# How do you keep recommendation carousels from showing the same five products everywhere? | KilterMuse

Source: https://kiltermuse.com/blog/2026-10-29-recommendation-carousels-same-products/

KilterMuse answers


October 29, 2026

**Short answer:** Cap how often one product can appear across the site in a day, reserve slots in each carousel for candidates outside the top sellers, and judge carousels on the revenue of the whole rail rather than the first position. Popularity should be an input to ranking, never the ranking itself.

## The loop that creates the problem

A carousel that ranks purely on predicted conversion discovers the same handful of products win almost everywhere. So it shows them everywhere. They collect more clicks and sales because they are everywhere, which confirms the prediction, which shows them more. Within a month the site has five products doing the work of a catalogue and a long tail that never gets the exposure needed to prove itself.

Nobody designed that outcome. Each carousel was tuned sensibly in isolation. The sameness is a property of the whole system, which is why fixing one rail at a time never fixes it.

## Diversity is a control, not a garnish

The standard controls are simple to state. Frequency caps per product across placements. A diversity rule inside each rail: no more than a set share of slots from one category, brand, or price band. And exploration slots, one or two positions per rail reserved for products with a plausible case and thin data, rotated so the system keeps learning what it does not yet know.

Exploration is the one teams cut first because its slot underperforms the champion it replaced this week. That accounting is rigged. The champion slot is spending an asset, exposure, and reporting the income as insight. Exploration buys the information that next quarter champions are made of, at a price of a few points on one slot.

## Judge the rail as a rail

Position level click reports make the first product look like a genius and the rest like passengers. Judge the carousel whole: revenue per session for visitors who saw it, against a holdout that shows a curated or category ranked alternative. Rails that win on that measure sometimes have a weaker first position and a stronger fifth, which is what a rail is for.

Also count unique products sold through recommendations per week. That number is the loop made visible. If it falls month after month while carousel revenue holds, the system is concentrating, and the catalogue is quietly becoming a shop window with a warehouse behind it.

## Start with the caps

You do not need a new platform. Set an appearance cap for the top products, add one exploration slot to the two highest traffic rails, and re read unique products sold in a month. The champions will still lead, because they are good. They will just stop being the entire recommendation strategy.

Keep reading: [How should recommendations change when a shopper keeps returning the same category?](https://kiltermuse.com/blog/2026-10-29-recommendations-repeat-returns-category), [Recommendations for gift buyers: a different intent needs a different feed](https://kiltermuse.com/blog/2026-10-29-recommendations-gift-buyers-different-feed), and [When to show bestsellers and when to show long tail picks](https://kiltermuse.com/blog/2026-10-29-bestsellers-versus-long-tail-picks) cover the nearby cases in this series.

## Frequently asked questions

### Why do the same products appear in every recommendation carousel?

Because ranking on predicted conversion alone creates a feedback loop. Popular products get shown, sell because they are shown, and their stronger data wins them even more placements. Without caps or diversity rules, the system concentrates on a handful of items.

### Do diversity rules hurt conversion?

In the short term an exploration slot may convert below the champion it replaces. Judged properly, on whole rail revenue per session and on unique products sold over a month, diversity controls usually protect revenue by keeping more of the catalogue discoverable and sellable.

### What is an exploration slot?

A reserved position in a carousel for products with a reasonable case but little exposure data, rotated regularly. It exists so the system learns what else could sell instead of only re confirming what already does.

### How can a store measure recommendation concentration?

Track unique products sold through recommendations per week and the share of recommendation revenue from the top five products. Rising concentration with flat unique counts is the warning sign, visible long before total revenue moves.