The Idea
Find out if an ad works before you pay to find out
Everyone knows the only way to learn if an ad converts is to spend money on it. You launch, you wait, you burn budget, and a few days later the numbers tell you what you could have guessed.
The Ad Predictor flips that. It builds a panel of synthetic buyers, researched to think and object like your real audience, and runs every ad past every one of them as an independent critic. Then it tells you which ad beats which, by how much, and exactly what to change on the weak ones.
It is not a crystal ball. It is a filter you run first, so the ads you actually spend on are the ones that already survived a room full of skeptics.
How It Works
Type one command. It does the rest.
You run /ad-predictor
Claude asks you six quick things: your product, the problem it solves, who you sell to, the price, the channel, and your ads. Paste the copy or hand it image files. It reads images too.
It researches your market
It pulls the real language buyers use, what competitors claim, and the gaps nobody is addressing. This is what stops the customer panel from being a pile of stereotypes.
It builds your customer panel (and shows it to you first)
A spread of buyers, deliberately skewed toward the skeptical, price-sensitive majority a real cold audience actually is. You get to edit the panel before anything runs, so the audience is right.
Every buyer reacts to every ad, alone
Each one reacts in its own private context, so one excited reaction can't infect the rest. Each gives a gut reaction, at least two reasons not to buy, the one change that would move them, and then ranks all your ads worst to best.
You get a simple, ranked verdict
A letter grade per ad, a directional conversion band, a one-line why, the top objections, and the exact copy fixes. Then it offers to rewrite your weakest ad using the feedback. The math stays under the hood. The report stays plain English.
Setup
Install it in 60 seconds
A skill is just a folder Claude Code reads. You drop it into a .claude/skills/ folder and it shows up. You need Claude Code for this. That is the only requirement to start.
Download and unzip
Inside you'll find a folder called ad-predictor (the one with SKILL.md in it) and a README.
Drop the folder into your skills
For one project, put it at your-project/.claude/skills/ad-predictor. To use it everywhere, put it at ~/.claude/skills/ad-predictor instead. Either way the path should end in .claude/skills/ad-predictor/SKILL.md.
Open Claude Code and run it
Type /ad-predictor, or just say "predict my ads." Answer the six questions and you're off. The README in the download has the same steps if you get stuck.
The Honest Part
It refuses to lie with a number
Most AI tools will happily tell you an ad "converts at 4.2%." That number is fake. AI personas are documented to be too agreeable, so any precise prediction is a lie dressed up as data.
So this tool will not do that. It tells you which ad beats which and why, because the relative ranking is the part you can actually trust. Every conversion figure ships as a directional band, stamped with one line:
"Directional filter, not a forecast. It narrows what to test and how to sharpen the copy. It does not replace a real A/B test with live spend."
Use it to decide what to test, not to skip testing. That honesty is what makes it safe to hand to someone else.
For Power Users
Want 100 customers and real statistics?
The default runs eight to twelve buyers right inside the conversation and needs nothing extra. Start there.
When you want the full statistical treatment, the download includes a Python engine that fans the same test across fifty to a hundred buyers, runs each one multiple times, groups ads that are too close to call into shared tiers, and reports how stable the ranking is across runs. It needs one package and an API key, both covered in the README inside ad-predictor/harness/. A full run usually costs a few dollars in API. You do not need it for normal use.
That's the whole thing. No course, no upsell, no redacted version. Download it, run it, and if it makes you money, come tell me.
Work with Me
Want me on your problem for an hour?
Book a 1:1 call and we build the fix live — the workflow you want automated, the tool you can’t crack, whatever’s stuck. Direct, hands-on, no pitch waiting at the end.