How to turn your knowledge into a lead gen tool

Yesterday I gave a presentation to the Claude Marketers group about something I've been building for a while: AI tools made out of my own thinking. The ones that are on this site. I want to use them somehow to earn income directly or indirectly, but I'm going to tell you how to do it anyway.

This is that talk, written up, with the exact prompts so you can build your own.

The idea

You can put your own knowledge into an app that people use for free, and it becomes a version of you. Not an AI wrapper. Not a chatbot. Your actual judgment, packaged. People get a real taste of how you think, and you get their attention without ever getting on a call.

Here's what one of mine does, so this isn't abstract. Marketing goes live on hope. Everyone has an opinion about why something isn't converting. Nobody has a diagnosis. So I built a second opinion. You give it a marketing asset, an email, a landing page, a piece of direct mail, and it grades the thing the way a senior strategist would, then hands back a letter grade, the actual problem, and fixes you can copy and paste.

The kind of thing it catches: your ad promises one thing and your landing page opens on another. You blame the page. The page was fine. The handoff was the problem, and almost nobody checks the handoff.

That's mine. Yours can be about anything you're good at. It works for any consulting expertise, not just marketing. If people pay for what you know, a free taste of how you think is the most honest way to get them in the door. And you can use it however you need: a tool for your own work when you're fried, a lead magnet for your consulting, a way to let the right people see how you think before they ever reach out.

The recipe

Four steps. The order matters, and each one builds on the last.

1. Get your thinking out of your head and into a written brain. This is the part only you can do. How you actually attack a problem, your frameworks, your strong opinions, what you think everyone in your field gets wrong. This becomes the brain of the whole thing, and you can use it on its own for your own work forever.
2. Build it like a consultant UI. Someone is going to answer a few questions, add some context, maybe upload a screenshot or a PDF. Figure out what the diagnosis actually needs before you think about a single line of code. You're designing the conversation, not the software.
3. Write your scoring down so it isn't just vibes. I started on instinct. The written feedback was great and the grades swung all over the place. A rubric holds the grades steady without touching the feedback. Everyone skips this step. It's the whole difference between something that sounds smart and something that's actually consistent. Don't skip it.
4. Test it hard before you trust it. This takes longer than the build. Run the same thing through several times and check the grades hold. Run things you already know are strong and weak and check the tool agrees with you. When it sounds smart but gets it wrong, fix the source, not the answer. Spoiler alert - this part actually never ends. I use my tools every day, and I find things that I update on the DL. They are small, but I can be a perfectionist sometimes.

The four prompts

I reverse engineered what it took me weeks/months to figure out from experimentation and trial and error. Run these in order. Each one builds on the last. Where you see a bracket, drop in your own expertise.

Prompt 1: Extract your thinking and turn it into a skill

I want to build an AI tool that acts as a consultant in my area of expertise: [your area]. The end product of this prompt is a reusable skill (a written instruction set) that becomes the brain of the tool. I'll also use the skill on its own for my own work, so it needs to be clear and complete enough for any AI session to load and think the way I think.

From everything you know about me from our conversations, write out how I think about [area]. My frameworks, what I look for first, what I believe others in my field get wrong, my strong opinions, my non-negotiables, and how I make judgment calls when things are ambiguous.

Then tell me where your picture of my thinking has gaps. Ask me a series of questions, as many as you need, so I can fill those gaps until you fully understand how I approach this. I'll also correct anything you got wrong.

Once the gaps are filled, write the final version as the skill. Be thorough, not polite.

Prompt 2: Design the tool as a consultant, including the UI

Now I want to turn that skill into a tool other people can use. The tool acts as me: a consultant who evaluates [asset or situation type], diagnoses what's wrong or what's about to go wrong, and tells the person specifically what to fix.

The tool needs a UI. Depending on what the diagnosis requires, that means places for people to answer optional questions, fill in context about their situation, and upload things (screenshots, PDFs, links, whatever the evaluation needs). You tell me what inputs the tool needs to do its job well, and which should be required versus optional.

Act as my product consultant. Walk me through this build from start to finish. Ask me every question you'd need answered: who the user is, what they upload, what context to collect, what the output looks like, what a great answer sounds like versus a generic one, and what the tool should refuse to do. Get as detailed as we need to get before anything is built.

Prompt 3: Lock in the rubric

The tool needs a scoring system that's written in stone, not vibes. If the evaluation runs on judgment alone, the written feedback is great but the grades swing all over the place. The rubric stabilizes the grades. It doesn't change the feedback.

Based on my skill, build me a scoring rubric. Define the dimensions the tool evaluates, why each one matters, how many points each is worth, and exactly what earns or loses points within each dimension. Every point gets a written rule behind it, so the same asset gets the same grade every time.

Prompt 4: Test it before you trust it

The tool is built. Now help me test it like it's going in front of real people.

Build me a test plan: run the same asset through several times and check that grades hold steady. Run assets I already know are strong and weak and check that the tool agrees with my judgment. Try edge cases: incomplete uploads, assets outside the tool's lane, and inputs designed to confuse it.

For every output that sounds smart but is wrong, help me trace it back: is it a gap in the skill, a loose rule in the rubric, or a missing input in the UI? Fix the source, not the symptom, and retest until it holds.

Then build it

Claude helped guide me on the best (and most cost effective) tools for my specific needs. For the record, mine runs on Next.js and Vercel for the site, Clerk for login, Airtable for the database, Stripe for payments, Loops for email, PostHog and Clarity for analytics, and the Anthropic API powers the diagnostics. Netlify is good too if you don't want to stand up a whole website.

You don't need to know how to code. You need to know what you want the tool to do. Open Claude Code, give it your skill and your rubric, tell it what you want, and let it walk you through the rest.

Why I'm handing this over

What I built here is me, no one else can build it. I learned by just figuring it out, trial and error. I can save you time to build whatever will help you with your goals. I've gained value from what I've built in terms including emails, job leads, and even some paid customers.

If you'd rather not build it yourself, I do this as consulting work here: BuildDx — Julie Irving. But that's not the point of this. The recipe is right here, and you can build your own version this week.

Build the thing. Then tell me what you put inside it.