Your AI Shopping Assistant knows the shopper’s current page.

A shopper on a product page asks “does this come in blue?” The assistant used to need the question restated with the product name attached. Now it knows what “this” is.

The exact product, the exact variant, the exact SKU the shopper is browsing all flow into the assistant’s context. Ambiguous questions resolve against the item on screen. Fewer dead ends, fewer restarts, more confident buyers at the decision moment.

The assistant answers about the page the shopper is on

Product, variant and SKU context flow into every response. “Will this fit?” gets answered for the exact item on screen, not the product family.

Visible confirmation for fitment-sensitive categories

For auto, appliance and industrial catalogs, a fitment banner sits inside the chat. Shoppers see the assistant is answering for their setup. Trust surfaces at the moment of decision.

Tune how the assistant finds and answers product questions

A new panel exposes per-store search behavior. A vitamins retailer and an auto-parts retailer no longer share a search model. Match the answer profile to how your buyers shop.

Cleaner editing for custom chat flows

The visual builder for automated responses, forms and live agent handoffs got a UX and reliability pass. The canvas is smoother, saves are more consistent, multi-turn journey design stops fighting you.

Test the full chat experience before it reaches a shopper

Test flows end to end in the Playground before activation. Same environment as production, no production risk.

Sharper product Q&A

Ambiguity handling improved across the product Q&A path. Responses are tailored to the shopper’s context rather than the catalog’s average case. Combined with contextual grounding, the assistant is meaningfully less likely to answer the wrong question.

Per-merchant prompt layering

Merchant-specific instruction blocks now layer onto every agent (support, orders, returns, inventory, search, Q&A, compare, RAG, welcome, router). Brand voice, domain terminology, business rules and promotions can be tuned without engineering time. The customization sits at the end of the system prompt so the cached portion stays cached. No latency or cost penalty.

AI-driven referrals generate 254% more revenue per visit than standard organic sessions, per Adobe for Business (January 2026). That revenue doesn’t show up as a loss. It just never shows up. The problem isn’t the catalog. It’s the delivery layer: non-browser rendering, crawler load, structured markup and whether your data is accessible to the models pulling it.