How Spectare works

Every visitor gets a personalised experience built from the same source material. When someone arrives, Spectare classifies their intent from referrer URLs, UTM parameters, and session context, then passes a context token describing their audience, buying stage, and confidence level to the assembly step.

Assembly runs on the server. Claude receives the context token alongside three ranked inputs: a vector search over your atom library finding the most semantically relevant candidates, a conversion history showing which atoms and arrangements have converted for this audience and stage before, and an image selection picking the best visual for the moment. From these signals Claude selects which atoms to include and maps them to one of nine component types.

Two of those nine components are written entirely by Claude in your brand voice: HeroStatement and ImageCtaHero. The remaining seven components, including StatGrid, FeatureList, TestimonialCard, ComparisonTable, CodeBlock, AtomCard, and CtaStrip, are drawn directly from your atoms without modification. Your words reach the visitor unchanged; Claude only writes the headline and hero layer.

The assembled page is cached at the edge so subsequent visitors with the same intent pattern are served instantly without another AI call. Every impression is recorded against the variant, audience, and stage that produced it, feeding back into the conversion history that informs the next assembly decision.