Main update: Shopify Winter Editions 2026 highlights new AI-driven tools, customization features, and operations upgrades, with several betas/early access sign-ups and a video overview. The Editions page includes interactive demos.
Sign-ups/early access:
Agentic Storefronts (list products in ChatGPT, Copilot, Perplexity), Online Store A/B testing (run experiments on variants), Tinker (AI playground), POS Hub (wired pointāofāsale device connector), Payment requests per fulfillment (Plus only).
AI & Sidekick:
Sidekick can create toādo lists, business recommendations, theme blocks across all themes, theme/app customizations, segments and Flow workflows; voice chat on mobile.
Sales channels:
Edit products/resources within the theme editor; generate themes in mobile; sell on WordPress; automatic discounts for specific customers; public AI theme builder (no account); Shopify Product Network to surface other brandsā products.
Checkout & payments:
Market-specific checkout/account customizations (early access for Advanced/Plus); Shop Pay Installments in the UK.
Operations & analytics:
Heatmaps, bot filtering, inventory adjustment history beyond 180 days, more flexible transfers, quick sale links, store credit email notifications.
Shop app & B2B:
Customizable Shop product pages; ACH bank payments and store credit for B2B; ERP integrations (NetSuite, BrightPearl, Fulfil, Sage, Acumatica).
Notes:
A/B testing app āsimgymā charges per simulation run. Discussion is informational; features rolling out, many require sign-up or specific plans.
Summarized with AI on December 10.
AI used: gpt-5.
Note: on the editions page some animations , background images are actually interactive for either on page clips or send you to take action in the admin. The UI is weird in places a bit , so best in fullscreen on desktop as a lot of the background chatbubbles , floating text, or animations are actually interactive to give mini demos that sometimes might appear further down the page /shrug.
TODO - Thereās a bunch of future features you can sign up for on the editions page to get notified about Sign up on the editions pages to be notified for:
āShopify Agentic Storefrontsā get products into ChatGPT, Copilot, and Perplexity, etc.
Online store A/B testing[1]
Tinker an ai playground for entrepreneurs
POS hub is on preorder, itās a wired connect for devices for more reliability than wireless
Payment requests per fulfillment. Send a separate payment request for each shipment of a multi-shipment order. Exclusive to Plus.
Did i miss any new betas?
[1] Note for a/b testing changes in the here and now see the new shopify made app thatās now available to test themes with simulated āAIā shoppers (pricing seems to be it charges per simulation run) : https://apps.shopify.com/simgym
Category Highlights
Sidekick is now able to generate todo lists, business recommendations, generate blocks in all theme store themes not just horizion themes, customize themes, generate apps, build segments or shopify-flow workflows. On mobile do with voice chat.
Online sales channel highlights
edit resources(products etc) right inside the theme editor
Super disappointed to see how heavily Sidekick is featured. I was excited, so went in to use sidekick to do the EXACT things you have advertised it can do, only for it to give me a 45 minute run around of lies and promises of things it canāt do. Itās like chatgpt now. Sidekick was better when it was in beta, not sure why it is so poor quality now.
It canāt run even the most basic inventory reportsā¦.but if you ask it, it will say it can! It will generate an empty report, then say āoh sorry, I canāt actually do thatā. AI has a lot of improvement to be done to it before it is actually useful.
Reminder Iām a shopify-partner, not shopify staff, so Iām not advertising what sidekick can hypothetically can do. For this thread only presenting some of the new āAIā based features.
Most modern āAIā is heavily overtuned to be nice helpful to give people good feelings; sycophantic AI.
So the tools can be kind but not nice vs nice but not kind ; e.g. glossy sentimental platitudes vs aid communicated effectively.
There is still a lot of utility in AI LLMs, it just needs to be treated as starting point or prototyping tool where your rolling the dice with your time for some possible output.
Treat it like an untrained intern that canāt be left to their own devices.
Even if you told the tool how to do it once and it did it once thereās a higher than non-zero chance itās gonna get it wrong yet the toolās still gonna try and give you a cheery disposition so you donāt fire it.
The are zero trust systems still needing actual human review, intervention or expertise.
āAIā Inconsistency
Since the tuning/training is ongoing what worked even just a few weeks ago might not work again without figuring out how to āprompt engineerā the problem statement for the current model. there is not a versioning or feature-freeze for sidekick afaik
Which is frustrating , because come on if it worked before then itās unexpected when it just stops working.
And thatās part of the trick, modern āAIā is based on unexpected behavior.
So that is a reality of using and living with technology thatās basically based on probability distributions in place of deterministic code.
Merchants that regardless of randomness find the ways in which such tools can be semi-consistently effective , without relying on it for deterministic results, those businesses may get further than others.
By not relying on previous experience with random outputs that happened to be useful at that time to set a baseline for all future behavior.
And those types will get waaay further than most who are fully buying into any type of hype about an āAIā creating an expectation it will always be consistent for anything and everything.
Test a thing, and have redundancy; and or use the extra time given by successful outputs to build that redundancy.
Modern āAIā just doesnāt work in a way to base expectations on.
Meaning previous results can literally be a fluke, some models can literally behave differently based on time of day.
Might be a good test to run having the āAIā generate a custom reporting app to try and lessen the effects of randomness by moving to code in a line of business app.
But the codes probability based too, so results are gonna vary ALOT as intent is still different from implementation.
Iām glad I found this summary, because although the official page is very pretty, thereās so much going on when I scroll something neurological is happening, and my brain just canāt make head or tail of it, just flatly refuses to take it in and causes some slight vertigo.
This update makes it clear that Shopify is moving from āstorefrontsā to āagent-ready commerce.ā
Agentic Storefronts is especially interesting ā listing products inside ChatGPT, Copilot, and Perplexity will only work if product data is machine-readable and trusted, not just SEO-optimized.
Weāve been analyzing this at Sixthshop while testing AI shopping discovery. One pattern weāre seeing is that products fail to surface in AI assistants even when they rank well on Google, usually due to missing schema, inconsistent price/inventory signals, or unclear variant modeling.
The Winter Editions features feel like Shopify is building the rails, while sellers still need better visibility into whether their products are actually eligible for AI-driven recommendations.
Curious how others here are preparing for AI agents selecting and purchasing products directly.
Glad it helped , didnāt want to do a full write up too many things and no where near enough time to invest on it.
Kinda figured that would happen for some , though theyād have logic for reduced motion but their idea of āreducedā is funny it seems. Try reading mode (nope doesnāt work in chrome browser ).
Try ctrl+p to in some browser to print the page to PDF for site designs that insist on themselves.
So I was gonna write up how to turn on device/os/browser settings that flag prefers-reduced-motion but I realized I ALREADY have that on!! and Iād have thought they had sane animation settings but I canāt even spot an on page button, nor a change in motion amount when I toggle my system settings though looks like thereās some code for it /shrug,
So itās a really cool golly gee design , might even win awards , but it is very much the type of landing page thatās disconnected from the everyman thatās full-tilt-because-all-the-designers-and-devs-have-company-issued-powerful-macbook-pros so it works good for them
Using prefers-reduced-motion on websites (click to open)
In the future on sale pages that insist on themselves like they are a superbowl commercial try prefers-reduced-motion accessibility settings, though itās not a guarantee the sites designers|accessibility-team were sane
Eeither at the OS level , somes browser, browser extensions, or using devtools and setting the flag. (test your current setting here) https://developer.mozilla.org/en-US/docs/Web/CSS/Reference/At-rules/@media/prefers-reduced-motion#user_preferences
(e.g. windows 10: Settings > Ease of Access (aka Accessibility ) > Simplify and Personalize Windows > Show animations in Windows in some modern browsers(e.g. chrome://settings/accessibility).
Or try reading mode, or ctrl+p to print the thing to PDF Sohe .
They way LLMās work at least for conversational surfacing(probabilistic / from raw text ingest) they donāt need structured data to make things up and luck out on pointing to something useful.
Itās a strength vs the wall regular SEO has backed itself into, but LLMās are still inconsistent if not outright random.
Or at least it was a differentiator as LLM providers try to get deterministic results out of the noise (read: pay wallable services).
I still think itās gonna be more about having structured data for the integration|feed|MCP etc to be able to present the merchandising in specific UI contexts when the LLM/agent uses tools(calls functions).
The info thatās in the conversational part can be plenty messy, otherwise people would have to talk in pedantic pseudo-code to get any effective results at all.
e.g. giving someone messy directions to a place can be work as long thereās road signs and addresses when the time comes.
Which in a way, with pseudo-code, is kinda how regular search engines have been deteriorating, with conversational queries your more and more likely to get noise thatās āoptimizedā to get in front of searchers but not at all relevant because your not pedantic enough on some things to filter out the machine-readable from the human made.
Itās just not clear if LLM optimization, GEO, is gonna be a mix of early wild west SEO with keyword stuffing in modern SEO machine-readable feeds thatās more about machine rigor to game algorithms than human appeal.
Or GEO has to be full optimization for machines, with content made for humans becoming a literal detriment.
Blocking entire countries, and data center IP addresses has become pretty common to fight bots.
So an irony is such normalcy may start costing sales, if not already, by blanket blocking agents as well because thereās no real way to tell them apart from bad actors.
Loudspeaker Winter Editions 2026 brings some solid upgrades for merchants especially AI-powered customizations, smoother app integrations, and improved automation. These updates make it easier to optimize stores, improve performance, and scale without heavy development. Excited to see how merchants use these features in real stores
Sign up on the editions pages to be notified for:
Customer sign-in with social accounts
Heatmaps in analytics
).
, but it is very much the type of landing page thatās disconnected from the everyman thatās full-tilt-because-all-the-designers-and-devs-have-company-issued-powerful-macbook-pros so it works good for them 
