The Draft-to-Publish Workflow That Actually Works
“Marketing owns messaging and positions release notes. PMs write drafts, but marketing finalizes it. We need final positioning control.”
A Head of Product Operations laid this out during a recent conversation. Her company had a defined workflow: product managers create drafts, marketing refines and approves, then the content goes out. Any automation had to fit into this process, not replace it.
This is a more common pattern than you might think. And it’s exactly the right way to think about automating release communication.
The Handoff Pattern
At most companies of reasonable scale, release notes involve at least two roles:
The drafters: Product managers or engineers. They know what shipped and have technical context. They write the first version.
The finishers: Product marketing or communications. They know brand voice, positioning, and audience. They polish and approve.
This division exists for good reasons. PMs are closest to the work. PMMs are closest to how customers should hear about the work. Neither can do both jobs perfectly.
Why Automation Often Fails Here
When teams try to automate release notes, they make one of two mistakes:
Mistake 1: Bypassing the workflow entirely. The automation generates content and publishes it directly. Marketing never sees it. The result is technically accurate but poorly positioned, off-brand, and sometimes embarrassing.
Mistake 2: Creating more work. The automation generates drafts that are so rough they need complete rewrites. Marketing now has two jobs: writing from scratch AND managing the automation output. Net time savings: negative.
Both failures share a root cause: the automation wasn’t designed to fit the existing workflow.
What “Draft-to-Publish” Actually Means
A workflow-aware approach works differently:
Step 1: Automated draft generation. The system creates a first draft based on code changes, commits, PRs, and any available context. This draft is not published. It goes into a review queue.
Step 2: Human review and refinement. Marketing reviews the draft. They adjust positioning, refine language, add context, and fix any inaccuracies. The automation makes this step faster, not unnecessary.
Step 3: Approval and publishing. Once marketing approves, the content publishes to configured destinations: changelog, Slack, email, help center, wherever.
The automation handles the tedious part (gathering changes, creating initial language). Humans handle the judgment part (positioning, tone, what to feature).
Where the Time Actually Goes
Most people assume writing is the time sink. It’s not. Consider what creating a monthly release recap actually involves:
Discovery (40% of time): What actually shipped? Checking Jira, GitHub, Slack, asking engineers… reconstructing three weeks of context.
First draft (20% of time): Actually writing the initial version.
Review and refinement (25% of time): Getting feedback, making edits, agreeing on positioning.
Distribution (15% of time): Copying to different channels, formatting for different destinations.
If you automate only the first draft, you save maybe 20% of the total effort. If you automate discovery AND first draft AND distribution, you save 75%.
The draft-to-publish workflow focuses on automating everything except the 25% where human judgment matters most.
Marketing Control
Here’s what marketing needs to maintain control:
Nothing publishes without approval. This is non-negotiable. Automated systems make mistakes. Brand-sensitive content requires human sign-off.
Edits are easy. The automation should produce output that’s easy to change, not locked in a format that requires starting over.
Tone adapts. Over time, the system should learn from edits. If marketing always changes certain phrases, the automation should adapt.
Context is available. When reviewing a draft, marketing should see the underlying changes, not just the generated text. This helps them refine accurately.
Distribution is controllable. Marketing should decide which channels get which content. A Slack update might be appropriate immediately; a blog post might need more polish.
What Automation Should Handle
Given these constraints, here’s what makes sense to automate:
Change aggregation. Pulling together commits, PRs, deployments, and tickets into a coherent list. This is tedious and error-prone for humans.
Initial categorization. Bug fix, new feature, improvement, security update. Automated systems can do this accurately.
First-pass writing. A draft that captures the essential information in readable language. Not perfect, but a starting point.
Technical-to-customer translation. Converting engineering language to customer language. This is where AI shines: it doesn’t get attached to technical jargon.
Multi-channel formatting. The Slack version, the blog version, the email version. Same content, different formats.
Distribution mechanics. Actually pushing content to destinations once approved.
What Humans Should Handle
And here’s what should stay with humans:
Strategic positioning. How does this release fit into the broader product narrative? What story are we telling?
Priority decisions. Which changes deserve headlines? Which are footnotes? Which get omitted entirely?
Brand voice nuance. The automation can approximate tone, but the final voice check needs a human ear.
Accuracy verification. Did the automation misunderstand something? Is there context it doesn’t have?
Timing decisions. Should this go out now or wait? Is there a better moment for this announcement?
The Learning Loop
The best draft-to-publish systems get better over time:
Edits become training data. When marketing simplifies jargon or adjusts phrasing, the system learns the preference.
Rejected drafts inform future drafts. If certain types of changes consistently need heavy rewrites, the system adjusts.
Company context accumulates. Product positioning, customer segments, competitive framing: the system should incorporate this over time.
Individual preferences matter. Different editors have different styles. The system can adapt to who’s reviewing.
This learning loop is what separates a useful automation from a novelty. Without it, you’re manually training the system forever.
Making the Handoff Smooth
The handoff between automated draft and human review needs to be frictionless:
Clear status indicators. What’s awaiting review? What’s approved? What’s published?
Easy editing interface. Marketing shouldn’t need to learn a complex tool. The editing experience should feel like editing a document.
Change tracking. What did the human change from the automated draft? This feeds the learning loop.
Preview capability. Before publishing, see exactly how the content will appear in each destination.
Rollback options. If something goes wrong, quick recovery matters.
The Realistic Expectation
Here’s what a well-implemented draft-to-publish workflow delivers:
First month: Drafts need significant editing. Marketing is learning what the system produces. Time savings are modest.
Third month: Drafts are usable with minor edits. Marketing is making positioning tweaks and corrections, not rewrites. Time savings are real.
Sixth month: The system has learned enough that most drafts need light editing only. Marketing focuses on strategy instead of mechanics.
This system improves with use. Teams that expect perfection on day one will find themselves underwhelmed. Teams that invest in the learning loop will see compounding returns.
The Bottom Line
Marketing should own final positioning. That’s the whole point of having a marketing function. Automation doesn’t change this.
What automation changes is how much effort it takes to get to the point where marketing can do their job. Instead of PMs spending days assembling context and writing first drafts, they spend minutes reviewing what the system produces.
The draft-to-publish workflow respects organizational roles while eliminating busywork. Marketing keeps control. Everyone saves time.
If your PM-to-marketing handoff is a bottleneck, let’s chat about how Changebot generates drafts that fit your existing workflow while keeping marketing in control of final positioning.
