On-Site Search & Merchandising: The 2026 Conversion Playbook

On-site search is the highest-ROI surface most ecommerce teams underinvest in. According to Constructor's "Beyond Relevance" study of 609 million searches across 113 retail sites, shoppers who use search make up roughly 24% of visitors yet generate about 44% of total revenue and 45% of add-to-cart activity. That is not a rounding error — it is the single biggest concentration of buying intent on your site.

The reason is intent. A shopper who types into the search bar has already decided, more or less, what they want. They self-select as high-purchase-intent users, which is why they convert at roughly 2.5 times the rate of non-searchers in Constructor's data — a gap independently echoed by an Econsultancy benchmark of 4.63% searcher conversion against a 2.77% site average. The catch: when search fails, that intent evaporates. Algolia reports that around 81% of US shoppers abandon a site after an unsuccessful search.

This guide covers what to measure, the merchandising rules that move revenue, the UX patterns that recover failed searches, how AI and vector search change the math, and the 2026 build-vs-buy decision now that the mid-market has consolidated. Every figure below is attributed to its source; the vendor-reported numbers are flagged as such, because the difference between a directional benchmark and your own measured baseline is the whole game.

Start with the structural fact that reframes everything else. Constructor's "Beyond Relevance" study — built on 609 million searches and roughly $9.8 billion in revenue across 113 global retail sites between October and December 2024 — found that search users are about 24% of visitors but generate around 44% of revenue. That is Constructor platform data, not an independent audit, so treat the exact figures as directionally reliable rather than settled. But the direction is unambiguous and consistent across the industry.

The mechanism is self-selection. A visitor who types a query has crossed from browsing into shopping; they are telling you what they intend to buy. That is the "intent premium" — searchers convert better than browsers almost regardless of how good your search is, because they arrive at the search bar already further down the funnel. The 2.5x conversion gap Constructor reports, and the 4.63%-vs-2.77% split from a circa 2022–2024 Econsultancy benchmark, are both reflections of that premium rather than proof that any one search engine is magic.

Here is the original analysis that matters for budget decisions: the intent premium means even a mediocre search experience converts better than browsing, which is exactly why search gets underinvested. It "works" well enough that nobody escalates it. But the revenue does not live in the floor — it lives in the delta between mediocre and excellent. A site already capturing 44% of revenue from search has proven the surface; the open question is how much of the leaked intent (the abandoned, zero-result, mis-ranked queries) it could recover. Constructor's sector data sharpens the point: in Health & Beauty, searchers convert at about 17% versus roughly 6% for non-searchers, and in General Merchandise, 41% of traffic from searchers translates into 61% of revenue.

The asymmetry is the whole story. Across the blended dataset, searchers punch nearly twice their weight: 24% of traffic to 44% of revenue is a revenue concentration of about 1.8x per visitor, and in General Merchandise the searcher share of revenue runs higher still — 61% of revenue from 41% of traffic. That ratio is what justifies treating search not as a feature you ship once but as a CRO surface you tune continuously — the same way you would never set a checkout flow live and walk away.

Most coverage treats on-site search as a relevance problem: did the engine return the right products? Constructor's study makes a sharper, less-cited point. Beyond a baseline of relevance, what moves the needle is result attractiveness — the visual design, price, availability, ratings, and personalisation of the results that come back. Per the study, each one-point increase in result attractiveness is associated with roughly a 4% rise in click-through rate. That is a merchandising lever, not an algorithm tweak.

The implication for how you staff and budget search is significant. If relevance is solved by your platform but your search results grid shows out-of-stock items, weak imagery, or no social proof, you are leaving the larger lever untouched. Merchandising the search results page — pinning hero products, demoting low-margin or low-availability SKUs, surfacing ratings and badges — is where teams with a competent engine still find double-digit gains.

This reframing also explains why a results page is a merchandising canvas, not a list. The same intent that brought a shopper to the search bar can be steered: a query for "running shoes" can lead with your best-converting, best-margin, in-stock models rather than a relevance-only sort that buries them. The discipline is the same one we apply in our ecommerce optimisation work — treat the search results page as the most valuable category page you own, because the shopper told you exactly what category they are in.

A zero-results page is a dead end at the moment of highest intent. Hello Retail's 2026 search statistics put the industry-average zero-results rate at 10–15%, with a best-practice target under 5%. The stakes are steep: Algolia reports that roughly 81% of US shoppers (and 80% globally) abandon a site after an unsuccessful search, and that about 82% avoid a site where they previously had a bad search experience — both figures are vendor-reported and widely cited, so treat them as directional rather than audited.

The cheapest wins are linguistic. Algolia's best-practice guidance notes that 20–30% of ecommerce search queries contain a misspelling or alternate phrasing. Typo tolerance, synonym dictionaries, and autocomplete are the first three fixes — and they are largely configuration, not engineering. Practitioner reports suggest typo and synonym handling alone can lift conversions in the high single digits to mid-teens, though that range is not independently quantified in a single study, so measure it on your own catalogue rather than banking a specific number.

The harder class is the use-case query. According to a Zoovu study cited by Algolia, around 63% of zero-results queries are driven by subjective, use-case-driven searches — "bike for commuting," "dress for a beach wedding" — that keyword engines misclassify because the words on the page do not match the words in the shopper's head. That attribution is cross-vendor (Zoovu via Algolia), so cite it as such. It is also the single strongest argument for semantic search, which we cover in section 05.

One pattern ties the recovery work to the rest of your funnel: a shopper who hits a zero-results page and a shopper who abandons a cart are the same lost-intent problem at two different stages. The fixes rhyme — surface alternatives, reduce friction, follow up. If you are already investing in cart abandonment recovery, a disciplined zero-results strategy is the upstream version of the same playbook, and usually cheaper to fix.

Vendors will happily show you their own dashboards. What they do not publish is a platform-agnostic scorecard you can use to evaluate any engine — or to hold your current one accountable. The table below is ours, assembled from Hello Retail's 2026 benchmarks, Algolia's stat compilation, and Constructor's sector data. The benchmark and threshold columns restate each row's own published band; the instrumentation column is how you wire the metric up regardless of platform.

Two of these deserve a closer look. Exit-after-search rate is the counterpart to zero-results: a query can return products and still fail if none of them are attractive enough to click — Hello Retail flags an exit-after-search rate above 25% as a relevance problem worth investigating. And p95 query latency connects search to the oldest performance benchmark in ecommerce: Amazon's much-cited finding, dating to a 2006 internal study, that every additional 100ms of latency measurably reduced sales. It is a two-decade-old figure and should be framed as a historical benchmark, not current research — but the principle still anchors the case for sub-100ms search to a finance team.

The scorecard also forces a funnel view. Search is the top of a chain that runs search → product page → cart → checkout, and each stage has its own leak. Pairing this search scorecard with a disciplined checkout optimisation guide gives you an end-to-end conversion baseline rather than a single-stage snapshot — which is the only way to know whether a search change actually carried through to revenue.

Keyword search matches strings. Vector (semantic) search matches meaning: it embeds both the query and your products into the same mathematical space and returns the nearest neighbours, so "something warm for winter hiking" can surface insulated jackets that share no literal keywords with the query. This is the structural fix for the use-case-query problem behind a large share of zero-results, and it is why search is now the top digital-investment priority for many retailers.

The investment signal is loud. In Algolia's 6th Annual eCommerce Search Report (November 2025), 49% of B2C retailers said they already use third-party search solutions, 42% planned to increase search spending in 2026, and 61% planned to implement agentic AI in search within twelve months. These are figures from a vendor-published survey, so weigh them as a direction of travel rather than an independent census — but the direction is consistent with the consolidation and product launches across the market.

On outcomes, be disciplined. Vendor case studies report real but self-reported gains: Algolia documents Decathlon achieving about a 50% conversion lift on personalised search queries and an early NeuralSearch beta cohort seeing roughly a 17% uplift in search-driven conversion within weeks — explicitly a small, early-beta sample. Beauty brand Tatcha is widely cited at 3x conversion and 38% higher average order value, but that is a vendor case study (via Alhena.ai) attributed to AI-powered discovery broadly, not search alone. Use these as evidence that the lever exists, not as a number you can promise. The honest version: AI and vector search can raise conversion meaningfully versus keyword-only matching, and you should benchmark the exact lift on your own catalogue before claiming a multiple.

Where does this go next? The forward projection worth making: the 2026 frontier is not better ranking but agentic discovery — search that holds a conversation, asks clarifying questions, and assembles a shortlist rather than returning a grid. Bloomreach's November 2025 launch of Personalized Media in-Grid, which turns static product listing pages into individually personalised storytelling surfaces, is an early signal of the same shift: the line between search, merchandising, and recommendations is dissolving into a single discovery layer. Teams that instrument their search analytics now will be the ones able to evaluate those agents on evidence rather than vendor decks. This is the same convergence we explore in our piece on agentic merchandising.

Here is the angle almost nobody publishes as a workflow: your search logs are the cheapest product-gap research you will ever run. Every zero-results query is a customer telling your buying and merchandising teams what they wanted and could not find. Most organisations stop at conversion and never route that signal anywhere. Turning the weekly zero-results report into an assortment and campaign input is a standing operational habit, not a project.

The operational cadence is simple and rarely done: pull the top zero- and low-results queries weekly, tag each into one of the four buckets above, and route them to the owning team with the query volume attached as a priority signal. Over a quarter, that single habit compounds — the catalogue-gap fixes recover revenue you already have inventory for, and the assortment-gap data gives your buyers a demand-ranked shopping list grounded in real customer intent rather than gut feel.

The build-vs-buy question used to be binary. It is not anymore. In January 2025, Klevu and Searchspring merged to form Athos Commerce (backed by PSG), collapsing two mid-market leaders into one entity. That consolidation, plus the maturing of open-source vector search, turned the decision into a three-way fork: enterprise discovery suites, mid-market SaaS, or open-source self-build. The 2025 Gartner Magic Quadrant for Search and Product Discovery named four Leaders — Algolia, Constructor, Coveo, and Google — per vendor press releases (the full report is paywalled, so read this as "named a Leader," not a ranking).

The matrix below is ours, built from vendor pricing pages, AWS Marketplace and competitor-comparison estimates, and the June 2025 Gartner press releases. Pricing is indicative and dated 2025 — search pricing is opaque and quote-driven at the top end, so verify live before you budget. Read each row, weigh it against your catalogue size and team depth, and the column you land in most often is your starting hypothesis.

You do not need a platform migration to start. The fastest gains come from instrumentation and configuration on whatever engine you already run. Sequence the work so each phase funds the next: measure first, fix the cheap leaks, then earn the case for a platform decision with data instead of a vendor pitch.

The reason this sequence works is that it inverts the usual order. Most teams buy a platform and hope for a lift; this playbook measures the leak, fixes what is free, and only then spends — by which point you can quantify exactly what a new engine has to beat. If you want a second set of senior eyes on the audit or the build-vs-buy decision, our AI transformation engagements start with precisely this kind of measured, evidence-first evaluation.