Shopify Recommendations: Manual vs AI

Manual and AI product recommendations solve the same customer question: what goes with this purchase? Manual selection gives you direct merchandising control. Data-driven selection can identify patterns across orders. The right choice depends on your catalogue, available evidence and what customers actually see.

What manual recommendations mean in Cart Uplift

Manual recommendations are products you select yourself. In the theme editor, open the Cart Uplift app embed and find Manual products in the recommendation settings. Select a short list, save, and check the published drawer across common baskets. Review any recommendation-mode settings in use; the picker guidance is to leave it empty for automatic suggestions.

The picker supplies a selected list. It does not by itself create a different “when product A is in the cart, show product B” rule for every item. For a specific trigger and discounted companion, review targeted cart offers and test their conditions.

When to use manual recommendations

For a new product, a gift collection or a small catalogue, you may already know which companion makes sense. Useful selection considers the occasion, routine, compatibility, available variants and the customer's likely budget. It can also avoid a product whose fulfilment cost makes it a poor add-on.

Manual does not mean set and forget. Review chosen products when stock, seasons or prices change. A pairing that was sensible last month may no longer be available.

What AI can contribute

Repeated co-purchases can reveal relationships that are hard to spot across a larger catalogue. Some systems also use product information or offer outcomes. Their value depends on the relevance and volume of the data, how unavailable products are handled and what the system optimises.

An algorithm does not necessarily optimise profit, understand every size relationship or learn reliably from a few orders. More time installed is not a guarantee of better recommendations. Review the output rather than relying on the label.

There is no universal order-count cutoff

A concentrated catalogue with repeated purchases can produce useful evidence sooner than a large catalogue with scattered orders. Promotions and one-off gifting orders can distort both. Instead of a fixed monthly-order rule, ask whether the main pairings recur and whether enough eligible shoppers see them to assess performance.

When evidence is thin, a few deliberate available pairings can be a sensible starting point. Keep automated fallback suggestions under review. Our FBT guide explains why order history needs interpretation.

Use manual and automated features deliberately

Use merchant-selected bundles or offers where you have a clear reason for a pairing. Use automated suggestions where purchase evidence is more useful or maintaining a large set manually is impractical. Check which controls your app supports: a general manual product list is not the same as a separate rule for every basket.

Compare equal opportunities

Review visible offer impressions, accepted adds and completed paid purchases. A manual offer with more exposure may generate more revenue without having a higher acceptance rate. Compare similar products, placements and periods, and record unavailable variants.

Attribution reports show tracked sales. They do not by themselves prove incremental revenue or that one recommendation method caused a better result. For small stores, use counts and customer feedback alongside rates.

Using Cart Uplift

Cart Uplift offers manual recommendation selection, bundles and automated recommendations. Start with the controls that express your merchandising choice, check the published suggestions and refine one offer at a time. You keep control of your products and promotion choices.