There is a version of product recommendations that makes a store feel attentive, like a good shop assistant who noticed what you were looking at and said here, you might also like this. And there is a version that makes a store feel like it is following you around with a clipboard, shoving unrelated junk in your face because an algorithm decided you were a conversion waiting to happen. Same feature, wildly different feelings. The gap between them is almost entirely about judgment, not technology.
I have set up recommendations on a lot of stores, and the merchants who do well with them are not the ones with the fanciest engine. They are the ones who understood what each type of recommendation is actually for, and resisted the urge to turn everything on at once. So let me walk through the types, when each one earns its place, and how to stay on the helpful side of the line.
The main types, and what each is good at
Think of these as different tools for different moments, not interchangeable widgets.
- Recently viewed. The most quietly effective of the bunch, because it is not really a recommendation at all. It is a reminder of what the shopper already chose to look at. There is no guesswork, no risk of irrelevance, and no creep factor, because the person watched themselves view those items. It shines on product pages, in the cart, and for returning visitors who need a way back to what caught their eye.
- Related products. The classic you also might like row on a product page. When it is good, it shows genuine alternatives and complements: other colors of the same shirt, the matching accessory, a similar item at a nearby price. When it is bad, it shows whatever shares a tag, which is how you end up recommending a phone case on a dog-food page.
- Cross-sell and upsell. This is the add the belt to go with the trousers move, best placed at the cart or just before checkout. Cross-sell adds a complementary item, upsell nudges toward a better version of what is already in the cart. Done with restraint it lifts average order value. Done greedily it feels like a timeshare pitch.
- Bestsellers and trending. The safe, social-proof play. Great for new visitors with no history, great on the homepage, great as a fallback when you have nothing personalized to show. The limitation is that it is the same for everyone, so it does no real personalization work. It is a strong default, not a strategy.
When each one works
The mistake I see most often is treating these as a stack you enable everywhere. They each have a natural home.
- New visitor, no context yet. Lead with bestsellers and trending. You do not know them, so show them what most people love. It builds trust and gives them an obvious entry point.
- Deep on a product page. This is where related products and recently viewed both belong. The shopper is evaluating something specific, so give them nearby options and a reminder of what else they considered.
- At the cart. Cross-sell the genuine companion item and surface recently viewed as a last-chance reminder. This is the right moment for I almost forgot, not for a wall of unrelated suggestions.
- Returning after a gap. Recently viewed, every time. Someone coming back three days later does not want to rediscover your catalog. They want the thing they were looking at.
Match the recommendation to the moment and it reads as service. Mismatch it, like upselling someone who has not added anything yet, and it reads as noise.
How to avoid the creepy and irrelevant trap
The two failure modes are different and both are deadly. Irrelevant recommendations make you look incompetent. Creepy ones make you look untrustworthy. Here is how I keep clear of both.
- Relevance beats cleverness. A simple rule that reliably shows items from the same category at a similar price will beat a sophisticated engine that occasionally produces something bizarre. The occasional bizarre result is what shoppers remember, and it taints the whole store.
- Do not reach past the obvious. If someone is looking at running shoes, show running shoes, socks, and insoles. Do not show a blender because some other runner happened to buy one. Thin association is where recommendations get weird.
- Keep the self-evident stuff front and center. Recently viewed never feels creepy because the logic is visible to the shopper. The more a recommendation explains itself, the safer it is. When in doubt, lean on the types the shopper can reason about.
- Respect the volume. One good related row and one recently viewed strip is plenty on a product page. Five stacked carousels is a store that does not trust its own merchandising, and shoppers feel that.
- Exclude the obvious misses. Out-of-stock items, the product currently being viewed, things the shopper already bought. Every one of these, left in, makes the whole feature look careless.
That last cluster of details is exactly why I usually reach for a dedicated app rather than hand-rolling the logic in a theme. The Scurix Recently Viewed Products app, for instance, handles the recently viewed piece, the one that is both the safest and the most consistently effective, and keeps it filtered, persistent, and matched to the theme so it never slips into feeling bolted on. Whether you build or buy, the discipline is the same: show things that obviously make sense, and nothing that does not.
Placement, briefly
Placement decides whether anyone sees these at all. A few rules I hold to:
- Product page, below the fold: related products and recently viewed, in that order. This is the densest intent on the whole site.
- Cart or cart drawer: one honest cross-sell plus a recently viewed reminder. Keep it tight.
- Homepage: bestsellers for new visitors, recently viewed for returning ones.
- Never the checkout fields themselves. Once someone is entering a card number, get out of the way. A recommendation at that exact moment is friction dressed up as helpfulness.
Good recommendations are not about showing people more. They are about showing people the right next thing at the moment they are open to it, and knowing when the right number of suggestions is zero. Get the judgment right and the feature feels like a helpful shop assistant. Get greedy and it feels like the clipboard. With BFCM around the corner and carts about to get busy, it is worth deciding which one your store is going to be before the traffic shows up.
