PRD: ChatGPT Traffic Optimization — Capture +3,364% Referral Wave

[jonit-dev]

Summary

ChatGPT referral traffic is exploding (+3,364%) but we have zero ChatGPT-specific attribution, no referral-optimized landing experience, and no way to measure conversion quality from AI search traffic.

PRD Location: docs/PRDs/chatgpt-traffic-optimization.md
Priority: 🟠 High (#2)
Complexity: 6 → MEDIUM

Phases

Phase 1: Referral Detection & Attribution

• Classify ChatGPT/Perplexity/Claude referrals in middleware via referrer header + UTM
• Add referral_source as first-class Amplitude user property
• Files: middleware.ts, analyticsClient.ts, types.ts, tests

Phase 2: Personalized Hero for ChatGPT Referrals

• Subtle "Recommended by AI" badge for ChatGPT visitors (server-rendered, zero CLS)
• Files: HeroSection.tsx, ChatGPTBadge.tsx, locale strings, tests

Phase 3: Optimize llms.txt for Recommendation Likelihood

• Rewrite to lead with problems solved, add competitive positioning, add utm_source=chatgpt
• Add X-Robots-Tag: noindex to prevent Google indexing llms.txt
• Files: llms.txt/route.ts, llms-full.txt/route.ts, tests

Phase 4: AI Search Structured Data & Measurement

• Add FAQPage + HowTo schema to homepage for AI extraction
• Files: schema-generator.ts, page.tsx, locale strings, tests

Key Design Decisions

• Server-side detection via referrer header (not client JS) — zero CLS, works with SSR
• Referrer-based, not just UTM — ChatGPT doesn't add UTM params
• No separate landing page — same page, conditionally enhanced hero
• Multi-engine from day 1 — tracks Perplexity, Claude, Google SGE alongside ChatGPT

Success Metrics

Metric	Target
ChatGPT referral identification	100% tracked
ChatGPT visitor → upload rate	70%+
ChatGPT visitor → paid conversion	2x organic rate

Non-Code Action: Official Listing

• Create custom GPT in GPT Store wrapping upscaling API
• Monitor ChatGPT recommendations regularly
• Optimize llms.txt continuously

---

PRD: docs/PRDs/chatgpt-traffic-optimization.md
Target: myimageupscaler.com

[jonit-dev]

Execution timed out after 7200s (via GLM-5). Moved back to Ready for retry.

[jonit-dev]

Timeout follow-up:

Execution hit the 7200s runtime limit while processing 87-prd-chatgpt-traffic-optimization-capture-3-364-referral-wave.
Progress was checkpointed on branch night-watch/87-prd-chatgpt-traffic-optimization-capture-3-364-referral-wave, so the next run will resume from the latest checkpoint.

Suggested slices for the next runs:

1. Phase 1: Referral Detection & Attribution
2. Phase 2: Personalized Hero for ChatGPT Referrals
3. Phase 3: Optimize llms.txt for Recommendation Likelihood

Recommendation: avoid huge PRDs. Slice large work into smaller PRDs/phases so each run can finish within the runtime window.

[jonit-dev]

Execution completed but no PR was found (via GLM-5). Moved back to Ready for retry.

[jonit-dev]

Execution completed but no PR was found (via GLM-5). Moved back to Ready for retry.

[jonit-dev]

Execution completed but no PR was found (via GLM-5). Moved back to Ready for retry.

[jonit-dev]

PR opened: #89 (via claude claude-sonnet-4-6 (fallback))