Instead of choosing upsell products by hand, you can let Atlas pick them per cart with Use AI-recommended products. There are two algorithms, Related and Complementary, and they get their data from completely different places. That difference decides which one will actually work on your store.
In this article:
Turning it on
In the cart drawer editor, under Product Upsells, open Configure Upsell Products and enable Use AI-recommended products. Then pick a Recommendation algorithm: Related or Complementary.
Related recommendations
Related works automatically from data Shopify already has, using three signals:
Note: The 'Related' algorithm uses the following to select products:
- Purchase history: Products that have historically been purchased together.
- Product description: Products with similar descriptions.
- Related collections: Products that share collections.
Nothing to configure. It works out of the box, and it improves as your store accumulates order history.
The catch: a new store has no purchase history, so early recommendations lean on descriptions and collections. Those are weaker signals, and the results can be loose until you have real sales behind them.
Complementary recommendations
Complementary uses relationships you define. The note is blunt about it:
Note: Complementary recommendations need to be set up manually.
They are configured in Shopify's Search & Discovery app, not in Atlas. You tell Shopify which products complement which, and Atlas surfaces those pairings in the cart.
To set them up:
Install Shopify's Search & Discovery app if you do not have it.
Open a product in the app and add its complementary products.
Repeat for the products you care about.
⚠️ If you select Complementary without configuring anything in Search & Discovery, no upsells will appear. Atlas has nothing to show, because the relationships do not exist yet.
Which to choose
Your situation | Choose |
Established store with plenty of order history |
|
New store with few orders | Manual selection, or |
You know exactly which products pair |
|
Large catalogue, no time to configure pairs |
|
Small catalogue, a handful of clear pairings | Manual selection |
💡 On a store with real order history, Related is the better default: it needs no maintenance and it improves on its own. Complementary is worth the setup effort when you have strong opinions about what goes with what and the data does not know them yet.
When manual selection is better
Automatic recommendations optimise for relevance, which is not always what you want.
Choose products yourself when:
You have one clear winner. If a single accessory attaches to most orders, offer it every time rather than letting an algorithm rotate.
Margin matters more than relevance. Recommendations do not know your margins and may surface your least profitable products.
You are clearing stock. Algorithms will not know you need to move a particular item.
Your catalogue is small. With a handful of products, choosing by hand takes minutes and beats any algorithm.
You can also start with AI recommendations, watch which products actually get added in your analytics, then switch to selecting those products manually.
If no recommendations appear
Work through these in order:
Check the algorithm. If it is set to
Complementary, confirm you have configured complementary products in Search & Discovery.Check your order history.
Relatedneeds data. A store with very few orders may not produce recommendations for every cart.Check the products are published to the Online Store sales channel and in stock.
Check the trigger. If your upsell is set to
Specific Trigger Products, it only shows when one of those is in the cart.Switch to manual selection to confirm the module itself works. If manually chosen products appear, the module is fine and the issue is the recommendation source.