Steps for building an AI content assistant workflow
From How Joy Joya went from 10 hours to 1 of content brainstorming per week, featuring Laryssa Wirstiuk of Joy Joya, published on the AI Lab by ActiveCampaign.
Get the checklist
Before you start
- Create accounts on Slack, Make, Notion, and OpenAI (GPT-4o access required)
- Set up a dedicated Slack channel for content idea requests
- Document each client’s brand profile in Notion — brand voice, content pillars, product categories, current promotions, upcoming events
- Budget for a developer familiar with Make scenarios (Wirstiuk used Upwork) or plan to build the scenario yourself
The workflow
Phase 1: Set up the knowledge base in Notion
- Create a Notion page for each client with brand voice, content pillars, product categories, and editorial calendar
- Add current promotions and weekly events to each client page so the AI can reference time-sensitive context
- Structure the data consistently across clients so Make can pull the right fields automatically
Phase 2: Build the Make scenario (Slack to AI to Notion)
- Add a Slack module in Make that monitors new messages in your dedicated content channel
- Add an AI parsing step that extracts the client name and start-of-week date from the Slack message
- Add filters to ensure the Slack message contains the required information before continuing
- Connect Make to Notion to retrieve the client’s brand context (voice, pillars, promotions, events)
- Add a GPT-4o module with a structured prompt that generates social media content ideas tailored to that client and week
- Add a second AI step to clean output — enforce second-person voice, platform-agnostic language, and consistent formatting
- Add routers to handle different clients, content calendar structures, or unique brand rules
- Add iterators to process multiple idea objects individually
- Add break modules to prevent Notion API throttling
- Write each idea to a Notion database (“Content Idea Review”) with status set to “In Review”
Phase 3: Build the approval-to-calendar automation
- Create a second Make scenario that monitors the Notion database for ideas marked “Approved”
- Configure Make to create entries in the client’s content calendar when ideas are approved
- Set the idea status to “Transferred” after calendar entry is created
- Add a notification step to alert the team of the status change
Phase 4: Establish the review cadence
- Schedule biweekly review sessions for the team to review ideas in Notion
- Edit and refine AI-generated ideas before approving — expect roughly 3–5 approvals out of 10–15 ideas (~33% acceptance rate)
- Place approved ideas strategically in the content calendar with storytelling flow
- Apply final edits to all content before clients see anything
Quick reference
Total time: 2–4 weeks for initial build (with developer); ongoing maintenance as roster grows
Tools needed: Slack, Make, Notion, OpenAI GPT-4o or more current model, Upwork (optional, for developer)
Key output: A system that generates 10–15 on-brand social content ideas per client per week, saving 5–10 hours weekly
Content types generated: Instagram, Facebook, Pinterest, TikTok ideas; storytelling, educational, product highlights, behind-the-scenes, seasonal themes, email themes, blog prompts, SMS concepts
Want the full story?
Read How Joy Joya went from 10 hours to 1 of content brainstorming per week, featuring Laryssa Wirstiuk, published on the AI Lab by ActiveCampaign.
Related
More data from the AI Lab.