Recommendation Fatigue: When Personalised Picks Start Hurting Sales
Recommendation fatigue sets in when shoppers see the same suggested products so often that they stop noticing them. Click rates decay while impressions stay high, dismissal actions rise, and the recommendations start competing with the shopper's own search intent instead of supporting it.
What fatigue looks like in the numbers
The pattern is distinctive: impressions hold steady, click through rate slides week by week, and conversion per click stays roughly normal. That last part matters. The products are still fine when people do click. The problem is attention, not assortment, and it means the placement has trained shoppers to look past it the way people look past banner ads.
For more on this, see this related guide and our earlier explainer on the same theme.
Segment the data before concluding anything. Fatigue concentrates in repeat visitors, who have seen the same carousel dozens of times. New visitors often still respond well to the identical module. If your blended metrics hide that split, you will redesign a module that is only broken for your most loyal audience, which is precisely the audience you can least afford to bore.
Why the same items keep resurfacing
Most engines rank by predicted affinity and barely account for exposure history. The shopper's strongest signal items win every auction for their attention, so the same five products appear on the homepage, the product page, the cart, and the email. Each system is locally rational. Together they produce a store that seems to sell almost nothing else.
Seasonality makes it worse. A gift bought in December keeps recommending itself in March, because purchase history is a strong signal and few engines age it out aggressively. The shopper has moved on. The model has not, and every stale recommendation teaches the shopper that the suggestions are about the store's memory, not their current intent.
Fixes that restore attention
Impression caps are the simplest cure: limit how many times an item can be recommended to one visitor within a period, and force rotation once the cap is hit. Pair caps with a diversity rule so a module cannot fill with near duplicates of one product type. Shoppers read variety as intelligence and sameness as laziness, even when the sameness is statistically justified.
Readers also found this companion piece from our archive useful.
Give real estate back to intent. On search and category pages, recommendations should support the task, with filters and comparisons, rather than interrupting it with personalised picks. Reserve heavy personalisation for the homepage, post purchase, and email, where the shopper has no stronger signal of what they want next. Matching the module's ambition to the page's intent ends most fatigue complaints.
Testing whether you actually fixed it
Judge changes on repeat visitor click rate and downstream conversion over several weeks, not on day one novelty spikes. Any new module gets a curiosity bump. The honest test is whether engagement is still higher after the same shoppers have seen it twenty times, which is the condition that produced fatigue in the first place.
Also watch hide and dismiss actions as a first class metric. A healthy module is ignored politely. A fatigued module is actively dismissed, and rising dismissals often precede the click rate slide by weeks. Catching fatigue at the dismissal stage lets you rotate creative and caps before revenue from the placement actually falls.
Frequently asked questions
How is recommendation fatigue different from a bad algorithm?
A bad algorithm shows irrelevant products from the start, so clicks are low for everyone including new visitors. Fatigue shows up as decay: strong early performance that slides as the same visitors see the same items repeatedly. The fix is caps and rotation, not a new model.
How many times should one product be recommended to a visitor?
There is no universal number, but most stores see returns diminish after a handful of exposures in a short window. Set a cap, then test it. The right cap is where repeat visitor click rate stops decaying week over week.
Should purchased products ever be recommended again?
Only when the product is genuinely replenishable or has a natural repeat cycle, and even then on a timed basis. For one off purchases, purchased items should be suppressed or shifted to complements and accessories instead of the item itself.
What metric reveals fatigue earliest?
Dismissal and hide actions from repeat visitors, followed by their click rate trend. Blended click rate moves late because new visitors mask the decay. Segmenting by visit count is what makes fatigue visible in time to fix it.