Recommendations on zero results pages: recovering the lost search
October 21, 2026
Short answer: The shopper who searches has done the store a favour: they stated exactly what they want. When the answer is a blank page and the words no results, the store has wasted the clearest intent signal it will get all day. Zero results pages deserve recommendation logic of their own, because the recovery only needs to get the shopper one step closer, not all the way to the exact product.
Diagnose why the search failed
Zero results have different causes and different fixes. A misspelling or a synonym gap means the store stocks the product under another name, and the recovery is a corrected suggestion: did you mean, with results attached. A genuinely absent product means the recovery is adjacency: the nearest category, the closest alternative, or the range the shopper most likely settles for.
A third cause is filter stacking, where the product exists but the shopper's combined filters exclude everything. The recovery there is relaxation: show results with the least important filter removed, and say plainly which filter was dropped. Each cause needs its own page behaviour, which is why a single generic zero results template underperforms.
What to recommend when there is nothing to recommend
Start with the query's intent, not the store's promotions. Token level matching can find products sharing part of the query; category inference can offer the department the query belongs to; and behavioural data shows what other shoppers who searched this term eventually bought, which is often the best recovery shelf available.
Popularity has a place here as a safety net, clearly framed as popular right now rather than pretending it answers the search. Shoppers forgive an honest dead end with helpful signposts. They do not forgive a page that insists unrelated best sellers are their results.
Capture the demand you cannot serve
Every zero results query is unmet demand written in the customer's words. Logging the terms, with counts, gives buying and content teams a list money cannot usually buy: products to range, names to add as synonyms, and guides to write. Stores that review this log monthly watch their zero results rate fall steadily.
Where the store plausibly will stock the item, a notify me option turns the dead end into a future sale and a measured demand signal. Where it will not, an honest pointer to the closest alternative category still beats a blank page.
Test the recovery like a landing page
Zero results pages get real traffic, often two to five percent of all searches, so they repay testing. Measure recovery rate, the share of zero results sessions that go on to view a product, and treat it as the page's headline metric. Small copy and shelf order changes move it noticeably.
Finally, keep prevention ahead of cure. Every zero results term fixed through synonyms, better titles, or new ranging removes a future dead end entirely. The best zero results page is the one fewer shoppers ever see.
Keep reading: Cold start recommendations: what to show shoppers you know nothing about, How should recommendation ranking weigh margin and stock levels?, and What belongs in checkout page recommendations, and what should stay off them? cover the nearby cases in this series.
Frequently asked questions
What should a zero results page show first?
A diagnosis based recovery. For a likely misspelling or synonym gap, the corrected search with results. For an absent product, the nearest category and closest alternatives. A generic list of best sellers should be the last element, framed honestly as popular items.
Why do shoppers abandon after zero results?
Because the page signals the store cannot help, at the exact moment their intent was highest. Shoppers rarely reformulate more than once. A recovery shelf that gets them one step closer keeps the session, and often the sale, alive.
How does behavioural data improve zero results pages?
It reveals what previous shoppers who typed the same term eventually viewed or bought. That observed path is usually a better recovery recommendation than anything guessed from the query text alone.
Should zero results queries be reported to buying teams?
Yes, monthly at minimum. The log is direct evidence of demand the store failed to serve, split between fixable naming gaps and genuine range gaps. Both are actionable, and tracking them steadily reduces the zero results rate itself.