Resources · 53
Search and facets: make a catalog usable
Connect customer language, reliable attributes and filter states to clear results and controlled URLs.
· 4 min
What this guide helps achieve
- Qualify attributes
- Test actual intents
- Preserve filter state
- Limit useless combinations
Quick check
- Are units and values coherent?
- Does a missing attribute exclude a product?
- Can one filter be removed?
- Does back navigation preserve state?
- Which facets deserve discoverable pages?
Step-by-step method
- 01
Collect intents
Build queries covering exact identifiers, needs, synonyms and typing errors. Have an offer specialist review expected results. Do not measure clicks alone: ambiguous labels can generate clicks without relevance.
Deliverable: queries and expected relevance.
- 02
Make attributes reliable
Normalise units, vocabulary, variants and missing values. Separate unknown, inapplicable and unavailable. Reconcile some results with authoritative catalog data before optimising ranking.
Deliverable: dictionary and data checks.
- 03
Make filters clear
Show active filters, result count and a way to remove each choice. Test keyboard, zoom and announcements after changes. Autocomplete, when used, needs a coherent accessible interaction model.
Deliverable: state and interaction acceptance.
- 04
Handle empty results
Explain which criteria produced no results and offer filter removal or reformulation. Do not present unavailable products as exact answers. Separate an empty catalog, an empty query result and search-service failure.
Deliverable: empty and unavailable states.
- 05
Decide crawlable URLs
Facet combinations can multiply URLs. Select pages serving an independent need and apply a consistent crawling strategy to others. Google specifies combination and no-result rules for facet URLs intended for crawling.
Deliverable: facet policy and URL tests.
- 06
Replay with current catalog
After imports or stock changes, rerun queries. Check variant availability through the cart, shared filters and back navigation. Classify failures as data, index, relevance or interface issues.
Deliverable: catalog regression report.
Four situations for testing the method
Fictional teaching cases. Each situation connects a signal, a check and a decision bounded by evidence; none represents a client outcome.
| Fictional situation | Concrete check | Decision and limit |
|---|---|---|
| Evaluate filtering before choosing a platform — Three store demos use short catalogues, while the project needs many sizes, brands and filter combinations. | Replay the same selection task on representative catalogue data. Examine generated URLs, backtracking and keyboard access. | Compare results and operating conditions. A CMS name does not establish the quality of a catalogue implementation. |
| Preserve selection context during a store redesign — An old URL represents a useful product selection, but the new site opens a general list without that context. | Inventory selections serving real demand, check new equivalents and test availability, filters and arrival links. | Retain or explicitly replace useful selections. Do not create an indexable page for every theoretical parameter combination. |
| A complementary recommendation must remain compatible — Cross-selling suggests an accessory unavailable for the selected variant because it uses only the product family. | Test variants and compatibility constraints. Check that filtering, product page and basket use the same identifiers and availability. | Remove unverified associations. Measure useful additions and compatibility errors rather than recommendation clicks alone. |
| Test returning from a product page — Someone filters a list, opens a product page and returns to find the selection and position lost. | Replay with multiple filters on mobile and keyboard. Check URL, visible state, result count and empty-list communication. | Preserve useful context and provide an understandable route out of no-match results. The presence of a filter does not establish usability. |
Reusable worksheet
Complete with your authorised observations. These fields are a working template, not observed results.
| Field | Information to record |
|---|---|
| Query | Intent, language and expected products |
| Filters | Attribute, unit, values and omissions |
| Result | Relevance, availability and empty state |
| URL | Intended crawling, canonical and back navigation |
Worked example
Illustrative situation
Fictional example: a size filter excludes a product whose value was imported in a different format.
Decision and expected evidence
The check fixes source vocabulary and rebuilds the index; hiding the filter would only conceal the defect.
Distinguish the mechanisms
| Mechanism | Purpose | Check or limitation |
|---|---|---|
| Search | Express free-form intent | Test synonyms and empty states |
| Facet | Narrow by attribute | Missing or incoherent values |
| Editorial category | Serve a stable need | Distinct content and relevant selection |
Management indicators
| Indicator | What it measures | First action |
|---|---|---|
| Zero-result queries | Valid queries without answers | Separate absent offers from broken indexing |
| Top-result relevance | Useful answers in the query set | Fix data before ranking |
| Crawled combinations | Facet URLs visited by crawlers | Review those without value |
Common pitfalls
- Creating a page for every filter
- Hiding active filters
- Confusing absence with failure
- Ignoring missing attributes
Frequently asked questions
Should every facet be indexed?
No. Choose those serving a distinct need with useful content; others require an appropriate crawling strategy.
Should an empty search return 404?
Internal search and crawlable facet URLs differ. Status follows the route contract; Google recommends 404 for crawlable no-result combinations.
How can a small team test this?
Start with important intents, exact references and edge cases, using reviewed expectations and dated data.
Robots.txt: crawling, indexing and evidence
A parameter variant can change the applicable rule
In teaching file B, /edition/brief is allowed by an exact rule but /edition/brief?view=print remains disallowed by /edition/. The $ suffix ends the match.
- Build URLs with and without parameters in the orders actually produced by filters.
- Separate crawl and indexing goals; noindex hidden by Disallow can remain unseen.
- Retain useful filter state for visitors and a consistent canonical strategy for variants.
Review the twelve robots.txt cases →
References reviewed
Official references
References consulted: . The method and worksheet propose checks to adapt to your context; they do not constitute certification.
