By the end of this, you'll be able to:
- Turn a generic first draft into something you would actually send, in two or three rounds
- Pull the answers you need out of long documents, and check them
- Research a question with Claude and pressure-test the conclusion
- Run a pre-mortem, a steelman, and a red-team on your own plans
The first output almost never ships. That is how the process works, not a failure. Writers revise, designers iterate. With Claude, each revision cycle takes seconds, so iteration is essentially free. The bottleneck is knowing how to iterate.
The most common mistake: restarting from scratch when the first output misses. Don't restart. Redirect. Tell Claude what is wrong and what to fix, and protect the parts that work.
The three-round structure:
Round 1: voice and structure. Does it sound like you, is the flow right?
Round 2: specific sections. The corporate paragraph, the buried lead, the conclusion that fizzles.
Round 3: polish words. Kill jargon, tighten sentences, cut redundancy.
Each round gets more precise. By Round 3 you are finishing, not fixing.
The Language of Effective Iteration
"Make it better" is the most useless feedback you can give Claude. It says nothing about what is wrong or what to preserve. Effective feedback has three parts:
1. Name what's wrong, specifically. Not "it's a bit flat" but "the second paragraph sounds like a press release."
2. Say what to keep. "The opening paragraph is perfect. Don't touch it." Without this, Claude may fix one thing and break three others.
3. Give direction, not just diagnosis. "The conclusion feels rushed" is diagnosis. "Expand the conclusion to tie back to the opening story" is direction.
All three at once: "Keep the opening. Rewrite the corporate second paragraph in the tone of the first. Add a final sentence that returns to the opening question."
When to Push Back vs. When to Accept
Not every output deserves another revision prompt.
Push back when:
The output changes something you said to keep. ("I told you to keep the opening.")
Claude hedged when you asked for a direct take. ("You gave me both sides. Make a call.")
The revision fixed one thing and broke another. ("Go back to the previous structure and just fix the tone.")
Claude added length when you asked for compression. ("Cut to under 200 words.")
Accept and move on when: the big issue is fixed and what remains is small, each revision gains less than the last, or the output is good enough for the context. The goal is a final product, not a perfect iteration log.
Three Rounds in Practice
Start: "Write a cold email to a VP of Marketing at a 200-person B2B SaaS company offering a content strategy audit. Under 100 words." The output comes back generic.
Round 1: "Rewrite completely. Drop the pleasantries. Open with a specific observation about what their content is missing. End with a yes/no question." Better, but the middle buries the offer.
Round 2: "The opening is great, keep it. 'Content strategy audit' is vague. Replace with 2 deliverables: a 10-page gap analysis and a 30-day editorial calendar." Almost there; one sentence still uses "leverage" and "synergize."
Round 3: "Kill 'leverage' and 'synergize', use plain English. Change nothing else."
Total time: four minutes. Result: an email you'd actually send.
Start broad (voice and structure), then narrow (specific sections), then polish (individual words). Name what's wrong, protect what's working, give direction not just diagnosis. Two or three rounds of targeted iteration beats ten rounds of "make it better."