“You may also like” is one of the oldest ideas in ecommerce, and one of the most often done badly. A shopper looking at a winter coat gets shown three more winter coats. Someone who just bought a phone case is offered the same phone case in another color. The block is there, but it isn’t helping anyone.

AI-based recommendations promise to fix this by learning from shopper behavior instead of relying on rules you set by hand. This post explains how they work in plain terms, where they make the most difference, and how to tell whether they’re earning their place on your pages.

Three ways stores recommend products

Before looking at AI, it helps to see the options side by side.

WooCommerce itself supports the first two out of the box. In the product editor, the Linked Products tab lets you choose upsells and cross-sells by hand, and the default related products block uses shared categories and tags. Our guide to WooCommerce related products covers those settings in detail.

How AI recommendations work (without the math)

“AI” in this context usually means a system that looks for patterns in shopper behavior. The most common methods are simple to understand:

“Customers who bought this also bought”

This is the classic approach, often called collaborative filtering. The system looks at orders and notices which products tend to appear together. If many people who buy a French press also buy a hand grinder, the grinder becomes a strong suggestion for the French press page, even if the two are in different categories.

“Products similar to this one”

This is content-based filtering. The system compares product attributes such as category, tags, price range, color and description text, and suggests items that resemble what the shopper is looking at. It’s useful for new products that don’t have sales history yet.

“Based on what you’ve looked at”

Here the system uses the current visitor’s own behavior: products viewed, items in the cart, past orders if they’re logged in. Two people on the same product page can see different suggestions.

Most modern recommendation tools combine these methods. For example, they might use “similar products” for a brand-new item and switch to “bought together” once there’s enough order data.

Where recommendations make the biggest difference

Not every placement is equal. Each one should match what the shopper is trying to do at that moment.

Product pages

The shopper is evaluating one item. Two kinds of suggestions work here:

• Alternatives for people who aren’t convinced (“similar products”)
• Complements for people who are (“goes well with”)

Keep alternatives below the main product information so they don’t distract from the Add to cart button.

Cart page and mini cart

The shopper has decided. Show only complements, and keep them cheap relative to the cart value: accessories, refills, small add-ons. Pushing a second expensive item here usually backfires.

Thank-you page

The purchase is done, so there’s nothing to lose. It’s a good place for “complete the set” suggestions or an offer that’s valid for the next few days. See our guide to customizing the WooCommerce thank-you page for how to edit that page.

Emails

Post-purchase and browse-based emails can include personalized suggestions. They work well a few weeks after purchase, when the shopper might need the next item.

Exit-intent and on-site messages

When a visitor is about to leave, a short popup with two or three products they viewed, or popular alternatives, can bring them back into the store.

A worked example

Consider a hypothetical kitchenware store with about 400 products. It currently uses WooCommerce’s default related products, which show items from the same category.

Before: On the page for a cast iron skillet, related products shows three more skillets.

After switching to behavior-based recommendations:

• Product page, “Goes well with”: a skillet handle cover, a chainmail scrubber and a seasoning oil. These came from order data showing they’re often bought together.
• Product page, “Compare with”: two other skillets at different sizes, placed lower on the page.
• Cart page: the seasoning oil and the scrubber, both low-priced.
• Follow-up email, three weeks later: a Dutch oven from the same brand, based on what similar customers bought next.

The store owner didn’t hand-pick any of these. The system learned them. The owner’s job was to decide the placements and check the results.

How to measure whether it’s working

Recommendation tools often report “revenue from recommendations,” but that number can flatter them, because some of those shoppers would have found the product anyway. Better ways to judge:

• Revenue per visitor across the whole store, before and after, or ideally in an A/B test with recommendations on and off
• Average order value and items per order
• Click-through rate on each placement, to spot the weak ones
• Conversion rate on product pages, to make sure recommendations aren’t distracting people from buying the main product

If your tool allows it, run a proper test: show recommendations to half your visitors and the old setup to the other half for a few weeks. That tells you what the recommendations actually add.

Getting better results from any recommendation tool

AI isn’t magic. It works better when your store gives it good material.

• Clean up your categories and tags. Content-based suggestions depend on them.
• Write real product descriptions. Text similarity is part of how many tools match products.
• Mark out-of-stock items clearly and make sure the tool doesn’t recommend them.
• Set exclusions for things you never want suggested, like gift cards, samples or discontinued lines.
• Give it time. Behavior-based methods need orders to learn from. A store with a few orders a week will see weaker suggestions at first than one with hundreds.
• Limit the number shown. Three or four good suggestions usually beat a carousel of twelve.

Choosing a tool

For WooCommerce, you can go several ways:

• Stay with native features if your catalog is small. Hand-picked upsells and cross-sells for your top 20 products can go a long way.
• Use a rule-based plugin to get more control over which products appear where.
• Use an AI recommendation plugin once your catalog and order volume are big enough that manual curation is no longer practical. iConvert Optimizer includes AI product recommendations built for WooCommerce, with placements on product pages, the cart and exit-intent messages.

Questions to ask before choosing:

• Does it use my WooCommerce order data directly?
• Can I control placements and exclude products?
• Does it slow down my pages? Check with a speed test after installing.
• Can I measure its impact with an A/B test?

Recommendations should feel like help

Good recommendations feel like a helpful shop assistant: they notice what you’re looking at and point you to the thing you didn’t know you needed. Bad ones feel like a random shelf. Start with the placement that fits your store best, usually the product page or cart, feed the system clean product data, and measure the effect on revenue per visitor rather than trusting a single “recommendation revenue” number.

If your catalog has outgrown hand-picked cross-sells, iConvert Optimizer is one way to add AI recommendations to WooCommerce.