Skip to main content
AI and automation

Shopify's New Flexible Collections (2026): What's Actually Native for AI Merchandising, and What Still Needs an App

Several blogs claim Shopify shipped native "AI Collection Sort" and "predictive cross-sell blocks" in 2026. Shopify's own documentation says something different. Verified against shopify.com, shopify.dev, and the Shopify Help Center as of September 4, 2026.

AIEC App Lab AI Editorial TeamPublished Updated Read 11 min#AI and writing#App comparisons

What you will learn

On this pageContentsCloseOpen

If you searched for "Shopify AI merchandising 2026," you've probably run into blog posts claiming Shopify shipped a native "AI Collection Sort" feature, "predictive cross-sell blocks," and a "merchandising insights panel" as part of its Summer 2026 Editions. Those exact phrases do not appear anywhere in Shopify's own changelog, help center, or Editions notes. Two of them are actually the marketing names of third-party Shopify App Store apps, not Shopify features.

What Shopify actually shipped in 2026 is real, but different: a rebuilt collections architecture, published July 16, 2026, that lets a single collection combine automated conditions, hand-picked products, other collections, and — for the first time — data supplied directly by apps. That last part matters more than it sounds. It's the plumbing that lets AI-driven merchandising apps feed trending, bestseller, or personalization signals straight into a native Shopify collection, without an app having to fully own or rebuild that collection.

This article checks what's actually confirmed in Shopify's own documentation, what the new architecture does and doesn't do, and which currently listed App Store apps fill the gap between "Shopify gives you the pipes" and "an algorithm decides what goes where." Research date: September 4, 2026. All pricing is in USD.

What Shopify actually shipped, in its own words

Two changes from 2026 are directly relevant, and both are documented on Shopify's own domains.

April 15, 2026 — Most Relevant sort order. Shopify's changelog added "Most Relevant" as a new collection sort option, based on sales performance, and made it the default for newly created collections. Shopify has not published the weighting or time window behind this sort, so treat it as a black box: useful as a default, not something you can explain to a client or audit.

July 16, 2026 — a new, source-based collections model. This is the substantial one. Shopify's own announcement describes it as replacing the old binary of "manual collection" or "automated (smart) collection" with a single model where a collection can combine multiple sources at once: automated conditions, manually picked products, exclusions, references to other collections, and — new — shareable sources published by apps. A parallel post on Shopify's partner blog, last updated July 1, 2026, spells out the developer side: the new model ships in GraphQL Admin API version 2026-07, and apps can now publish "shareable app-owned sources" that merchants add to a native collection alongside their own rules.

Concretely, this update adds:

  • Multi-source collections. One collection can pull from automated conditions, a manual product list, other collections, and an app-owned source at the same time, instead of forcing a single mechanism.
  • Variant-level targeting. Collections can include specific variants (for example, only red and pink variants for a Valentine's promotion) rather than entire products.
  • Collections as reusable building blocks. A collection can be built from other collections and stays in sync as the source collections change.
  • App-owned "sources." A reviews app, a subscriptions app, or a merchandising app can publish a source — for example, "trending this week" — that a merchant adds to a native collection. Shopify evaluates the combined sources as catalog data changes; the app doesn't have to keep rewriting the product list itself.
  • Sidekick-assisted collection building. Shopify's Help Center documents that Sidekick, the admin's AI assistant, can create and edit collections from natural-language instructions — for example, setting up a "New Arrivals" collection that tags new products and automatically removes the tag (and the product) after a set number of days.

Eligibility: according to Shopify's Help Center, collections — including the new model — require the Basic plan or higher, or the Agentic plan for merchants building an online store. Starter-plan stores don't get collections at all, native or otherwise. The Help Center also notes the new model is rolling out gradually: if it isn't yet available in a given store's admin, that store keeps using the legacy manual/smart collection split until it arrives.

What "AI merchandising" actually means in this update

Here's the honest read: nothing Shopify shipped in 2026 is itself an AI ranking or recommendation engine for collections. Shopify did not publish a native feature that autonomously detects trending products, predicts conversion probability, or generates cross-sell pairings on its own. The "Most Relevant" sort is sales-based, not disclosed as AI-driven, and Shopify's own wording never uses "AI" to describe it.

What Shopify built instead is an extensibility layer. The Collection Sources API exists specifically because, in Shopify's own words on its partner blog, "partners had product signals merchants wanted to use, but no clean way to contribute those signals to a native Shopify collection." The intelligence — the actual "which products are trending," "which products convert best for this segment," "what to rank first" logic — still has to come from somewhere. As of this research, that somewhere is an app, not Shopify's core platform.

So when a merchant asks "does Shopify now do AI merchandising natively," the accurate answer is: Shopify now gives apps a native place to plug AI-generated merchandising decisions into a collection your customers see — but Shopify itself isn't generating those decisions.

What the native tools still don't do

Four limits matter most when deciding whether the native architecture is enough on its own:

1. No native ranking algorithm you can inspect or tune. Most Relevant is a fixed, undocumented sort. The new sources model lets you combine typed conditions (tag, price, inventory, metafield, and similar attributes) with manual picks — but typed conditions are rule-based, not learned. There's no native setting for "rank by predicted conversion probability" or "rank by margin-adjusted demand." That logic has to arrive as an app-owned source.

2. No per-visitor personalization. A collection built with the new model still renders the same product order to every visitor. Variant targeting changes which variants appear; it doesn't change what a specific shopper sees based on their browsing or purchase history. Personalized collection ranking is still app territory.

3. No native A/B testing of collection composition or sort. Neither the collections announcement nor the Help Center documentation mentions experimentation tooling for collections. If you want to test whether an app-driven "trending" source outperforms Most Relevant, you need the app (or a separate testing tool) to measure it.

4. Sidekick is task automation, not continuous optimization. Sidekick can set up a collection, its tagging rules, and its automation from a prompt — but it executes what you asked for once and stops. It does not continuously re-rank a collection based on live conversion or margin signals; that requires a source that keeps updating on its own, which is exactly the app-owned source mechanism, not Sidekick.

There's also a practical migration limit: apps that haven't updated to GraphQL Admin API 2026-07 can't read or write collections that use any new-model feature — multiple sources, exclusions, sub-collection sources, or variant targeting. Older third-party apps interacting with a store's collections may silently stop seeing them until their developer migrates.

Where the app layer still earns its price

Given that Shopify supplies the pipes but not the intelligence, the apps worth evaluating are the ones that either generate the ranking signal itself or add capabilities (personalization, testing, checkout-level cross-sell) the native model doesn't touch. Three currently listed apps illustrate different parts of that gap.

AppDeveloperPricing (USD)What it adds beyond native
Entaice ‑ AI Collection SortEntaiceFree to install; free for 30 days, then $48/month under $250K in annual sales, $150/month up to $1M, custom above thatGenerates an actual ranking signal — sales, inventory, and trending data recalculated automatically — instead of Shopify's undisclosed Most Relevant sort
Bestsellers reSort ‑ OrganizerEGNITIONFree (up to 49 products/99 collections); $9.99–$39.99/month by catalog size (up to 4,999 products); higher tiers availableRule-based sorting by revenue, profit, views, and tags with per-collection rules and scheduled placement — useful where you want transparent, auditable rules rather than an opaque algorithm
Zoorix: Bundles & Cross SellZoorixFree plan available; paid tiers $7.99–$29.99+/month, billed against revenue generated through the appAI-assisted cross-sell on product pages and in the native cart drawer, plus bundle and volume-discount logic — none of which the collections update touches, since it's cart-level, not collection-level

A few things worth noting about this table. Entaice is explicitly branded and described as AI-based sorting: it recalculates rankings from sales, inventory, and product data on an ongoing basis, which is the "continuously learning" capability Sidekick and Most Relevant don't provide. Bestsellers reSort, by contrast, is rule-based automation — useful and currently listed, but its own listing doesn't claim AI-driven ranking, so don't mistake "automated" for "AI-driven" when comparing options. Zoorix operates at a different layer entirely: cart and product-page cross-sell, which sits outside collections altogether and is unaffected by the July 2026 update either way.

If you're already running a recommendation or upsell app — Rebuy, LimeSpot, Wiser, or similar — the July 2026 update doesn't replace what those apps do for on-page or cart recommendations. It only changes how collection pages can be assembled, and it gives those same vendors a cleaner way to feed a "trending" or "personalized" source into a native collection instead of rendering a separate widget.

Decision guidance

You have a small catalog and don't need explainable ranking logic. Most Relevant plus manual curation, using the new multi-source model, is enough. No app required.

You need auditable, rule-based sorting (by revenue, margin, or tags) without paying for an AI subscription. Bestsellers reSort or a comparable rule-based sorting app covers this; you're not paying for machine-learned ranking, just automation of rules you already understand.

You want a ranking signal that adapts on its own as sales and inventory shift, without you rewriting conditions. This is where an AI-branded sorting app like Entaice fits — evaluate it against your actual catalog size and revenue, since Entaice's pricing scales with annual sales.

You need cross-sell or bundle logic in the cart or on product pages. That's not something the collections update touches at all. Zoorix and similar cart-layer apps remain the relevant category, independent of anything native shipped in 2026.

You're a developer or you rely on third-party apps that manage collections. Confirm those apps have migrated to GraphQL Admin API 2026-07. If they haven't, they may not see collections built with multiple sources, exclusions, or variant targeting.

What not to do

Don't pay for an app whose entire pitch is "automatically sort collections by sales" if your real need is met by Most Relevant plus a manual pin or two — that specific capability is now free and native. Don't assume "AI merchandising" claims about Shopify itself without checking Shopify's own changelog or Help Center first; the specific feature names circulating in some 2026 recap articles ("AI Collection Sort," "predictive cross-sell blocks," "merchandising insights panel") could not be verified as native Shopify features as of this research, and at least one of those names belongs to a third-party app, not Shopify. And don't confuse "automated" with "AI-driven" when comparing collection apps — read the actual feature description, not just the category tag, since Shopify's own App Store categorizes rule-based and machine-learned sorting tools under the same headings.

Sources

Pricing, ratings, and feature sets change. Confirm current details on each app's App Store listing and Shopify's own documentation before making a decision.

RelatedChatGPT Never Got In-Chat Checkout for Shopify Merchants — Google AI Mode, Copilot, and Meta Quietly Did

OpenAI's September 2025 Instant Checkout launch promised Shopify's 1M+ merchants native in-chat purchases. By March 2026 that plan was scaled back for ChatGPT specifically. Shopify's current Help Center documentation confirms ChatGPT is discovery-only with a redirect to your own checkout, while Google AI Mode and Gemini, Microsoft Copilot, and Meta got real Shopify-powered direct checkout instead — and none of your GA4 or Meta pixels fire when a customer buys there. Researched from Shopify Help Center and official sources, September 4, 2026.