How to Automate Social Media: A Weekly Workflow That Runs Without You

A founder running four accounts (X, LinkedIn, Threads, Instagram) and posting once a day on each is publishing about 120 posts a month. Typed into native apps, that’s somewhere around 15 hours of copying, cropping and logging in. Most of those hours have nothing to do with having something to say.

The workflow below gets that down to about two hours a week. It uses three tools: Claude to draft, SchedPilot to publish, and n8n to run the parts that repeat on a clock. None of it is exotic, and the order you build it in matters more than which tools you pick.

Decide what gets automated before you buy anything

The single most useful rule I’ve found comes from automationsocialmedia.com, a practitioner site that reads the actual platform rulebooks instead of guessing at them. Their split is three-way: automate distribution, assist production, keep conversation human.

That one sentence settles most tool decisions.

Scheduling and publishing can run with nobody watching. Drafting can be done by AI as long as a person reads it before it goes out. Replies, DMs, complaints and anything that happens during a bad week stay with a human, permanently. The platforms agree with this, too. When the site went through the rules for X, LinkedIn, Meta, TikTok and Pinterest, none of them banned automation outright. What they target is automated engagement (auto-likes, auto-follows, bot replies) and the same content repeated at high frequency.

So the workflow here automates publishing completely, automates drafting partially, and doesn’t touch the inbox at all. If you want automated replies, this isn’t your article, and honestly I think you’d regret building it.

The stack, and why it’s these three

Claude is the drafting layer. It’s good at turning a brief or a blog post into platform-shaped drafts, and it holds a voice reasonably well once you give it a style file with ten or so of your real posts.

SchedPilot is the publishing layer, and it’s the piece that makes the whole thing work with an AI assistant. It ships an MCP server, so Claude can connect to it directly and list your accounts, upload media, create and schedule posts, and pull analytics without you opening a dashboard. It also has a regular API for n8n to call. Pricing starts at $21 a month, and the $99 plan covers 100 accounts, which matters if you post for clients.

That last bit is the reason SchedPilot sits in the middle instead of being one option among many.

n8n is the clock. Anything that should happen on a schedule without you starting it (a new blog post going out as social posts, a Friday analytics pull) lives here. Self-hosted, it costs whatever your server costs. A $6 VPS is plenty.

Why not wire Claude straight to each platform’s API? Because getting publishing access is slower than people expect. LinkedIn routes posting through a vetted API product. TikTok keeps posts from unaudited apps private. Pinterest starts new apps in a sandbox where only the creator sees the pins. Letting a scheduler that already holds approved connections do the publishing skips months of applications, and automationsocialmedia.com has the platform-by-platform rules if you want the details.

The weekly workflow, step by step

This is the cycle. It assumes one person, a handful of accounts, and roughly a post a day per platform.

Monday morning: 20 minutes of audit and brief

Open last week’s numbers and look at two things only: which posts got replies, and which formats got ignored. Ignore follower counts. They move too slowly to steer a week by.

Then write the brief. Mine is usually five lines in a note: the one thing I want people to know this week, two pieces of source material (a blog post, a customer call, a product update), and anything that should not be posted about. That last line matters more than it looks. A queue that doesn’t know about a bad news week will post cheerful product content straight through it.

Monday, still: draft the week in one Claude conversation

Paste the brief and the source material into Claude along with your voice file, and ask for the week’s posts per platform. Be specific about counts: five X posts, three LinkedIn posts, five Threads posts, two Instagram captions.

Don’t ask for one post and four reformats. Ask for each platform separately, because the formats don’t translate. Threads caps text at 500 characters. A LinkedIn carousel is really a PDF document post. An Instagram carousel crops every slide to match the first image. One payload copied everywhere is how automated accounts end up looking automated.

Expect to throw out a third of the drafts, which is normal and still faster than writing from zero.

The approval step (don’t skip this one)

Read every post before it gets queued. Edit, cut, rewrite the hook. This is the only step in the workflow that needs your actual judgement, which is exactly why it’s the one people skip once things feel routine.

Here’s the reason it’s not optional.

SchedPilot’s MCP server exposes eight functions, and its delete function refuses to remove a post that has already been published. An agent can publish in a second and can’t unpublish at all. The irreversible action is the one the AI is fastest at, so the human gate goes right before it, never after.

Queue it from the same conversation

Once the drafts are approved, tell Claude to schedule them through SchedPilot. Something like: “Schedule the approved X posts at 9:00 and 17:00 local time, Monday to Friday, and the LinkedIn posts at 8:30 on Tuesday, Wednesday and Thursday. Use the images I uploaded to the media library.”

Claude calls the account list, uploads or picks the media, and creates each scheduled post. You get a summary back and can check the calendar in SchedPilot if you want to see it laid out. Total time from approved draft to full queue is a few minutes.

The MCP server has no reply, comment or DM functions, so an assistant wired to it can’t touch your inbox however you prompt it (I’d count that as a feature).

The n8n pipelines: things that run without a Monday

Two pipelines are worth building once the weekly cycle feels boring.

The first is blog-to-social. n8n watches your RSS feed. When a new post lands, it sends the article to Claude’s API with your voice file and asks for platform drafts, then posts those drafts to a Slack channel (or an email) with approve and reject buttons. Approved drafts go to the SchedPilot API and get scheduled into the next open slots. Rejected ones get dropped. You still review everything, but you review it in Slack in 30 seconds instead of starting a new writing session.

The second is the Friday report. n8n pulls analytics for the week’s published posts through SchedPilot, has Claude summarise what did well and what flopped in plain sentences, and drops the summary into your inbox. Monday’s audit then takes five minutes because the reading is already done.

Build blog-to-social first, since that’s where the hours are, even though the report is easier.

Friday afternoon: 15 minutes of review

Skim the report. Check that nothing failed to publish, because failures are per-post and usually silent. Kill anything still queued that no longer fits the week. Then close the laptop.

What it costs

For one person running four or five accounts, the monthly bill looks roughly like this:

Compare that with per-channel pricing elsewhere. Buffer charges $5 per channel, which is cheap at four channels and starts to sting at forty. If you’re an agency, the 100-account plan is the line that changes the maths.

Prices change often in this category. Check them before you commit to anything annual.

What I’d never put in this workflow

Replies to complaints. DMs. Anything posted during a crisis, a layoff, or a news event you haven’t thought about. Auto-likes and auto-follows, which are a separate business from publishing and the exact thing X and LinkedIn wrote their rules against.

Also: anything opinionated that carries your name and that you haven’t read twice. AI drafts are confidently wrong often enough that “I’ll skim it later” turns into a post you have to explain.

And here’s my one stronger opinion. Most people who say they want an “AI agent” for social media actually want a scheduler plus a good drafting session. Agents that decide what to post on their own sound great in demos, but for an account with fewer than ten thousand followers the decision space is small enough to write down as rules. Write the rules. Keep the gate.

Where to go next

If you want to go further than this setup, automationsocialmedia.com is the resource I’d point you to first. It covers the five-layer automation stack, what each platform’s API will and won’t publish, the posting ceilings that break queues (Instagram’s 24-hour cap, YouTube’s upload quota), and how to keep a human in the loop once volume picks up. Their guide to social media automation workflows and the order to build them pairs well with the cycle above, and the walkthrough on connecting an AI assistant to your social accounts covers the MCP setup in more detail than I have here.

Start with one account and one week of posts you’ve already written. Queue them through SchedPilot from a Claude conversation, read every one before it goes, and only build the n8n pipeline once that cycle has run for a month without surprising you.