Where Product Recommendations Earn Their Place on a Store Page
Recommendations earn their space where shoppers are deciding, not everywhere they fit. Product pages suit related and alternative items. The cart suits true complements only. Post purchase suits one clear add on. Homepages suit returning visitors with history. Each placement needs its own logic and its own test.
Product pages: alternatives and completions
On a product page the shopper has chosen an item but not committed. Two recommendation jobs matter here: similar alternatives for the undecided, and completing items for the decided. Mixing both into one undifferentiated carousel serves neither. Separate them, label them honestly, and order by relevance to the viewed item.
For more on this, see this related guide and our earlier explainer on the same theme.
Watch cannibalisation. A recommendations block that pulls clicks away from the main item and converts none of them is decoration with a cost. Measure the placement on total page conversion, not on its own click rate, which rewards distraction beautifully.
The cart: complements or nothing
Cart recommendations have the highest interruption cost in the store, because the shopper is trying to pay. Only genuine complements belong here: the battery, the refill, the case that fits this model. Anything exploratory belongs earlier in the journey, not between the customer and the payment button.
Keep the cart block small, priced low relative to the order, and removable in one tap. Stores that test this honestly often find one tight complement block outperforms a full carousel, and that removing cart recommendations entirely beats a weak implementation.
Post purchase: the underused placement
After payment, attention is high and interruption cost is zero. One relevant add on, offered once, with a clear decline, converts well and annoys almost nobody when it is genuinely related to the order. This is also the safest place to test recommendations for new categories.
Readers also found this companion piece from our archive useful.
Respect the moment. The post purchase offer should extend the order's purpose, not start a second shopping trip. Frequency caps matter here too: a customer who declines twice has answered, and the system should remember it.
Testing placements one at a time
Change one placement, hold the others, and run long enough to cover weekday and weekend behaviour. Recommendation tests are noisy because a small number of high value orders can swing results. Decide the metric and the minimum run before starting, and write both down.
Kill placements that do not beat the control on revenue per session. Screen space and shopper attention are finite, and a neutral block is not neutral: it pushes everything below it down. The stores that win with recommendations run fewer, better placements than the stores that run them everywhere.
Frequently asked questions
How many products should a recommendation block show?
Usually four to eight on desktop product pages, fewer on mobile and in the cart. More choice rarely converts better and slows the page. Test block size explicitly rather than assuming the widest carousel wins.
Do recommendations hurt SEO or page speed?
They can if they load heavy scripts and large images for every visitor. Load recommendation blocks efficiently, size images properly, and avoid layout shift. A fast relevant block has no meaningful SEO downside.
Should new stores use recommendations at all?
Yes, but start with rule based logic: best sellers in the same category and true complements. Algorithmic personalisation needs traffic and history to beat simple rules. Earn the data first, then let the models take over.
What metric proves a placement works?
Revenue per session for the page or step where the placement sits, compared against a control without it. Clicks on the block alone mislead, because clicks can come at the expense of the main purchase.