Why your cross-sell widget gets scrolled past
Every platform ships a "frequently bought together" row. In most categories it earns its place. In wellness it usually does not, and merchants are blunt about why.
- Co-purchase logic recommends what is popular, not what fits. Shoppers read it as a popularity chart.
- Merchants report that accessories and refills work, and that unrelated bestsellers do not.
- Placement matters more than logic: after add-to-cart beats the product page for many stores.
- Stacking prompts in one session is the fastest way to get all of them ignored.
“generic cross-sell widgets often just get scrolled past”
“Accessories and refills worked better than unrelated best sellers.”
“Put the recommendation after add-to-cart, not on the product page.”
“Most upsell fatigue we’ve seen comes from stacking three prompts in one session”
“relevance is more important than volume”
All five are merchants in public Shopify Community threads, August 2026. None of them is talking about a vendor’s features. They are talking about whether the shopper noticed.
- 1Why this, for me? Something about this shopper, not about your catalogue.
- 2Why now? "Works with what is already in your basket" is a reason. "10% off today" is a deadline.
- 3What should change if it works? Naming the measurement that should move turns a purchase into something they can check.
- Put a reason beside every recommendation
- Start with accessories, refills and consumables
- Test one placement at a time
- Make the prompt easy to dismiss
- Rank by margin — shoppers work it out
- Stack three prompts in one session
- Add a countdown to a recommendation
- Train people to wait for the bundle discount
The merchant bar is simple and quoted often: uninstall anything you cannot name a measured result for. The problem is that most stores cannot run a clean A/B test, because they do not have the traffic to reach significance in a sensible time. One vendor says so openly to merchants in public, which is more honest than most.
What works at small volume:
- A holdout. Turn the surface off for a share of sessions and compare. Slower, but it survives a seasonality argument.
- Measure the chain, not the total. Shown → connected → shortlist returned → added to basket. The weak step tells you what to fix.
- Say what you cannot prove. At twenty thousand sessions a month, a two per cent difference is not a finding.
Humotron holds no order permissions, so it reports what it can see rather than claiming credit for revenue it cannot verify. Order-level attribution belongs in your own analytics.
Does this replace our existing recommendation row?
It does not have to. Co-purchase logic is good for accessories and consumables. Use something else where the right answer depends on the person.
Can we exclude products?
Yes, as a step in setup. Anything you exclude never appears. Prescription lines, clearance and gifting are the usual ones.
Can we push higher-margin products up?
No. Position is decided by fit against the shopper’s own signal. Better product data is the legitimate route to appearing more often.
Where should the prompt go?
It is an app block, so you place it. Put it on an unpublished theme and walk the flow yourself before publishing.