Abstract
Prism is a real-time visitor intelligence layer that detects referral source at the edge, classifies it into a channel and topic, and passes that context downstream to adapt CTAs and the AI concierge system prompt. One visitor signal splits into multiple adaptive behaviors — hence the name.
Problem
Most sites serve the same hero copy and call-to-action text to every visitor regardless of how they arrived. A visitor referred from an AI answer engine who just read about a specific topic is in a different mental state than someone who clicked a paid ad or arrived through organic search.
The referral header is often stripped before it reaches server components, and client-side `document.referrer` is an unreliable fallback. A reliable solution needs to capture the signal at the network edge, before any header stripping occurs.
Approach
Edge detection
A Cloudflare Worker runs on the site's zone and inspects the `Referer` header on every inbound request. Detected AI sources (Perplexity, Gemini, ChatGPT, Claude, Copilot, and others) set a short-lived ai_ref cookie with a 1-hour TTL.
A client-side component provides a `document.referrer` fallback for sources where the header survives.
Context resolution
A broader _apex_src cookie (30-day first-touch) tracks all referral channels — organic, paid, social, email, and referral. A resolution function checks both cookies in priority order and returns a typed context object carrying tier, medium, source, and topic.
Tiered adaptation
Full-tier sessions (within 1 hour of arrival) receive channel-specific CTA copy and an enriched concierge system prompt. Light-tier sessions (1 to 7 days) receive only a label swap.
Direct or stale sessions receive no adaptation. All adaptations fire a GA4 event for downstream analytics.
Status
- Live on isimplifyme.com with both AI and broad-channel detection (v2).
- Other network sites carry the v1 AI-only variant; v2 rollout follows production validation.
- Concierge context injection is active on isimplifyme.com.