'%20clip-path='url(%23canvas-clip)'/%3e%3cg%20clip-path='url(%23canvas-clip)'%3e%3cg%20transform='rotate(0,%20256,%20256)'%3e%3cg%20transform='translate(87%2087)%20scale(14.083333333333334%2014.083333333333334)%20translate(0%200)'%3e%3cpath%20fill='none'%20stroke='%23FFFFFF'%20stroke-linecap='round'%20stroke-linejoin='round'%20stroke-width='2'%20d='m18.064%2010.877l-4.89-7.26c-.42-.625-1.287-.803-1.936-.397a1.4%201.4%200%200%200-.41.397l-4.893%207.26C4.24%2013.715%204.9%2017.318%207.502%2019.423a7.13%207.13%200%200%200%204.168%201.572M15%2018a3%203%200%201%200%206%200a3%203%200%201%200-6%200m5.2%202.2L22%2022'/%3e%3c/g%3e%3c/g%3e%3c/g%3e%3c/svg%3e)
ParticleSearch Metrics Guide: What Each Search Analytics Signal Means
A search dashboard is useful only when its numbers lead to better merchant decisions. “Products found” can hide irrelevant results. A click can mean interest, confusion, or normal comparison. An empty result can be a catalog failure, a navigation request, or a query no store should satisfy.
This guide defines the ParticleSearch metrics, explains what each one cannot prove, and shows how to move from a pattern to a storefront inspection. For the product workflow and review queue, read the ParticleSearch analytics dashboard guide.
Product contract checked July 28, 2026. Definitions were verified against the current ParticleSearch dashboard, storefront event contract, and reporting builders. Available metrics depend on the installed surfaces and recorded events.
Chapter 1 · The measurement chain
Start with the event behind the metric
ParticleSearch connects a rendered search state to the shopper actions that follow it. That context matters. A product click belongs to the active search journey rather than becoming an unexplained sitewide click.
The chain still records behavior, not motive. The dashboard can show that a shopper refined a query or left after seeing results. It cannot know why without stronger evidence.
Search rendered
A ParticleSearch surface shows the current result set.
Result observed
Products and available filters become part of the measurable search state.
Shopper responds
The shopper opens a product, adds directly, filters, refines, or leaves.
Merchant diagnoses
The dashboard connects the query, products, context, and recorded response.
Chapter 2 · Read the journey in order
A funnel is a sequence of questions, not a single conversion score
The most common analytics mistake is jumping from a headline number to a solution. A query can return products and still fail because the first result is wrong. A direct add can be valuable and still say little about a product that normally requires a detail-page visit. Start at the first stage that changed, then follow the same journey forward.
ParticleSearch keeps these stages separate so a merchant can ask a narrower question: did the store fail to retrieve, fail to present, fail to earn an action, or simply lack enough order evidence? That distinction determines the responsible next move.
Search session
Did a shopper use search during the period?
Sets the denominator for search activity. It is not a count of people or orders.
Product result
Did the query return an eligible product set?
Shows retrieval coverage. It does not say the products were the right answer.
Product action
Did the shopper open or add a product from that search?
Shows a recorded response to the visible result. It is not purchase conversion.
Order evidence
Can a later Shopify order be responsibly connected?
Adds commercial context without turning an observed association into causation.
Chapter 3 · Metric contracts
What each metric means, and what it does not
Use a metric as a question prompt. The definition establishes the measured fact. The next inspection determines whether the pattern deserves a repair.
Search activity
Measured fact
Recorded search executions in the selected period.
Does not prove
Unique shoppers, sessions, or high-intent demand.
Inspect next
Compare the period, distinct query wording, campaigns, releases, and event health.
Products found
Measured fact
Searches where the primary result set contained at least one eligible product.
Does not prove
The result was relevant, visible near the top, or commercially useful.
Inspect next
Inspect the actual products, order, variants, availability, and card information.
Product CTR
Measured fact
The share of observed searches that led to a recorded product click.
Does not prove
Raw clicks divided by impressions, purchase conversion, or satisfaction.
Inspect next
Read it beside result coverage, query intent, click position, and direct adds.
Direct adds
Measured fact
Add-to-cart actions completed from a ParticleSearch result surface.
Does not prove
All add-to-cart activity influenced by search.
Inspect next
Check the query, selected product or variant, surface, and later order evidence.
Zero-result rate
Measured fact
The share of recorded searches whose primary product result set was empty.
Does not prove
Every empty result is a defect or lost sale.
Inspect next
Separate product demand, navigation intent, support questions, noise, and spelling.
Low-result rate
Measured fact
The share of searches that returned only a thin product set.
Does not prove
Thin always means poor. A precise identifier may correctly return one product.
Inspect next
Read the query and determine whether precision or assortment breadth was expected.
No-click rate
Measured fact
Searches with products where no product click or direct add was recorded.
Does not prove
The shopper disliked every result or abandoned the store.
Inspect next
Inspect result order, price, imagery, availability, query type, refinements, and exits.
Average click position
Measured fact
The average rank of products that received a recorded click.
Does not prove
The ideal ranking depth or proof that higher results are better.
Inspect next
Compare exact queries with discovery queries and inspect outliers individually.
Engine latency
Measured fact
Measured search-processing time for the recorded request.
Does not prove
The full delay visible to a shopper.
Inspect next
Also inspect network, theme code, rendering, imagery, and device performance.
Products found + Product CTR
High coverage with low product action means retrieval is not the whole problem. Inspect the first visible products, price, availability, variant clarity, and whether the query is exploratory.
Product CTR + Direct adds
A click with no direct add can be completely healthy for products that require configuration or consideration. A direct add without a product open can be useful for a simple, known item, but verify the selected variant and later order evidence.
Search volume + query shape
A high-volume head term and a lower-volume identifier need different acceptance tests. Do not average them into one score and then apply the same repair.
Search time + visible experience
A fast engine measurement does not prove that the shopper saw a fast result. Check request timing alongside theme rendering, network work, image loading, and device conditions.
Chapter 4 · From signal to diagnosis
Do not optimize the number before locating the problem
Zero results repeat for the same exact identifier
First question
Confirm the product and variant exist, then inspect searchable source fields.
Avoid
Create a broad synonym before proving the identifier is indexed.
Likely path
Catalog or field coverage, normalization, variant identity, then a controlled query test.
Products are found, but actions stay low
First question
Open the result set and judge the first visible products as a shopper would.
Avoid
Treat Products found as evidence that retrieval is healthy.
Likely path
Relevance, ranking, product-card information, availability, or query intent.
Filters are used and empty combinations repeat
First question
Check whether values are compatible at the product and variant level.
Avoid
Remove a useful filter because one catalog combination is malformed.
Likely path
Facet data, count logic, value naming, mobile visibility, or impossible combinations.
A query has clicks but no direct adds
First question
Check whether direct add is appropriate for the product and selected variant.
Avoid
Call the query weak without checking product-page continuation.
Likely path
Product handoff, option selection, product detail quality, price, or normal consideration.
Search exits rise after a storefront change
First question
Verify the change, event health, result rendering, and comparable time window.
Avoid
Assume the relevance engine caused every exit.
Likely path
Interface regression, slower visible response, result quality, seasonality, or instrumentation.
Chapter 5 · Read a change before repairing it
A metric movement is useful when it narrows the next decision
A percentage does not diagnose itself. Read the stage that moved against the stages around it, then run the smallest test that could prove or disprove the likely explanation. This prevents a legitimate catalog or storefront issue from becoming an unnecessary search rule.
If a release or a freshness warning is part of the pattern, use the ParticleSearch catalog health guide before treating the trend as shopper behavior. A clean period comparison starts with a known working search surface.
Search activity is steady, but products found falls
What it narrows
Demand is still reaching search, while fewer recorded searches return an eligible product set. That is a coverage question before it is a relevance question.
Controlled check
Pick a small set of repeated queries that previously returned products. Re-run them on the same surface, then compare product eligibility, variant identity, market availability, and catalog freshness.
Decision boundary
Pause ranking and synonym changes until the product set is trustworthy. Repair catalog data, availability, or indexing evidence first.
Products found is steady, but product CTR falls
What it narrows
Retrieval may still be working, but the first visible results may no longer answer the query well enough to earn a product open.
Controlled check
Compare the first page for a repeated query with the previous expected result. Check product order, title clarity, image, price, availability, variants, filters, and any storefront release.
Decision boundary
Change only the layer the test isolates: source data, result presentation, a focused ranking decision, or the query class itself. Keep a working query as a guardrail.
Product opens hold, but direct adds fall
What it narrows
Shoppers are still willing to inspect the product. The break may be variant selection, price, availability, product-page information, or whether direct add is appropriate for that product.
Controlled check
Follow the exact product handoff from search. Confirm the selected variant, available quantity, required options, visible price, and the expected add-to-cart behavior.
Decision boundary
Do not call this a ranking failure from the dashboard alone. Fix the handoff or product decision surface if that is where the promise breaks.
Activity or actions shift sharply after a release
What it narrows
Timing is evidence, not a cause. A theme, consent, catalog, widget, or tracking change can alter what was rendered or recorded without changing shopper intent.
Controlled check
Run one controlled journey on the affected surface and compare health, event order, consent state, catalog version, and the release boundary before comparing periods.
Decision boundary
Re-establish measurement trust before publishing a search rule or reporting a performance story. Use the health record as the release gate.
Chapter 6 · Diagnose the query class
No-result, thin-result, and no-action queries need different repairs
Treating every weak query as a ranking problem creates noisy rules and hides catalogue defects. The useful question is not “how do I improve this percentage?” It is “what was the shopper trying to do, and which part of the path failed?”
For example, table under may be a missing interpretation or a malformed product field. table may return enough products but show an unhelpful first page. A SKU may correctly return one exact variant. The dashboard can surface these patterns, but the merchant still has to inspect the query and the storefront state.
No products returned
Was the shopper asking for a product, a navigation destination, content, or something the catalogue cannot satisfy?
Start with the exact query. Check product and variant data, searchable fields, spelling, market availability, and whether a redirect or content result is the better response.
A thin set returned
Is one precise result correct, or is the query asking for a broader assortment?
Inspect the result count and product identity together. A SKU should be precise. “Black sofa” should usually be judged against the store’s available assortment and filter model.
Products returned, no action
Did the shopper reject the result, need more information, or simply continue researching?
Review the first page and card content before changing ranking. Compare product opens, direct adds, refinements, exits, and query intent.
Action recorded
Was the action appropriate for the product and did the journey continue?
Protect the query and product as a regression fixture. Use revenue attribution later if you need order-level commercial evidence.
Chapter 7 · Choose the right view
Move from overview to the evidence that can answer the question
Performance
Establish the period, denominators, broad movement, and operational context.
Queries
Find repeated wording patterns and inspect the products behind each query.
Products
Compare search exposure with product opens and direct adds.
Filters
See which narrowing paths are used and where combinations become thin or empty.
Behaviour
Follow recorded search, refinement, product, and exit sequences without inventing intent.
Audit
Verify event-level evidence when a metric looks surprising or instrumentation is in question.
Use the Audit view when the measurement itself is the question. If a metric changes unexpectedly, verify duplication, consent state, surface coverage, timing, and event order before explaining the change as shopper behavior.
Chapter 8 · A weekly operating rhythm
One well-inspected pattern is more useful than a long export
Check trust first
Confirm catalog freshness, storefront delivery, reporting period, and recent releases.
Choose one repeated pattern
Prefer a query or product issue with recurring evidence and a falsifiable question.
Inspect the live experience
Review result identity, order, variants, card information, filters, and destination.
Make the smallest responsible change
Repair source data, adjust a focused control, change the interface, or document no change.
Measure a comparable period
Wait for enough new evidence and verify that another important query did not regress.
If the diagnosis points to a merchant control, use the ranking, synonyms, redirects, and safe publishing guide. If the question is commercial value after search, continue with the ParticleSearch revenue attribution guide.
The goal is not to make every metric rise. It is to understand the shopper job, preserve searches that already work, and improve a specific failure without creating a broader one.