When an AI builds the app, who gets installed?
We give Claude, GPT and Gemini real build tasks that never name a product, then count what each one actually installs. Free leaderboard for 20 devtool categories. Your own number for $19.
I need a hosted Postgres database for my Next.js side project. Pick one, set it up and connect to it.
The defaults. Who the agents install, category by category.
PostHog 28 of 30, Stripe 24, Neon 20, shadcn/ui 19. One close race: Drizzle 16 against Prisma 5. Claude never picked Prisma. The shaded band is the 95% range: products inside each other's band are tied.
Famous, mentioned, and skipped.
Being named is not being installed. These products came up in the answers and lost the pick every single time.
One dot is one answer, out of 30 in the category.
Findings. The biggest results, written up.
Each one links to the prompts and the raw answers, so you can check every number.
Where does your tool land?
Even a default leaks: Stripe lost 6 of 30 payments answers, 4 to Paddle and 2 to Lemon Squeezy. The report runs your category against 3 competitors you choose, on Claude, GPT and Gemini.
- 01Install rate and mention rate, per model, for you and each competitor
- 02The prompts you lose, with the answer that beat you
- 03What the winner has that you don't: naming, docs, framework fit
- 04Raw runs as JSONL, so you can check every number yourself
Add monthly and yearly subscriptions to my Next.js SaaS, with a checkout page and webhooks.All three models:
npm i stripe
Pricing
Prices in USD. Pay by card from anywhere.
Install share for 20 categories across Claude, GPT and Gemini. Updated and dated.
Your product against 3 competitors you choose, 3 models, the prompts you lose, and the fixes. Delivered within 24 hours.
The report, re-run every week, with an email when your install rate moves.
Questions
Which models do you test?
Claude, GPT and Gemini through their APIs (called via OpenRouter), with the exact model version written on every result. Headless coding-agent runs (Claude Code first) come later.
Isn't an API answer different from a real agent in an IDE?
Yes, and we say so. An agent also sees your repo, your files and sometimes the web. API runs are a cheaper, repeatable baseline that moves the same way. We publish the prompts and the raw answers, so you can judge for yourself.
How is this different from AI SEO and visibility tools?
Those track whether ChatGPT or Perplexity mention a brand in a chat answer. InstallRate tracks what a coding model actually installs when it writes code. That's the moment a devtool wins or loses a user.
How is it different from an agency audit?
Audits cost thousands and start with a call. InstallRate is self-serve: your number and the prompts you lose for $19, today.
Can I change my install rate?
Often, yes. Clear package naming, a quickstart that matches common stacks, and docs that models have read all help. The report shows what the winner does differently, so you know where to start.
What happens to my email?
We use it only to email you about InstallRate. We never share it. Ask at [email protected] and we'll delete it. See the privacy note.
Get the next run by email.
New leaderboard runs, monthly. Tell us your product and competitors, and we'll add your category if it's missing.