Ship Whole Projects With Reverse Prompting

Reverse prompting is the one move that flips prompting on its head. Instead of you writing a perfect prompt, you make the AI interview YOU and write the prompt itself. The result is a project that ships in one shot, not 47 prompts and a half broken output.

What You'll Learn

• The exact reverse prompting template you paste into any Claude or ChatGPT session before you start a project
• Why your projects keep coming back wrong even when you think the prompt was clear, and the one structural fix that catches every blind spot
• The 12 question interview Claude runs on you so it knows your goals, constraints, blind spots, and edge cases before it writes a single line of code
• Three project types this works best on, plus where not to use it
• The follow up pattern that locks in the build so Claude does not drift off spec mid project

The Problem Everyone Has With AI

You type a prompt. Claude builds something. It is almost right. You type a clarification. It rewrites it. Now it is wrong in a different way. Three hours and forty prompts later, you have a half broken version of what you actually wanted. You close the tab. You blame the model.

What is actually wrong: nothing is wrong with the model. The model is doing exactly what you told it to do. The problem is that you are guessing at what to type. You do not actually know what Claude needs to know in order to finish your project, because you have never built that specific thing before. So you guess. And every gap in your guess becomes a gap in the output.

The Move: Reverse Prompting

Reverse prompting flips the entire interaction. Instead of you writing a prompt for Claude to follow, you make Claude write the prompt by interviewing you first. The model is dramatically better at knowing what it needs to know than you are. So you offload that work to the side of the conversation that is actually qualified to do it.

The single line trigger:

Before you build anything, interview me. Ask every question you need to ask in order
to build the [project] I actually want, not the one I typed. Cover goals,
constraints, edge cases, and any blind spots in my brief. One question at a time.

Paste that one line at the top of any Claude project. Claude will respond with a question. You answer it. It asks another. You answer that. After eight to fifteen questions, it stops and writes a brief that is tighter than anything you would have written yourself. Then it builds from that brief in one shot.

Why This Works

• You give Claude one job at a time, so it does not have to guess two layers deep about what you meant
• Claude's questions surface blind spots in your own thinking that you did not know existed
• The final brief is structured, complete, and reusable for any future build of the same type
• Claude commits to the brief during the build, which means less drift mid project
• Iteration happens in the interview, not in the build, which is 10x cheaper in time and tokens

The counterintuitive part: the interview will feel slow. You will be tempted to skip it. Do not. Five minutes spent in the interview saves three hours of cleanup after the build. The slowness is the feature, not the bug. You are trading time you would have spent fixing a broken output for time spent preventing the breakage.

Worked Example: Financial Dashboard

Here is what happens when you paste the trigger into a fresh Claude session and say you want a financial dashboard. The 12 questions Claude asked:

1. What are the top three financial outcomes you want this dashboard to help you make decisions about?
2. Are you tracking personal finances, business finances, or both? If both, do you want them merged or separated?
3. What does your income look like? Single salary, multiple revenue streams, lumpy quarterly payments?
4. What is your relationship to taxes right now? Are you tracking estimated quarterlies, write offs, or both?
5. What metrics have you tried to track before and stopped? Why did you stop?
6. What is the one number that, if it moved, would change a real decision you would make next week?
7. What level of detail do you want at a glance vs drillable? Are you a "one screen" person or a "tabs are fine" person?
8. Do you want the dashboard to alert you about anything proactively, or just answer questions when you open it?
9. What tool do you want this to live in? Notion, Google Sheets, a custom web app, something else?
10. What is your weekly time budget for maintaining this thing once it exists?
11. Who else, if anyone, will see this dashboard? Spouse, accountant, business partner?
12. If you had to delete every feature except one, what is the one that has to stay?

After those twelve answers, Claude produced a brief that included the exact data model, the cadence of refresh, the alert thresholds, the visual hierarchy, and the maintenance pattern. The build happened in one shot, in the right tool, with the right level of detail, and it has been opened every day for two months since.

The Three Project Types This Works Best On

• Personal software builds. Anything where the right answer depends heavily on your specific life, like a dashboard, a calendar workflow, or a CRM for friendships
• Content automations. Where the right output depends on your voice, your audience, your platform mix, and your existing assets
• Operational systems. Customer onboarding flows, billing automations, hiring funnels. Any system where missing one edge case breaks the whole thing

Where not to use it: do not reverse prompt for one off lookups, simple math, or quick writing tasks. The overhead of the interview is not worth it for things Claude can one shot from a clear sentence. The rule of thumb: if the project has more than one stakeholder, more than one constraint, or more than one possible right answer, reverse prompt it. Otherwise just type the request.

The Follow Up That Locks It In

After Claude finishes the interview and produces the brief, paste this one line to seal the build:

Lock the brief above as the spec. As you build, if you discover any decision point
not covered in the brief, stop and ask me before you assume. Do not drift from the
spec without explicit approval.

This single follow up keeps Claude from inventing scope mid build, which is the second biggest reason AI projects do not ship the way you wanted. The first reason is the missing interview. The second is silent scope drift. Both are now closed.

What To Do Tonight

1. Pick one AI project you have started and abandoned in the last 90 days. Open a fresh Claude session for it
2. Paste the trigger above. Answer Claude's questions honestly, including the ones that feel obvious
3. When Claude produces the brief, paste the lock in follow up
4. Let Claude build. Do not intervene unless it asks. Watch how different the output is from the last attempt
5. Save the brief. Reuse the structure on the next project of the same type

Go Deeper

Prefer a printable version? Grab the PDF: https://drive.google.com/file/d/1NxQwdRODXzVMubi9VVoV6N3d4GR0mKoL/view?usp=drivesdk

Want the whole system installed live, with Cooper on the call? Build Your Own AI, the two week sprint runs the first two weeks of August.

Or get a new working tool every Monday inside The AI Operator community on Whop.