ai-marketing-skills/autoresearch

Autoresearch for Marketing

Karpathy-inspired iterative optimization loops applied to conversion content.

Inspired by Andrej Karpathy's autoresearch concept — autonomous research loops that generate, evaluate, and evolve solutions — this skill applies the same pattern to marketing copy. Instead of optimizing ML experiments, it optimizes landing pages, emails, ads, and forms.

How It Works

┌────────────┐    ┌─────────────┐    ┌──────────┐
│  Generate  │───►│  Score with  │───►│  Evolve  │
│ 10 variants│    │ 5-expert     │    │ top 3    │
└────────────┘    │ panel        │    └────┬─────┘
                  └─────────────┘         │
                        ▲                 │
                        └─────────────────┘
                          repeat until 85+

1. Generate 10+ variants of each content element (headline, CTA, body copy, etc.)
2. Score every variant with a 5-persona expert panel (CMO, skeptical founder, CRO expert, copywriter, your CEO)
3. Evolve the top performers — analyze what won, push those patterns further
4. Cross-breed winning elements into complete units
5. Output the best version + full experiment log

No traffic needed. No A/B test infrastructure. Minutes instead of weeks.

Quick Start

# 1. Install dependencies
pip install -r requirements.txt

# 2. Set your API key
export ANTHROPIC_API_KEY="your-api-key-here"

# 3. Run the optimizer
python3 autoresearch.py --input landing-page.html --type landing_page --min-score 85

# 4. Check results
cat data/landing-page-optimization-report.md

What It Optimizes

Content Type	Elements	Score Dimensions
Landing Pages	Headline, subheadline, CTA, problem section, social proof	First impression, clarity, trust, urgency, conversion
Email Sequences	Subject, opening, body, CTA, PS line	Open rate, read rate, click rate, reply rate, spam risk
Ad Copy	Headline, description, CTA	Scroll-stopping, clarity, click-worthiness, relevance, differentiation
Form Pages	Headline, subtext, bullets, button, field order	First impression, trust, completion likelihood, lead quality

The Expert Panel

Every variant is scored by 5 simulated experts:

1. CMO — "Would this make me stop and engage?"
2. Skeptical Founder — "Do I believe this?"
3. CRO Expert — "Is this clear and action-driving?"
4. Senior Copywriter — "Is this compelling and differentiated?"
5. Your CEO — Configurable. Define their voice in references/founder-voice.md

Output Files

Each run produces:

• {name}-optimized.{ext} — The winning content
• data/{name}-experiments.json — Full experiment log with all scores
• data/{name}-optimization-report.md — Human-readable summary

Configuration

Option	Default	Description
--min-score	80	Target score (stops when reached)
--rounds	3	Max optimization rounds
--variants	10	Variants per round
--elements	all	Specific elements to optimize
--type	auto	Content type override

Credit

Concept inspired by Andrej Karpathy's autoresearch — applying autonomous iterative optimization to marketing instead of ML research.

Requirements

• Python 3.10+
• Anthropic API key ($ANTHROPIC_API_KEY)
• See requirements.txt for Python dependencies

---

🧠 Want these built and managed for you? →

This is how we build agents at Single Brain for our clients.

Single Grain · our marketing agency

📬 Level up your marketing with 14,000+ marketers and founders → (free)