Social media automation for agencies across multiple brands
A single system that generates and publishes content for several businesses across three platforms, with optional human approval.
The problem
Creating, scheduling and posting content for several businesses across three platforms takes a great deal of manual effort. The agency wanted to coordinate it without a dedicated full-time person.
Challenges
Several brands, one system
Each business has its own tone, topics and audience, and content that suits one brand can damage another. The system had to keep each brand's profile separate and apply the right one every time it generates, so posts never blur together.
Three platforms with different rules
Facebook, Instagram and LinkedIn each expect different formats and offer different publishing routes. The build had to generate platform-specific variations and send each through the appropriate route, rather than posting one identical piece everywhere.
A platform limit on long-form articles
LinkedIn's API does not allow third-party tools to publish long-form articles. We could not automate that final step, so the system had to draft the article well enough that a person can paste and publish it quickly.
How the system works
Brand profiles
Each business has a stored profile in a database covering tone of voice, topics, target audience and platform preferences.
Scheduled generation
On a recurring schedule, the automation pulls the relevant brand profile and sends it to an AI model to produce platform-specific content variations.
Optional approval step
Content can go through a lightweight human approval before publishing, or run fully hands-off, depending on how much risk the business will accept.
Platform-specific publishing
Facebook and Instagram posts use official publishing tools. LinkedIn short posts and company page content go through a dedicated LinkedIn automation layer.
Long-form LinkedIn articles
LinkedIn's API does not allow third-party publishing of long-form articles. The AI drafts the article and a person pastes and publishes it manually, which is quick.
Design decisions
Brand profiles stored in a database
Tone of voice, topics, audience and platform preferences live in stored profiles that the generation step reads each time. Adding or adjusting a brand means editing a profile, not rebuilding a workflow, which keeps the system manageable as the agency grows.
Approval that can be switched on or off
Content can pass through a lightweight human approval before publishing, or run fully hands-off. We made this configurable because how much risk a business accepts on public posts differs from one client to the next.
Human publishing for long-form articles
Rather than looking for a workaround to LinkedIn's limit, we accepted it. The AI drafts the article and a person publishes it, which is quick and keeps the system inside the platform's rules.
Result in production
One system manages content for multiple businesses across three platforms. Automation covers the repetitive work, and people focus on approvals and long-form publishing.
One system covers content for several businesses across three platforms.
Automation handles the repetitive work of drafting and posting, so the agency does not need a dedicated full-time person for it.
People spend their time only on the parts that need them, namely approvals and long-form publishing.
What we would do differently
We would formalise the internal operations as small SaaS-style surfaces earlier, rather than composing n8n and Airtable. It ages better as the agency grows its headcount.
Who this pattern fits
Teams with similar volume, compliance or coordination challenges can use the same approach. The stack is stable, self-hosted where it matters and designed for a small team to operate.
Tools used
n8n
Workflow runtime and orchestration.
Claude API
Reasoning and structured outputs for content generation.
Supabase
Database, authentication and storage, including brand profiles.
Zernio
Publishing integration in the production stack.
Unipile
The integration layer used for LinkedIn.
Related
Frequently asked questions
Does every post need human approval?
Not necessarily. Content can go through a lightweight human approval step before publishing, or run fully hands-off, depending on how much risk the business is willing to accept. The choice can differ from one brand to another.
How does the system keep each brand's voice separate?
Each business has a stored profile covering tone of voice, topics, target audience and platform preferences. On a recurring schedule, the relevant profile is sent to an AI model, which produces platform-specific variations for that brand.
Can it post long-form articles to LinkedIn?
Not directly. LinkedIn's API does not support publishing long-form articles, so the system drafts them and a person publishes them manually, which takes minutes rather than hours.