Most marketing is a guess.
Autoresearch Playbook puts AI to work testing your cold emails, landing pages, and prices — it keeps the versions that win and drops the rest. Higher conversions, without the expensive guesswork.
The hidden cost of autopilot: run these loops unattended on a paid AI and the per-token bill is ~$15 a night — about $450/month run nightly. The same prompt, copy, and SEO loops run on your own computer for $0.
- No code
- if you can fill a form
- $0
- to run locally
- 12 + guide
- templates + playbook
Testing your marketing shouldn't take a team you don't have.
Write a subject line. Ship it. Hope. Or pay thousands a year for an enterprise testing suite built for big teams — and still do the work yourself.
- Guesswork dressed up as a "strategy"
- $30k/yr tools like VWO & Optimizely
- Hours of manual A/B busywork
Hand the AI one thing to test and one metric to chase. It writes versions, tests them against real conversions, keeps the winner and reverts the rest — while you sleep.
- A one-time $97 toolkit, not a subscription
- Runs for $0 on your own computer
- Scored on real outcomes, not raw clicks
Try a change. Keep it if it won.
It's the open-source idea the AI researcher Andrej Karpathy shared for free (88,000+ GitHub stars), packaged for marketing: try a change, see if it did better, keep it if it won, undo it if it lost — then try again.
Pick what to test
One thing at a time — a cold-email subject, a headline, a price. Set the single metric that means "better" (replies, opt-ins, sales).
Let the AI try versions
It generates variations, runs them against real outcomes, and scores each one — never against raw visitor counts, only results that matter.
Keep the winner
Winners are promoted, losers reverted automatically — a ratchet that only moves forward. Then it goes again, while you sleep.
Run it through an AI tool you already use — Claude Code or Cursor — or fully local on Ollama. No GPU, no PyTorch, no coding.
Twelve fill-in-the-blank templates.
Each one is a ready-to-run instruction set for a real marketing job. Copy it, fill in your numbers, hand it to your AI — and it runs the loop. Grouped across the funnel:
- Cold-email subjectsmore positive replies
- Ad headlines & hookslower cost per acquisition
- Lead magnetshigher email opt-in rate
- Titles & metamore organic clicks · runs $0 local
- Landing-page headlineshigher conversion rate
- Pricing pagesrevenue per visitor · 14-day guard
- Checkoutfewer abandoned carts
- Webinar signupshigher registration rate
- Onboarding emails7-day activation rate
- Email sequencescompletion & reply rate
- Trial → paidhigher conversion to paid
- Referralsmore referrals per customer
One subject line, run through the loop.
You fill in your goal, your baseline, and your metric. The AI does the rest — one variable, scored on real replies, winner kept.
Illustrative run — your numbers, your copy.
Lift the positive-reply rate on my cold-email subject line. One metric: replies. Keep it under 8 words.
The loop promotes the highest-reply subject and discards the losers — so the version you ship is the one that earned it, not the one you happened to like. A cost guard caps any API spend before it starts.
Buy it once. Run it forever.
One-time digital downloads on Gumroad — no seats, no subscription. Not sure yet? Start free.
Just the 3 core templates. Start here.
- 3 core templates — cold email, landing page, pricing
- Clean, ready-to-edit markdown
- Free lifetime file updates
- Setup guides not included
The toolkit, plus the guide that runs it.
- All 12 templates + the 28-page guide
- Cost-savings appendix (Ollama / MLX)
- Model-routing matrix + 7 failure modes
- 30-day execution roadmap
- 14-day money-back guarantee
Fix the prompts, then automate the testing.
- Everything in the Full Playbook
- 77-page AI Prompt Engineering Playbook
- 1h 5m audio guide + 5 ready-to-use tools
- The free assessment, included
Instant download on Gumroad · free lifetime updates · 14-day money-back guarantee
Before you buy.
What if I'm not technical at all?
That's exactly who this is for. Everything is plain English, walked through step by step. You fill in templates with your own details — if you can fill out a form, you have every skill you need.
Do I need a GPU or to write code?
No GPU, no PyTorch, no coding. (The GPU requirement is for Karpathy's original ML project — this applies the same loop to marketing.) You run it through an AI tool you already use, like Claude Code or Cursor, or fully local with Ollama.
What does it cost to run?
Free if you run it locally on your own computer with Ollama. If you'd rather use a paid AI for higher-confidence scoring, it's roughly $20 a month — and every template has a hard cost cap so you never get a surprise bill. The guide shows you both ways.
Isn't this just prompts I could find free?
No. Free prompts are one-off guesses. This is a system — fill-in-the-blank templates, a 28-page guide, and a testing loop that keeps the winners and reverts the losers automatically. It's a repeatable process, not a lucky paste.
How is it delivered, and what if it's not for me?
Instant download on Gumroad right after checkout — yours forever, with free updates. And it's backed by a 14-day money-back guarantee, no questions asked.
Is this affiliated with Andrej Karpathy?
No. "Autoresearch" is his open-source idea, shared freely. This product is independently created and is not endorsed by or affiliated with Andrej Karpathy.