Zero-Results Pages: Turning a Dead End Into a Recovery
The highest-intent moment to lose someone
A shopper searched, got nothing, and most of them leave. A zero results search page is one of the highest-intent moments in the whole session, and a blank "no results found" throws it away.
Why it still happens with semantic search
Semantic search cuts the zero-results rate; it doesn't eliminate it. The remaining causes:
- A product you genuinely don't stock.
- A misspelling the model can't bridge.
- An over-tight filter combination.
- A brand-new SKU not yet in the index.
- A query for something that isn't a product ("delivery times").
What to show instead of a blank page
- Relaxed results. Drop the least important term or the tightest filter and show those, clearly labelled: "No exact matches — showing similar boots."
- "Did you mean" for genuine likely misspellings.
- Nearest categories or approximate matches, even if below the relevance floor, labelled as approximate.
- A back-in-stock notification offer, if it's a real product that's out.
- A visible path to search help or a human.
The one thing to always do
Capture the zero-result query, with the filters applied, into the analytics the merchandising team reviews. These queries are a direct list of demand you're not meeting — sometimes a stocking decision, sometimes a synonym to add, sometimes a content gap. On our semantic search project the merchandising console surfaces exactly this feed, and it's one of the most acted-on reports the team has.
Query relaxation, step by step
Relax in a fixed order and always label what you did:
- Drop the tightest active filter first — a price band, then a secondary attribute. Re-run. If there are results: "No exact matches — showing boots without the price filter."
- Only then drop query terms, least important first — adjectives before nouns. "warm waterproof boots" becomes "waterproof boots", then "boots".
- Never silently. The shopper needs to know these are approximate, or they'll judge your search on results you didn't claim were exact.
The back-in-stock capture
For a real product that's genuinely out of stock, the zero-result page is the best place in the whole site to capture a notify-me. The intent is unambiguous — they searched for exactly this — and it converts a dead end into a re-engagement you control. Wire the "did you mean this out-of-stock product?" path to a one-tap notify option.
Measure the recovery
Zero-results rate as the headline, and the recovery rate on the zero-results page — do the relaxed results actually get clicked? A page full of "similar" results that nobody engages with isn't a recovery, it's a slower dead end.
The metric that tells you the recovery works
Zero-results rate is the headline number, but the one that tells you the recovery page is doing its job is the click-through rate on the relaxed results. If shoppers see "showing similar boots" and don't click, the relaxation is either too loose (the results genuinely aren't relevant) or not labelled clearly enough (they don't trust what they're looking at). A recovered session isn't one where you showed something — it's one where the shopper engaged with it.
Where this stops being right
- Query relaxation that's too aggressive shows irrelevant results and trains shoppers to distrust the search — label it clearly and keep it conservative.
- "Did you mean" on a term that's actually correct is annoying — only offer it on genuine likely misspellings.
- A tiny catalogue will have more legitimate zero-results — the recovery UX matters more there, not less.
FAQ
Why do I still get zero-results with semantic search? Absent products, things you don't stock, misspellings, over-tight filters, unindexed new SKUs. Semantic search reduces the rate, it doesn't remove it.
What should the page show instead of nothing? Relaxed results labelled as approximate, a "did you mean," nearest categories, and a back-in-stock option for real out-of-stock items.
What do I do with the failed queries? Capture them into the merchandising analytics — they're a list of unmet demand: stocking gaps, missing synonyms, content holes.
Should a zero-result page ever show a human contact option? For a high-value or complex catalogue, yes — a shopper who searched precisely and found nothing is a good candidate for assisted selling. For a simple catalogue it's usually enough to relax the query and capture the term.
ISTRALLEN builds search where a zero-result page recovers the session and feeds the merchandising team; see AI for Retail.