I use AI constantly as we build Stravu. It is an invaluable assistant. But, its output needs editing, review, and approval. Here is how I work with AI and edit/review/approve its output.
The Dangers of Unreviewed AI Output
- Hallucinations are common. AI systems confidently generate information that sounds authoritative but is completely fabricated.
- Inaccuracies creep in constantly. Even when AI isn't hallucinating outright, it frequently gets things wrong. Yesterday, it matched the
- AI takes your content in unexpected directions. I had AI use a detailed outline for this blog post (I will include it below) to write a first version for me. It talked about all kinds of AI for all use cases, whereas I wanted the blog focused on working with AI on business content. I rejected this blog post, gave more context, and tried again.
- AI's length is off. The first blog AI wrote was a book. I like things shorter.
- Important! You and your team need to understand what you are writing and what is happening. Seriously.
Best Individual AI Workflow
The most effective individual workflow treats AI as a sophisticated drafting partner, not a final authority:
- Start with your own thinking. Before engaging AI, write down you goals, some key points, and open questions. Here is the initial goal I wrote down.
- Or sometimes get kickstarted with AI: Often AI brainstorming can help get me started as it did for this blog post, but I usually then throw away everything it wrote and move to the next step.
- Build your detailed outline. Build a full outline or bullet points or a prompt for what you want. Be clear about your goals. The more specific guidance you provide, the more useful the output will be. The example outline I used for this blog post is in the appendix.
- Have AI generate a first response.
- Accept what is good. Reject what is bad. Edit ruthlessly.
- Iterate until it's right. Use multiple rounds of AI generation and human editing to create content that truly meets your standards.
Scaling Review for Team Success
Team workflows require more coordination but produce better results:
- Harness diverse perspectives early. Have each team member brainstorm independently—treat AI as another team member contributing ideas, not the primary source. This prevents groupthink and ensures multiple viewpoints inform your content.
- Centralize and evaluate all inputs. Bring all brainstorms together in one shared space. Evaluate ideas collectively, identifying the strongest concepts and approaches.
- Build consensus on structure. Create a detailed outline that the whole team agrees on before any AI generation begins. This prevents scope creep and ensures everyone works toward the same goals.
- Use AI for expansion, not direction. Let AI flesh out your agreed-upon outline, but maintain human control over strategy and key messaging.
- Enable collaborative editing. Team members should add their expertise, refine AI-generated sections, and ensure the content reflects collective knowledge.
- Implement review processes. Have team members review each other's work, then use AI to identify potential issues or improvements you might have missed.
- Iterate as a team. Use feedback loops where human edits and AI suggestions inform each other, gradually improving the content through multiple rounds.
Practical Tools for Effective Review
- Red/Green Diff tracking with Approve/Reject makes it easy to see exactly what changed between versions, especially what edits AI made. This visual comparison helps teams understand the evolution of content and ensures important changes don't get lost.
- Undo functionality provides safety nets for experimentation. Teams can try AI suggestions knowing they can quickly revert if the changes don't work.
- History with Revert. If things get too messed up, revert to an earlier version.
These tools transform iterative collaboration with AI from a burden into an efficient process that improves both quality and team alignment.
The Bottom Line
AI is a powerful partner, but it's not a replacement for human judgment. Teams that build robust review and approval processes into their AI workflows produce better results, avoid costly mistakes, and maintain the collaborative culture that drives innovation.
The question is how to build review processes that are thorough enough to catch problems but efficient enough to maintain momentum. Get this balance right, and AI becomes a true force multiplier for your team's capabilities.
Appendix: Outline I gave AI
Why need to review and approve AI output
- Hallucinations
- Inaccuracies
- Took it in direction you didn't want
- Too much detail or too little
- Understand what is being said
- Whole team understand what has changed
Best individual workflow
- (sometimes: Use AI to brainstorm and get you started. Then throw away most of this)
- Write some hooks, context, bullets, an outline
- Have AI generate based on that
- Then review and edit that
- Iterate
Best team workflow
- Each team member brainstorm independently. AI can be another team member
- Bring brainstorms together in one place and evaluate
- Build an outline
- Have AI flesh out outline
- Team members edit and add
- Team members review and AI reviews
- Iterate
How to review and edit AI's output
- Red/Green Diff
- Undo
- History