Welcome to Prompt, Tinker, Innovate, my AI playground. Each edition gives you one hands-on experiment and the prompt to run it.
This week's playground: Have AI diagnose your writing. You do the rewriting.
You've seen the prompt guides going around. "Rewrite this so it sounds like a real person." "Strip out the AI tells." "Make it hit."
They hand the draft back to the machine and ask it to fix itself.
And you get back something that reads differently bad. Shorter sentences. A few fragments dropped in for texture. A studied casualness that reads like a LinkedIn post doing an impression of a person. The em dashes are gone but the writing is still empty.
Here's the thing nobody says out loud: the tell was never the em dash. The tell is that the writing doesn't know anything. No numbers, no name, no moment where somebody did something and it went badly. AI can't add that, because you never gave it to AI in the first place.
So a rewrite pass can't save you. What you need is a diagnosis.
Use case spotlight: the editor who marks but doesn't touch
Think about the best editor you ever had. They didn't rewrite your paragraph. They put a line in the margin and wrote "prove it," "you're stalling," or "this is the only good sentence on the page, start here." My favorite, from one of my Forrester editors: “WTF.”
That editor made you better because they made you do the work. The fix came out of your head, so it sounded like you.
That's the job to give AI. Instead of asking it to rewrite, ask it to show you what's wrong and get out of your way.
Every prompt below follows that rule. The machine diagnoses. You revise.
Your AI experiment: match the prompt to what's actually broken
This is not a seven-step process, so please do not run your Slack update through seven prompts.
When something is wrong with your draft, pick the prompt that finds it. Most drafts need one pass. A big piece might need two.
Find the symptom and take the prompt:
| What's bugging you | Grab this |
|---|---|
| It's too long and I can't tell what to cut | 1. The Deletion Audit |
| It sounds confident but says nothing | 2. The Specificity Interrogation |
| The paragraphs are fine, the piece isn't | 3. The Reverse Outline |
| Nobody's getting past the first line | 4. The Opening Cut |
| It's technically correct and it isn't mine | 5. The Voice Fingerprint |
| I don't know if this was worth writing | 6. The Hostile Reader |
👉 Time to tinker: Grab a draft you're not happy with—something you wrote with AI help, or something you wrote cold that came out flat. Pick your symptom. Paste it in.
If you want to address two issues, do the big ones first. The Reverse Outline and the Hostile Reader tell you whether the piece survives at all. There is no sense auditing sentences in paragraphs you're about to delete.
📝 Prompt: The Deletion Audit
You are my editor. Do not rewrite anything. Do not improve anything. Go through this draft sentence by sentence and list every sentence that could be deleted without losing meaning, evidence, or argument. For each one, name what it was doing: throat-clearing, restating the sentence before it, hedging, transition padding, or summarizing something I already said. Then tell me the total word count I'd save if I cut all of them, and what percentage of the draft that is. I'll decide what goes. Just show me the list. Text: [paste or attach]
Most drafts come back somewhere between 20% and 40% deletable. That's a thinking problem, not a writing problem, and now you can see exactly where it's hiding.
📝 Prompt: The Specificity Interrogation
You are a skeptical editor. Find every sentence in this draft that makes a claim with no number, no name, no example, and no consequence attached to it. Do not invent the missing detail. Do not fill the gap yourself. Instead, ask me the question that would pull it out of me. Format: [the vague sentence] → [the question you'd ask me] Then stop. Text: [paste or attach]
AI wants to fill the gap. When it fills the gap, it fills it with air. Make it interview you instead.
📝 Prompt: The Reverse Outline
Reverse-outline this draft. Give me one sentence per paragraph that states what that paragraph argues, not what it's about. Then show me the outline as a plain numbered list, so I can read the argument without the prose in the way. Tell me three things: where the argument repeats itself, where it jumps a step, and which paragraph is doing no work at all. Change nothing in the draft. Text: [paste or attach]
Often, a "voice problem" is a structure problem. You can't fix the flow of a piece whose argument doesn't hold. Sentences are the last thing to fix, not the first.
📝 Prompt: The Opening Cut
Read this draft and find the first sentence that could only appear in this piece, and no other piece on this topic by any other writer. Everything before that sentence is warm-up. Show me what you'd cut. Then show me the draft starting from that line, with the smallest possible edit to make it stand as an opening. If no such sentence exists anywhere in the piece, say so plainly. Text: [paste or attach]
That last instruction is the important one. If the draft doesn't contain a sentence worth opening with, I want to know. I don't want AI inventing one for me.
📝 Prompt: The Voice Fingerprint (two steps, worth it)
Step one: Here are things I wrote before I ever used AI. Do not compliment them. Extract the rules of how I actually write: typical sentence length and how much it varies, how I open, how I close, my punctuation habits, words I reach for, words I never use, how I handle transitions, how much I hedge, how often I use "I." Give it to me as a list of rules an editor could enforce. Samples: [paste or attach at least 3 samples] Step two: Now hold this draft against those rules. Show me a table: [my line] | [the rule it breaks] | [the fix]. Leave alone anything that already follows the rules. Do not smooth. Do not polish. Do not add. Text: [paste or attach]
The rules list is the real asset here. Save it or better yet, drop it into a custom instruction, GPT/Gem, or a Skill. It's a voice guide built from evidence instead of adjectives, and it is more useful than "write in a friendly, professional tone."
📝 Prompt: The Hostile Reader
You are the reader I'm most worried about: [describe them, e.g. "a CFO who has sat through nine AI pitches this quarter and approved none of them"]. Read this draft. After every paragraph, write the one thing you'd actually be thinking. Be blunt. Be unfair if that's honest. Mark the exact line where you'd stop reading, and say why. At the end, tell me the one sentence that made you keep going. If there wasn't one, tell me that. Do not edit anything. Text: [paste or attach]
Every other prompt improves the writing. This one tells you whether the writing was worth doing.
The rewrite prompt, and its leash
You want the one that actually rewrites. Here it is, and it comes with conditions.
"Make it sound human" fails because it gives AI nothing to aim at. It swaps its own defaults for a different set of defaults: casual instead of formal, but still not you.
That's why this one is the only prompt on the menu with a prerequisite: run the Voice Fingerprint first. Without your rules list, this is "make it sound human" with extra steps.
📝 Prompt: The One Rewrite You're Allowed
You may rewrite this draft, under four conditions. 1. You may only cut, tighten, and reorder. You may not add a single new claim, example, statistic, or flourish. If a sentence is thin, leave it thin. That's my job to fix, not yours. 2. Every change must trace back to one of these rules about how I write: [paste your rules list from Prompt 5]. 3. Show me a change log before the rewrite: [what you changed] | [which rule] | [why]. If you can't name the rule, don't make the change. 4. Do not smooth the parts that already sound like me, even if you think you could improve them. Text: [paste or attach]
The change log is the key. Every line is one you can reject. You stay the editor. AI stays the hands.
💡 Pro tips: Make the prompts work harder
- Add the read-aloud test when a sentence bugs you and you can't say why: "Flag every sentence I couldn't say out loud in one breath in a normal conversation without sounding like I was reading. Show me where the breath runs out. No rewrites."
- Push back when it goes soft. AI wants to be encouraging. If a diagnosis comes back gentle, say: "You were being nice. Do it again, and this time assume I have thick skin and a deadline." The second pass is usually the more honest one.
- Do the revision yourself. This is the whole point. Do not let AI apply the fixes.
What did you discover?
Run the Deletion Audit and tell me your number. What percentage of your draft turned out to be air?
Now try the harder one. When the Specificity Interrogation asked you a question, did you already know the answer? Or did you discover that your draft was vague because you didn't know the specific yet?
I'd love to hear what surprised you most. Drop it in the comments.
📩 Not subscribed yet? Hit the button at the top. Next week, we'll try another experiment.
