Where Product Recommendations Belong: Placement That Earns Clicks
Product recommendations earn their keep where they match the shopper current intent: alternatives on a product page when the item is not quite right, complements in the cart when the main decision is made, and rediscovery modules on the homepage for returning visitors. Placed without regard to intent, the same module becomes page decoration that shoppers learn to ignore.
Intent changes by page
A shopper on a product page is deciding about that product. Recommendations there should answer the two live questions: is there a better version of this, and what goes with it. A shopper in the cart has decided; only low friction complements belong. A homepage visitor returning after a week wants their thread picked back up.
For more on this, see this related guide and our earlier explainer on the same theme.
One global module pasted everywhere ignores all three contexts. The fix is not more modules but assigned jobs: each placement exists to answer the specific question shoppers ask on that page, and its algorithm and heading should say so plainly.
Headings and honesty
The heading sets the contract. You may also need, Pairs well with, and More in this collection make different promises, and shoppers click when the promise matches the row beneath it. Generic headings like Recommended for you over random rows train visitors to skip the section entirely.
Relevance should be inspectable at a glance: same category logic, compatible sizes, in stock, and not the item already on the page or in the cart. Every obviously wrong tile taxes the credibility of the correct ones beside it.
Placement traps that waste good modules
Below the footer links, buried under reviews nobody reaches, or competing with three other carousels, good recommendations die of placement. Equally wasteful is the interruptive version: recommendation popups that block a shopper mid decision to suggest something unrelated to the task in front of them.
Readers also found this companion piece from our archive useful.
On mobile, placement discipline matters more. A module that pushes the buy button or key product information below several swipes of suggestions has traded the main conversion for a speculative one. Keep the primary action sacred and let recommendations support it.
Testing placement like it matters
Move one module at a time and measure click through, add to cart from the module, and overall page conversion, because a module can win clicks while cannibalizing the main product purchase. Holdouts settle arguments that dashboards full of attributed revenue cannot.
Retire or redesign modules that fail two honest tests. Page space is inventory; every weak module occupies a slot a stronger one, or simple breathing room, could use. Placement is a portfolio decision, reviewed quarterly, not a one time design choice.
Frequently asked questions
How many recommendations should a module show?
Four to eight is the usual working range on desktop, fewer effectively visible on mobile. More choice is not more help; long carousels mostly go unscrolled. Show the strongest few, make them visibly relevant, and let search and category pages carry the full catalog.
Are frequently bought together modules worth it?
Yes, when the pairing is genuinely complementary and priced as an easy add. They fail when the algorithm pairs random co purchased items or duplicates what is already in the cart. Review the top pairings manually; the embarrassments are always visible at a glance.
Should recommendations appear on the homepage for new visitors?
For new visitors you have no behavioral data, so bestsellers and category entry points outperform fake personalized picks. Save true personalization for when there is a signal: a viewed category, a past order, or an email arrival context. Honest generic beats pretend personal.
How do I know a module is actually working?
Measure module attributed revenue against a holdout where possible, and watch click quality: clicks that bounce straight back are not engagement. A module nobody clicks is not neutral; it costs page weight and attention, so the bar for keeping it should be real usage.