Content pipeline

The content pipeline turns signals into publishable content across channels.

Pipeline stages

Signals → Brief → Generate → Humanize → Queue → Approve → Publish

1. Brief

Claude reads up to 20 recent signals and writes a 1-paragraph summary with an editorial angle. The brief captures what’s happening and why it matters to the company.

2. Generate

From the brief, Claude generates content for each configured channel:

Channel	Format	Typical length
`linkedin`	Professional post with hook	200-300 words
`x_thread`	Thread of short posts	3-5 tweets
`facebook`	Social post	150-250 words
`blog`	Long-form article with headers	800-1500 words
`release_email`	Email with subject line	300-500 words
`newsletter`	Digest format	500-800 words
`yt_script`	Script with hook, sections, talking points	1000-2000 words

Generation is voice-aware. Claude reads the org’s voice profile (persona, audience, tone, brand keywords) and uses approved content as examples and spiked content as anti-patterns.

3. Humanize

The humanizer removes AI-generation patterns:

Significance inflation (“pivotal”, “landmark”, “game-changing”)
Vague attribution (“experts say”, “many believe”)
AI vocabulary (delve, tapestry, paradigm, leverage, nuanced)
Structural patterns (em-dash overuse, rule-of-three lists)
Promotional language (“revolutionary”, “cutting-edge”)

It then adds voice-specific texture — opinions, acknowledged gaps, concrete examples.

4. Queue

Generated content enters the queue with status queued. From here you can:

Approve — Move to approved, ready for publish
Spike — Reject (content becomes anti-pattern memory)
Edit — Modify headline or body text
Humanize — Re-run the humanizer
Schedule — Set a future publish datetime

5. Publish

Approved content is published to connected platforms:

LinkedIn — Via OAuth API
Facebook — Via page token
HubSpot — Via CMS API
Medium — Via publishing API
YouTube — Script upload with metadata

Running the pipeline

Full pipeline (scout + generate)

POST /api/pipeline/scout    → pull signals
POST /api/pipeline/generate → brief + generate + humanize

Or in one call via MCP:

pressroom_full_pipeline(org_id=1, channels=["linkedin", "blog"])

Generate from specific story

Instead of using the latest signals, generate from a curated story:

POST /api/stories/{story_id}/generate
{"channels": ["linkedin", "blog"]}

Regenerate a single item

POST /api/pipeline/regenerate
{"content_id": 42, "channel": "linkedin"}

Content memory

Pressroom learns from editorial decisions:

Approved content is included as positive examples in future generation prompts
Spiked content is included as anti-patterns (what to avoid)
This creates a feedback loop where content quality improves over time

Scheduling

Schedule content for auto-publish at a specific time:

POST /api/content/{id}/schedule
{"publish_at": "2025-01-15T09:00:00Z"}

The background scheduler checks for due content and publishes it automatically.

Email drafts

Content with channel release_email or newsletter can be composed into email drafts:

POST /api/email/drafts/compose
{"content_id": 42}

This creates a draft with subject line, HTML body, and text body ready for sending.