The short version
- Claude social media work goes best when Claude has a standing brief: who you are, who you speak to, how you sound and words you never use.
- Claude Projects keep that brief and your reference material in one place, so every new chat starts on brand.
- Specific prompts that name the platform, format, audience and one message beat open requests like “write a post about our launch”.
- Through MCP, Claude can use a design tool directly and create real, editable designs rather than describing them.
- Gradiently connects to Claude as a remote MCP server, so Claude can search Marks, read your brand and make designs that open in the Studio.
On this page
For Claude social media work, treat Claude as a writer you brief once, not a slot machine you pull every morning. Set up a Project holding your voice, audience and banned words, ask for specific posts with the platform and format named, and connect a design tool over MCP so Claude can lay its words onto real designs. Done that way, Claude writes posts that sound like you and arrive at the right size, ready for a human edit.
Set up a Project with your house style
Claude's Projects let you store instructions and files that apply to every chat inside them. That is the single most useful thing you can do for social media, because the brief you would otherwise retype each time is always there. Upload a few of your best past posts, your brand guidelines if you have them, and write project instructions along these lines.
We are [brand], a [what you are] for [audience].
Voice: [three words, e.g. calm, precise, warm]. Short sentences. British spelling.
We never: use exclamation marks, say 'revolutionary', promise results, invent numbers.
We always: say one thing per post, end with one clear action.
Platforms we use: Instagram (portrait 1080x1350, Stories 1080x1920),
LinkedIn (1200x1200 or 1080x1350), X (1600x900).
When asked for a post, return: the words on the image (under 8 words),
a supporting line, the caption, alt text. Ask before inventing any fact.If you want help writing the voice part, brand guidelines your AI assistant can follow walks through it. The aim is rules Claude can apply, not adjectives it has to guess at.
Prompts that get usable posts from Claude
With the house style in place, the prompt itself can be short, but it should still be specific. The difference shows most on the second draft: a specific prompt gives Claude something to correct against.
Vague
- Write a post about our new service.
- Make it more engaging.
- Give me some LinkedIn ideas.
- Write a caption for this photo.
Specific
- Write a LinkedIn portrait post announcing same day repairs to office managers. One message: booked before 10, fixed by 5.
- Cut the hook to six words and open with the benefit, not our name.
- Give me eight LinkedIn post angles about onboarding mistakes, each as a single line I could put on an image.
- Here is a photo of our new studio. Write alt text, then a two sentence caption inviting people to the open evening on Thursday.
Prompt starter
Why it works
Prompt starter
Why it works
Prompt starter
Why it works
Prompt starter
Why it works
Prompt starter
Why it works
There are fifty more in AI prompts for social media, grouped by format.
From words to designs: connecting Claude over MCP
Writing is half a post. The other half is a design at the right size, with text that reads. Claude can do that half too when it is connected to a design tool through MCP, the Model Context Protocol, an open standard for giving assistants real tools. What MCP is explains it without jargon, and Claude's custom connectors are how you add one, where your plan allows.
Gradiently works inside Claude this way. With a scoped API key, created by a workspace owner or admin in Settings › API and agents, Claude can search Marks, read your workspace and brand kit, and create designs that open in the Studio as ordinary editable pages, without spending Gradiently credits. Each design wears a Mark rather than a stock gradient (once you claim one, only you can export with it), and the Mark moves its calmest region behind the words so they stay readable. The full walkthrough is designing with Claude through Gradiently.
- 1
Create a scoped key
In Gradiently, open Settings › API and agents and create a key with only the permissions Claude needs. Name it after where it is used.
- 2
Add Gradiently as a connector
Add the remote MCP server to Claude with the address shown in Settings. Technical details are in the developer docs.
- 3
Ask for designs, not descriptions
“Make Monday's post as a LinkedIn portrait design and an Instagram Story in our brand.”
- 4
Review in the Studio
Open each design, read every word, adjust anything and export the sizes you need.
A LinkedIn portrait post written and laid out by Claude from one prompt.
The Story version from the same request: same words, recomposed for the tall frame.
A weekly routine that holds up
- Monday, 15 minutes: ask Claude for the week's plan against your content pillars. Swap anything that feels off.
- Monday, 30 minutes: have Claude draft every post and create the designs through your connector.
- Monday, 15 minutes: read each design in the Studio. Fix facts, tighten one line in your own words, export.
- Through the week: schedule as normal. Note which posts did well and paste that note into the Project.
- Friday: ask Claude to compare what worked against your pillars and suggest next week's angles.
Where Claude needs a human
Claude is reliable at structure, variants and editing against rules. It cannot know what happened in your shop today, which customer story is worth telling or what your audience is tired of hearing. Keep those choices yours, and let Claude do the drafting, resizing and checking around them. That split is also the honest answer to whether your posts will start sounding generic: they will only if the ideas come from the model rather than from you. Brand consistency with AI covers the longer term.
Questions people ask
Is Claude good for social media posts?
Yes, particularly for drafting, rewriting for different platforms and checking posts against your voice rules. Give it a standing brief in a Project and real facts in each prompt.
Can Claude make images for social media?
Claude does not generate photos or illustrations the way image generators do, but connected to a design tool over MCP it can create real designs. Gradiently is one such tool: designs Claude makes open in the Studio for editing.
How do I connect Gradiently to Claude?
As a workspace owner or admin, create a scoped API key in Gradiently under Settings › API and agents, then add Gradiently as a custom connector in Claude. Setup details are in the developer docs.
What should I put in a Claude Project for social media?
Your audience, voice rules, words you never use, the platforms and sizes you post to, a few of your best past posts, and an instruction to ask rather than invent facts.
Written by Gradiently
The team behind Gradiently, a design tool built around Marks: living gradients that make everything you design look like yours.
See our profile