How to Design Recommendation Placements: Where to Put Them on Homepage, Detail, and Cart

Summary: The three decisions in recommendation placement design are: which page to place it on, what content to recommend, and how much attention it takes. The homepage serves weak-intent exploration, the detail page assists decision-making with similar and complementary items, the cart drives attach purchases, and the search results page needs careful mixed placement; every placement needs per-slot metrics, and cannibalization measurement confirms it brings incremental gains rather than just traffic reshuffling.

Recommendation systems get plenty of algorithmic discussion, but what actually drives results is often a more basic question: where the placement sits, what it shows, and how much space it occupies. The same algorithm can deliver significant incremental lift as an attach recommendation on the cart page, but can directly break relevance if mixed into the top slot of search results. This article breaks down recommendation placement design page by page: what each position's user intent is, what type of recommendation fits, where the quantity ceiling is, what metrics to use for acceptance, and finally how to allocate traffic and measure cannibalization across placements.

Key Takeaways

The first principle of placement is to match page intent: whatever purpose the user brings to the page, the recommendation fills the corresponding information gap.
Homepage recommendations are a weak-intent exploration ground: category and popular-item recommendations dominate, with naturally low click-through rates – do not judge them by the detail page's metrics.
Detail page recommendations assist decision-making: similar items help comparison, complementary items help bundling, and the two should be presented separately, not mixed.
Cart recommendations are among the highest-converting positions: low-decision-cost attach items plus low-stock prompts deliver direct returns.
Be cautious with placements on the search results page: mixing them in when intent is clear hurts relevance trust; only zero-result and broad-query cases suit fill-in placement.
Account for each placement separately, and use "cross-entry conversion" to confirm incremental gains – if total conversion rises but cross-entry conversion stays flat, it is mostly traffic reshuffling.

1. The Three Decisions of a Recommendation Placement

Designing a placement starts with three questions.

Where to place it. The basis is page intent and user state: what the user is thinking on this page, where they are stuck, what information is missing. The placement should appear in the gaps where "the user has a need but the page lacks the information."

What to recommend. Content type follows page intent: exploration positions push categories and popular items, decision positions push similar and alternative items, conversion positions push attach and replenishment items. Wrong type placement cannot be saved by a better algorithm.

How much space. Too large a placement squeezes out the main content; too small gets no exposure. The common constraint is "the placement must not push the page's main-task completion path off the first screen," and each placement gets a quantity ceiling to avoid the page becoming a recommendation platter.

The shared principle behind the three decisions is: a placement supplements the page's main task, it is not a competitor. Whether it competes is a data question: after adding the placement, does the page's main conversion metric go up or down.

2. Homepage: The Weak-Intent Exploration Ground

Homepage users have fuzzy goals; the placement's job is to spark interest, not to hit precisely.

What to recommend: category-navigation recommendations (category entries sorted by user interest), popular and new-item picks, and history-based "continue browsing." This is where "broad and stable" content fits – personalized category ranking suits the homepage better than single-item recommendations, because single-item hit probability is low, and a category entry retains navigation value even without a click.

How big: reserve the first screen for brand and main actions, placing the recommendation module in the secondary area below or to the right of the first screen. Judging homepage recommendations by the detail page's click-through standard is a common misjudgment – the weaker the intent, the lower the recommendation click-through, and each position should have its own baseline.

Acceptance metrics: homepage recommendation module click-through rate, depth after entering from a recommendation (whether the user continues to a product page), and changes in homepage bounce rate.

3. Detail Page: Decision Assistance

Detail page users have clear intent (evaluating this item); the placement assists decision-making.

Similar and complementary recommendations: help comparison and evaluation, answering "are there other options." Place in the middle-lower area of the detail page. Note that ranking must respect a similarity floor – mixing unrelated bestsellers into the similar slot looks good for short-term clicks but teaches users to ignore the module long-term.

Complementary recommendations: answer "what else is missing," the main battlefield for attach purchases (recommend a tempered-glass film when buying a phone case). The bundling logic must match real usage scenarios; pure margin-based bundling gets spotted by users quickly.

Bundle and exchange: push low-decision-cost small items to already-decided users. This position directly and stably lifts average order value.

Acceptance metrics: each module's click-through rate, add-to-cart rate, attach conversion rate, and whether the detail page's main conversion is harmed by the placement – detail pages with too many recommendation modules commonly show a drop in the main item's add-to-cart rate.

4. Cart and Checkout Page: The Final Push

Cart page users have expressed clear purchase intent, making it the highest-converting position for recommendations.

What to recommend: low-decision-cost attach items (consumables, accessories, categories often bought together with cart items), bundle items (for threshold discounts), and stock and offer prompts ("an item in your cart is low on stock"). Recommendation content must be strongly relevant to cart items – relevance requirements here exceed any other position, and users have very low tolerance for "irrelevant items pushed on the cart page."

How big: one to two modules, below the checkout button or in the sidebar. Never push the checkout button to a position that requires scrolling to find – Baymard Institute's research on checkout flows repeatedly confirms that every bit of friction on the checkout path is amplified by the abandonment rate.

Acceptance metrics: placement add-to-cart rate, attach conversion rate, and changes in cart-to-payment conversion rate (must confirm the placement does not slow checkout).

5. Search Results Page: Careful Mixed Placement

The search page is the strongest-intent page; placements here must be extremely restrained.

For explicit queries: result relevance is everything; mixing a placement into the result list directly breaks the "search returns what you queried" trust. If placed at all, put it only in a separate area below the result list or in the sidebar, visually distinct from the main results.

For zero-result and broad queries: this is where the search page suits fill-in placement – when nothing is found, give related category recommendations and bestseller fill-ins, turning "a wasted trip" into "browse while you're here." Zero-result fill-in is the typical seam of search-and-recommendation synergy; the site's earlier zero-result optimization topic gives specific fill-in strategies.

Acceptance metrics: fill-in module click-through rate and recovered conversion rate, and whether the main result's top-slot click-through is affected.

6. Traffic Allocation and Cannibalization Measurement

Once multiple placements coexist, the real question is: do they bring incremental gains, or just reshuffle traffic among themselves.

For allocation, give each placement an initial exposure weight, then adjust dynamically by per-slot returns: raise the quota for positions with high click-through and conversion contribution, lower or retire those with both low. Netflix's recommendation architecture practice treats scenario-based online decision-making as a core capability; the same idea on the operations side is: recommendation content allocation should adjust dynamically by real-time scenario, not be statically filled.

For measurement, beyond each placement's clicks and conversions, you must look at cross-entry metrics: the ratio of users who enter from slot A but eventually convert at slot B or the main path. The method for judging incremental gains is detailed in the site's search-recommendation synergy topic: if total conversion rises but cross-entry conversion stays flat, it is mostly traffic reallocated between entries; a rise in cross-entry conversion shows the positions are cross-driving traffic. Another useful signal is "average browse depth per user" – incremental placements deepen browsing, while cannibalizing ones only change the entry distribution of clicks.

7. Common Pitfalls

Packing more and more placements on a page: each slot yields a little, but combined the main-path conversion keeps sliding and the total account actually loses. Set a total ceiling.
One strategy for every position: the algorithm outputs uniformly, positions only change skin, ignoring page-intent differences, letting weak positions drag down strong ones.
Only measuring click-through: high click-through does not equal high contribution; bundle-position clicks are naturally high but amounts naturally small, so set a metric mix per slot.
Ignoring cannibalization: declaring success just because total conversion rose after adding a placement, when it actually credited natural conversions to the placement. Apply cannibalization measurement.

FAQ

Q: Should the homepage placement push single items or categories?

A: Categories and scenarios first, single items second. Homepage users have weak intent and low single-item hit rates, but a category entry retains navigation value even without a click; curated single items suit a supplementary module showing popular and new items. Judging homepage recommendations by the detail page's click-through baseline is a common source of misjudgment.

Q: Can the detail page's similar and complementary items be mixed in one module?

A: Not recommended. The two serve different user tasks – similar items serve comparison and substitution decisions, complementary items serve attach purchases; mixing lets users who "want to compare" get distracted by bundles. Split into two separately named modules ("viewed also viewed," "buy together") so each module's clicks and conversions can be attributed separately.

Q: Will placing recommendations on the cart page affect checkout conversion?

A: It depends on position and content. When the recommendation module sits below the main checkout button or in the sidebar, content is strongly relevant, and module count is limited to one or two, it usually drives attach purchases without hurting checkout; pushing the checkout button off the first screen, or pushing irrelevant items that draw users out of the cart, directly drags down payment conversion. Watch the main metric "cart-to-payment conversion rate" before and after launch.

Q: How should placements in search results be placed to avoid hurting experience?

A: For explicit queries, do not mix into the result list – place only in a separate area below the list or in the sidebar; open fill-in for zero-result and broad queries, giving related categories and bestseller recommendations. The criterion is that the main result's relevance trust is not impaired – if top-slot click-through and query-rewrite rate rise, the fill-in strategy needs tightening.

Q: Is there an empirical upper limit on the number of placements?

A: Set it inversely by the importance of the page's main task: the detail page's main task is evaluating the main item, so two to three recommendation modules; the cart page one to two; the homepage can have more, but no more than one main recommendation module per screen. Rule of thumb: before adding a new placement, confirm the existing per-slot return curve has not turned downward.

Q: How do you tell whether a new placement is real incremental gain or just stealing traffic from elsewhere?

A: Look at three sets of data: per-slot metrics (clicks, add-to-cart, conversion), page main-task metrics (whether main conversion is harmed), and cross-entry metrics (the mutual traffic-driving ratio between placements and the main path). If total conversion rises but cross-entry conversion stays flat, it is mostly traffic reshuffling; a rise in cross-entry conversion with no drop in main-task metrics is the attributable incremental gain.

About Tengence

Tengence Search addresses the recommendation and search scenarios described in this article, providing scenario-based recall and ranking capabilities, supporting zero-result fill-in, cart attach, and cross-entry behavior feedback, with placement strategies configurable and verifiable per position. Tengence (TENGENCE) is the enterprise growth engine of the AI era, focused on public-traffic acquisition and private-domain conversion; its core products are Tengence GEO (dual-engine GEO + SEO) and Tengence Search.

Data Sources

Netflix TechBlog "System Architectures for Personalization and Recommendation": the division of offline, near-line, and online three-layer architecture, with the online layer supporting scenario-based real-time adjustment of recommendation allocation.
Baymard Institute "Checkout Usability" research topic: systematic research on checkout-path friction and abandonment rate, supporting the cart placement's position constraints.
Nielsen Norman Group "Conversion Rates: How to Measure and Improve": the methodology basis for interpreting conversion effects by segment position and behavior path.