Trust & safety
The assistant speaks for your brand, so it’s built not to make things up. Four properties protect that.
Grounded only in your catalogue
Section titled “Grounded only in your catalogue”Every recommendation is grounded in your real catalogue, with live prices and stock. The assistant will not invent product names, SKUs, prices, or links. If it isn’t in your catalogue, it isn’t in the answer.
On-topic and on-brand
Section titled “On-topic and on-brand”It stays on the subject of shopping your store and answers in your brand’s voice. It politely declines off-topic prompts and resists attempts to be talked into a different persona or to ignore its instructions.
Moderated input
Section titled “Moderated input”Anything a shopper uploads or types is validated and moderated for safety before it’s processed. Unsafe or out-of-scope input is declined rather than acted on.
Per-storefront isolation
Section titled “Per-storefront isolation”Each storefront is fully isolated. Sessions, keys, and conversations never bleed across brands — one retailer’s traffic, catalogue, and configuration are never visible to another.
Data handling
Section titled “Data handling”Conversations are scoped to a session for the purpose of answering follow-ups in context. For specifics on data retention and processing for your contract and locale, speak to your THG Ingenuity representative — they can provide the data processing details relevant to your storefront.