time?
Build a voice sheet from your own posts with prompt 1, put it and your rules into a Claude Project's instructions, and upload your best posts as project knowledge. Then save your most used writing prompt as a Skill so the format stays the same in every chat.
Where These Facts Come From
Every outside number, rule and price on this page links to the page it was taken from.
- Prompting best practicesClaude Docs
- Prompt engineering overviewClaude Docs
- What are projectsClaude Help Center
- What are skillsClaude Help Center
- How to create custom skillsClaude Help Center
- How large is the context window on paid Claude plansClaude Help Center
- Get started with custom connectors using remote MCPClaude Help Center
- Claude pricingAnthropic
- Can Claude produce imagesClaude Help Center
- Public skills repositoryAnthropic on GitHub
- Prompt engineering guideOpenAI Developers
- Projects in ChatGPTOpenAI Help Center
- Connecting custom MCP servers in ChatGPTOpenAI Help Center
- Images in ChatGPTOpenAI Help Center
- Instagram Platform media referenceMeta for Developers
- Threads API posts guideMeta for Developers
- View your creator analyticsLinkedIn Help
- Account and content insights on Instagram with Meta OneMeta Help Center
- Video: What are skillsClaude
- Video: Prompting 101Anthropic
Researched and written by Spreadcast. Last checked on October 17, 2026.