Milly Chat in June 2026: Ask Claude About Your Store, Chat Inside the Page, and Smarter Answers

June was a build-heavy month, and most of it points at the same thing: the assistant gives better answers, lives in more places on your store, and asks less of you to get there. The bigger items — a one-click Claude connector that answers questions about your store, an in-page widget that puts product discovery right inside the page, tunable search relevance, a rebuilt mobile experience, a broad anti-hallucination pass, per-segment B2B routing, and quizzes that now tie back to revenue. Billing got more honest too, with real projected overage numbers instead of a surprise at month's end.

For full details on any one feature, see its dedicated post on /writing.

Ask Claude about your store

You can now connect your store to Claude in one click and ask it about your chat performance — top customer questions, where your knowledge base has gaps, which products are pulling their weight, what the chat is actually returning on. Claude reads it directly and answers in plain language.

• One-click connect. Authorize once, pick your store, done. No keys to paste, no config file.
• Read-only by design. It can read your analytics, questions, knowledge gaps, product performance, and an honest ROI summary — it can't change anything about your store. The one thing it can write is a bug report, straight from the conversation.
• Ask in your own words. "What are people asking that we don't have a good answer for?" "Is the chat actually driving sales?" A real answer grounded in your data, not a dashboard you have to go read.

Available on Enterprise. There's a deeper write-up in its own post: Ask Claude about your store's chat performance.

Chat inside the page

The assistant no longer has to live in a floating corner bubble. You can now drop it inside the page — an embedded grid or list that turns any section into a product-discovery surface: a collection page, a landing section, a product page, wherever discovery belongs. On Shopify themes it's a native theme-editor block ("Inline Chat Widget") — add it like any other section, no code. The floating formats (bubble, slide-out, search bar, banner) are all still there; inline is just a new place the same assistant can live. Full write-up: the inline widget post.

Smarter, safer answers

The single biggest quality story of the month is one you'd never see in a settings page: the assistant got noticeably harder to push off your real catalog.

• It sticks to what's actually true. No invented prices, specs, stock, links, or store locations — if it isn't in your catalog or knowledge base, it won't make it up. It applies the right facts to the right product, searches before it says it can't help, surfaces an out-of-stock variant instead of dead-ending, and when something genuinely isn't there it says so politely instead of erroring out.
• Date-aware instructions. A time-boxed instruction ("say the Alpha restocks June 1, but only until May 15") is now honored to the day — the assistant knows today's date in Pacific Time, so stale instructions stop applying on their own.
• Honest handoffs. When a shopper asks to be contacted, the assistant only submits an email the shopper actually gave it — and if a handoff doesn't go through, it says so and hands over your real contact instead of pretending it worked. Support, warranty, and policy answers stay grounded in your knowledge base and your sanctioned channels.
• Cleaner formatting and SKU answers. Mid-sentence links render inline instead of as full-width buttons, the assistant can cite SKUs on request, and the widget's styling is shielded from your theme's CSS so it looks right on any store.

Tunable relevance

Search and recommendations are now steerable. A new editor lets you tell the assistant which product attributes should drive a match — device fit, material, whatever matters in your catalog — instead of a fixed, invisible boost. It's validated against your real catalog as you set it, so "why did it recommend that?" becomes something you can actually answer and adjust.

B2B-aware chat

If you sell wholesale and retail from the same store, the assistant now tells them apart and answers accordingly — using the customer tags you already have, so there's nothing new to set up.

• It branches on who's logged in. A recognized wholesale buyer gets B2B answers — pricing, minimums, policies — while a regular shopper keeps the standard experience. Cached responses stay segment-safe, so a retail shopper never gets a wholesale answer by accident.
• Per-segment product routing. A rule can point different segments at different Shopify collections automatically — wholesale shoppers to your wholesale collection, retail to retail, off the tags you already use. You describe the outcome; the assistant scopes its search to match.

Quizzes that tie back to revenue

The guided shopping quiz became a full loop this month: open the widget straight into a quiz (store-wide or per-page), read every quiz transcript in Replay, feed the answers into the same-session chat — and now, see what it earned.

• A new card on the Impact dashboard reports the orders, revenue, and conversion rate from shoppers who completed a quiz before buying. Open → transcript → chat → funnel → revenue, end to end.
• The card stays hidden until a store has quiz-driven orders, so it shows up exactly when it has something to say.

A rebuilt mobile experience

Most of your shoppers are on a phone, so we rebuilt how the widget opens there. Every format now slides up into a proper mobile bottom-sheet — swipe it down to dismiss, and it stays out of the way of the on-screen keyboard instead of fighting it. No more scroll jank, no more input hidden behind the keyboard. It's the same assistant; it just finally feels native on mobile.

Headless storefronts

If you're on a headless or single-page storefront — React, Hydrogen, WordPress — the widget is a first-class citizen there. Mount and refresh controls that follow client-side navigation, a product grid that adapts to the container it's dropped into, and session tracking that stays unified across frontends.

Transparent billing

Billing got more honest. The dashboard shows your real projected overage charges instead of leaving you to guess, conversation overages meter automatically onto your normal Shopify bill, and plan limits read as a clear "X of Y used" so you always know where you stand.

Smaller things

The assistant can now recommend your content, not just your products — point it at your blog or press feed and relevant articles surface alongside product picks (in-page rendering for headless storefronts is still on the way). You can deep-link the widget to open straight into a mode — a quiz, say — from any link or button. Analytics day-boundaries now bucket in your store's timezone, so daily numbers read cleanly. The product-catalog sync got safer and more accurate — metafield enrichment that survives the hourly sync, higher custom-field limits, and a guard so a truncated feed can't wipe your catalog. And we ran a broad security and trust pass across the platform this month — nothing you have to do, it's just sturdier underneath.

What's next

• More for the Claude connector. This month is the first cut — read-only insights and bug filing. There's more we want an agent to be able to do with your store.
• Inline article rendering for headless. Article recommendations ingest today; rendering them in-page on headless storefronts is the next step.
• Sharper answers, still. The anti-hallucination work isn't a one-month project — expect the assistant to keep getting harder to push off your real catalog.
• Live-agent handoff. Routing a conversation to a human when the assistant hits the edge of what it should answer — a merchant asked for this directly, and it's on deck.

The pattern under most of June: fewer decisions, smarter defaults, and an assistant you can talk to about your store instead of a dashboard you have to learn — now living wherever discovery happens, not just in a corner. We've started writing down why. More on that soon.

If something on your store would benefit from a feature we haven't built yet, say so — drop a note via the contact page, or reply to whatever thread you're already on with us. The request list is where most of the next month's roadmap comes from.