1. Define what visitors can search
Our fictional library contains guides and practical fact sheets. Start with an inventory of public content to retrieve, including its title, language and address. Decide whether search covers the whole library or only the current section. Show that scope near the field: “Search resources” avoids making an overly broad promise.
Select a few concrete needs, such as finding the library opening-hours guide. For each need, record a query and the relevant expected page without imposing an arbitrary ranking. Check that the page exists in the tested version. Missing content first needs its editorial owner’s attention; changing the search engine does not create it.
2. Prepare a clear field and action
Review the field label, any example and the search button together. The WAI tutorial calls for a label describing the control’s purpose and correctly associated with it. In our example the label remains available during typing; disappearing text should not be the only explanation of the search scope.
MDN explains that an HTML search input accepts a search query; it does not itself provide the mechanism that retrieves content. Do not infer suggestions, spelling correction or searching inside PDFs from the field’s appearance. Confirm the features actually planned and their behaviour separately.
Source : W3C WAI — Labeling Controls · Source : MDN — input type search
3. Make each result identifiable before opening it
Prepare a distinct title and a useful short excerpt instead of a series of vague names. For our opening-hours guide, the excerpt describes its coverage without promising real-time availability. If several editions coexist, distinguish them with an accurate identifier. Test the destination: an attractive card opening the wrong sheet is still a failure.
The W3C page-title criterion requires a title describing the topic or purpose. Check continuity between the selected result and the page opened, then the return to the list. If the query and filters should be retained, record that expectation in acceptance criteria and observe it; the Back button alone does not guarantee retention.
4. Distinguish empty queries, no matches and failures
Write three separate expectations: no term entered, a successful search with no match, and a search that could not complete. For “No results for this search”, offer a useful action such as changing the words or removing an active filter. Do not present unavailability as evidence that content does not exist. In preview, request a controlled failure case if the team can provide it safely.
The W3C status-message explanation says that a brief search message updated without focus moving or a page refresh needs to be available for assistive-technology announcement when it is a status message. The result list itself is not that message. Have the announcement checked for the delivered behaviour; this check alone does not certify accessibility.
5. Repeat realistic queries in each language
Build a small query set from the content rather than using only a word already known to work. Try a known title, a common term, several words and an absent term. Add an accented variant or a typo only if handling it is part of the agreed need. Record expected and observed results; do not assume search automatically translates the entered words.
Repeat with a keyboard and on a phone. Open a sheet, return, change the query and then clear it. If filters or several result pages exist, combine them with search. In Arabic, read the entered words, numbers and any foreign titles in their actual reading direction. Use public, fictional queries without personal data.
6. Keep useful evidence for every mismatch
Keep the query, language, filters, tested version, device and starting and destination addresses. Record the steps and observed result, including the state after returning and clearing. A screenshot of a list does not prove that its link reaches the right document. Tie each mismatch to a precise check before requesting a correction.
Distinguish a missing sheet, unclear wording and a search defect. After correction, repeat the same query and a nearby case. Accept the agreed scope while keeping unverified limits visible. This acceptance exercise does not measure conversions or demonstrate a SEO ranking gain; internal search and a public search engine’s results are different journeys.
Internal-search acceptance worksheet
| Case | Expectation to agree | Evidence to keep |
|---|---|---|
| Known title | Correct sheet reachable | Query, result and destination |
| No term | Explicit behaviour | Initial state and action |
| No match | Message and useful next step | Absent term and proposed action |
| Unavailability | Failure distinct from no matches | Controlled case and message |
| Return and clear | State matches the decision | Steps and final result |
| Language and mobile | Usable reading and controls | Language, device and observations |
Frequently asked questions
Should we add search suggestions?
Only if they serve a defined need. They add a journey to prepare and check; a visible field does not prove that suggestions exist.
Does a query with no matches always indicate a problem?
No. Compare it with the agreed content and scope. An absent term can legitimately return nothing, provided the message and next step are understandable.
Should search find other languages?
Decide explicitly. A search limited to the current language and a multilingual search need different expectations, which should be announced and tested.
Prepare your next acceptance check
Choose a public section, inventory the content to retrieve and agree a few reference queries. Have the expectations approved before checking the preview in each published language.
Sources and scope
Primary W3C and MDN sources checked on November 7, 2026. The library and its wording are fictional. These steps are an editorial acceptance proposal, with no claim of tests on your website or new Dev Ring capabilities.
