AI Social Media Manager
Plans, writes, and schedules a month of posts in one sitting.
Overview
A pipeline that takes a brand's positioning and a few ideas and turns them into a full content calendar: captions, hooks, hashtags, image prompts. Then it schedules everything. What used to be a weekly scramble becomes a quick review of a month you can approve in one go.
The problem
Small teams know consistent posting grows their audience, but writing something for every platform every day is the first thing that slips. Agencies charge thousands a month for what is mostly a repeatable process.
And generic AI captions read as generic. The hard part isn't producing text. It's staying on brand and native to each platform, day after day, without someone babysitting it.
The solution
I split it into stages. One step works out the themes from the brand profile, another writes versions tuned to each platform, and a third scores every draft against the brand and format rules before anything gets queued.
Approved posts go into a scheduler that publishes at sensible times per platform. A person stays in the loop through a single board where they approve, tweak, or ask for a regenerate, instead of writing from a blank page.
Architecture
A staged pipeline (strategy, then writing, then a brand check, then scheduling) running in n8n, with a Next.js board as the place a human signs off.
- Prompts tuned per platform so posts read native, not copy-pasted
- A brand-voice check that blocks off-brand drafts before scheduling
- A scheduling queue with retries so nothing double-posts
- The whole calendar approved from one board
Screens & diagrams
Key features
A plan, not one-offs
Works out content pillars first, so posts ladder up to something instead of being random.
Native to each platform
Writes a different version per channel, with the length and hook each one rewards.
A brand-voice gate
Every draft gets scored against the brand and format rules before it can be scheduled.
Approve in one go
Review a month at a glance and approve, edit, or regenerate without leaving the page.
Implementation highlights
- Kept strategy, writing, and review as separate stages, so I can tune or swap any one without breaking the rest.
- Built the scoring gate to reject off-brand drafts on its own, which cut the time spent reviewing right down.
- Made scheduling safe to retry, so a failed publish never turns into a double post.
- Rate-limited every platform integration with backoff to stay inside their quotas.
Challenges solved
Getting past 'clearly AI'
A brand profile plus a few examples plus the scoring gate lifted the output from obviously generated to something you'd approve with a glance.
Scheduling that doesn't misfire
Publishing has side effects, so idempotency keys and a durable queue mean a post goes out exactly once, even under retries.
Every platform is different
Each API has its own limits and formats, so an adapter layer keeps those quirks away from the core pipeline.
Technologies
Future improvements
- Feed engagement back in so it learns what actually lands
- Generate the images and drop them straight into the calendar
- Multiple brands in one workspace for agencies