The site uses ChatGPT as an example of a shopping agent that may visit a store. Maitre is not an OpenAI integration or partnership. It does not submit product feeds, set up checkout, or guarantee that ChatGPT visits, uses or recommends a store. Better-grounded recommendations are the intended benefit of answering an agent from the live catalog. They are not a measured search-ranking or AEO result.
All numbers on the site come from the Hermes trials described below, not from ChatGPT. The bag example comes from a synthetic trial shopper. It is not a real ChatGPT conversation.
October 27, 2026. Native Hermes shopping agents used GPT-6 Luna to shop a locally captured Philo Sophie storefront. Each session began at the homepage. The store used a compact merchant chat backed by GPT-6 Luna and its captured product catalog.
Two invitation wordings were tested against 50 matched synthetic shopper profiles each: 100 sessions. 58 reached the chat; 47 sent a nonempty shopping message. That is 47/100 from the homepage, or 47/58 (81%, rounded) among chat visitors.
Of 50 tasks explicitly authorizing collection updates, 28 supplied an authorized synthetic email. No real enrollment or purchase took place. Other context varied; unknown, missing and unsupported details were recorded separately.
The displayed brief comes from a synthetic trial. Demand grouping is a product design, not measured request volume. Results depend on agent, store, invitation and task. These trials measured replies, not sales or revenue lift.
Source: retained 20261027-philosophie-chat-clean study, compact-chat arms (23/50 and 24/50 replies). These statistics are separate from later GPT-6 Astra discovery studies.