Atlas Experience Builder (Beta)

Describe the experience you want, and Atlas builds it, from the modification to the test groups and audience targeting.

The Atlas Experience Builder takes a prompt and builds a complete content test or personalization for you. It supports all content modification types today and passes context back and forth with the Atlas Visual Builder, so you can keep editing by hand at any point.

• Builds tests or personalizations end-to-end

• Works with the Atlas Visual Builder

• Currently in beta, with support for more experience types coming next

Reply Hi Hannah to request beta access →

Onsite Upsells (Beta)

A new experience type that puts testable upsells on your product and cart pages.

Onsite Upsells are a new experience type under Merchandising. To build one, describe what you want to the AI Visual Builder. It reads your site and creates the upsell for you, so there is no code, no apps, and no theme editing involved.

• Place upsells on product pages and in the cart, with discount support

• Recommendations skip the product the shopper is already viewing

• Dedicated onsite upsell analytics, with filters and key metrics

Reply Hi Brandi to request beta access →

Buy X Get Y Offers

Run "buy 2, get 1 free" style promotions and test them against your other offer types.

Buy X Get Y is a new discount type. Shoppers can mix and match across the products you make eligible, and the reward can be free, a percentage off, or a set amount off. You can A/B test it against your other promotions to see which drives more revenue.

• Rewards can be free, a percentage off, or a set amount off

• Choose whether the offer applies to one-time purchases, subscriptions, or both

• The buy and get items come from the same product selection today. Buy one product, get a different one is coming.

Build a BXGY experience→

Create Price Tests from Claude

Price test creation is generally available, and Intelligems is now listed in the Claude Directory.

The full price test lifecycle can now be managed through Claude. Anyone using Claude can search "Intelligems," connect in one click, and start creating and analyzing tests.

• Create and seed price tests through the MCP, now available to all customers

• Create and update custom events through the MCP and external API

• Set a custom metric as a key metric on new or existing experiments

• Test history is tagged with its source, so you can see whether a change came from the app, the API, or the MCP

Connect Claude →

Other Updates

• Make content edits exclusively with the AI Visual Builder. The legacy onsite editor can no longer create or edit content modifications. Preview and price tagging still work there.

• Add recurring countdown timers to your checkout blocks. Timers can reset on an hourly, daily, or weekly schedule, following your store's timezone.

• Add timers and expiration to your post-purchase upsells. Add a countdown to the offer, plus a 1- to 60-minute timeout that redirects the shopper to the thank-you page.

• Launch mutually exclusive tests. Now available to all accounts. No support request needed to turn them on.

• Intelligems tracks on any page where it is installed. That includes landing pages not hosted by Shopify, with identity carried across subdomains of your primary domain.

• View new upsell analytics. A Checkout Upsell analytics page for tests with upsell blocks, plus a dedicated Onsite Upsell page in beta.

• Better price test CSVs. Files include SKUs, and you choose which test group to download.

• Automatic currency targeting for Post-Purchase Upsells. Offers lock to your store's default currency, so they reach shoppers who can actually see them.

• Target merchandising experiences by subscription status. Show or hide upsells and checkout offers based on whether the shopper's cart contains a subscription.

• Add more volume discount tiers. The limit went from 5 tiers to 7.

• Read smarter experience summaries. Summaries know when a test is too early to call, and a refresh button keeps them up to date.

• Read clearer test verdicts. Verdicts now say when the underlying test data is filtered.