The same sentence in four boxes is not four posts
Most repurposing produces one paragraph pasted into four fields. It underperforms everywhere for opposite reasons: a LinkedIn opener reads as too corporate on X, and an X opener reads as unserious on LinkedIn.
The failure is not effort. A single generation cannot hold four conflicting instruction sets at once. Asked for all four in one call, a model converges on a neutral register that satisfies none.
Separating the platforms makes the output usable. Each gets its own generation with rules about length, opening line, link placement, and how much context a reader needs before the claim lands.
What stays shared is the substance. Claims worth repeating are extracted once before any post is written, so four different voices say the same thing.
How the ActionFlow runs on the canvas
The post enters as pasted text through Start or through a Scrappey node when you pass a URL.
A Generate Object node extracts the three claims worth repeating and returns them as a structured array. This step writes no copy. It ensures every platform generation downstream reads from the same three claims rather than rereading the full post and picking differently each time.
A Switch node routes to one branch per platform. Each branch holds its own Generate Text node with its own instruction block: character limit, opening style, hashtag policy, link placement. Keeping them as separate nodes rather than prompts in one node lets you change LinkedIn rules without touching X.
A Merge node collects the branches. The assembled pack goes to Notion as one page so a reviewer sees every variant beside the extracted claims, making inconsistencies visible at a glance.
A Slack node from the Human in the Loop category holds the pack for approval. Only after approval does an X (Twitter) node publish. Nothing reaches a public account without human release.
The extraction and writing steps can use different model services. A cheap, fast model is usually enough to pull three claims from a post, and the per-node cost is visible before the flow runs.
Nodes this ActionFlow uses
- Start: receives the post text. Scrappey (Util) fetches it instead when you pass a URL.
- Generate Object (AI Core): extracts the three claims worth repeating, as structure rather than prose.
- Switch (Control): routes to one branch per platform so each keeps its own rules.
- Generate Text (AI Core): one node per platform, each with its own limit and opening style.
- Merge (Control): collects the branches into a single pack.
- Notion (Integration): receives the pack with the extracted claims beside the variants.
- Slack (Human in the Loop): holds the pack for approval before anything publishes.
- X (Twitter) (Integration): publishes only after approval.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.