How a clean skincare brand restructured its ingredient story to win the high-intent "niacinamide + bakuchiol" queries on ChatGPT — and ran 4.2× its matched control over ninety days.
Modern skincare shoppers don't trust hero shots. They ask AI to reconcile claims, mechanisms, and tolerability — the kind of question a polished campaign image can't answer. AI assistants only cite a brand when its PDP exposes ingredient claim, biological mechanism, and study reference in a structure the model can parse. Most luxury beauty PDPs bury that information inside design — gorgeous to humans, invisible to language models.
Vermille makes 84 clean-skincare SKUs — serums, moisturizers, masks built around active-ingredient pairings. Every PDP was beautifully written for human shoppers: hero imagery, an evocative claim, a small ingredient block at the bottom. The problem was that the ingredient story — claim, mechanism, study — lived inside design rather than structured data. ChatGPT couldn't quote it, so when buyers asked "what serum pairs niacinamide with bakuchiol without irritation," Vermille wasn't in the answer.
Agentic Page restructured every Vermille PDP around a three-part pattern: ingredient claim → biological mechanism → study citation. The mirror surfaces the "why" before the "what" — exactly the structure AI assistants quote when they cite a recommendation. Within four days, batch ingestion was detected from Claude and GPT knowledge-base systems. By day twelve, the first cited recommendation landed: a user asked about niacinamide + bakuchiol, and ChatGPT cited Vermille's Lumina serum by name.
Over the following two months, cited recommendations expanded into adjacent query clusters — vitamin C for sensitive skin, ceramide moisturizers with study backing. By day ninety, AI-channel traffic ran 4.2× above a matched control group, and two Vermille hero serums held cited-recommendation status across the highest-intent long-tail queries in the category.
Beauty buyers research slowly — they trial, compare, return to the same questions a few weeks apart. The Vermille deployment shows the longer arc that ingredient-led categories tend to follow: slow first weeks, sharp acceleration once the model associates the brand with a query family.
Agentic Page deployed across all 84 Vermille SKUs. Every PDP restructured around the ingredient claim → mechanism → study citation pattern.
First batch ingestions detected from Claude and GPT knowledge-base systems. The new structure is parseable; the ingredient logic now lives in machine-readable form.
First cited recommendation in ChatGPT. A user asks about niacinamide + bakuchiol pairings; ChatGPT names Vermille's Lumina serum and quotes the mechanism paragraph from the mirror.
Cited recommendations expand to vitamin C and ceramide query clusters. AI-channel traffic running 2.8× the matched control.
AI-channel traffic stabilizes at 4.2× control. Two hero serums hold cited-recommendation status across the highest-intent long-tail queries. AI Visibility Score reaches 84.6 / 100.