Find in document: keep matches tied to place

Show which match the arrows will visit

Craft presents an active query with a 3/4 result indicator, previous and next controls and View All Results. NaturalReader shows a query with a 1/1 indicator and navigation arrows while the keyboard remains visible. A compact result navigator can preserve reading space, but only if the person can understand the selected occurrence and the next action.

Give the current match a treatment distinct from other matches and from unrelated document highlights. In a reading app, a colored passage could also represent speech playback or an annotation. Avoid using the same color alone for every role. Navigating to a result should reveal enough surrounding text to confirm it is the intended passage, not align the word underneath the keyboard or a persistent toolbar.

Define what happens at the first and last occurrence. Wrapping to the beginning can work if the interface announces it, while disabling an arrow can work if its state is understandable. Keep the query intact during navigation and do not move focus back into the search field after every result. Reading a found passage should not require repeated keyboard dismissal.

Make result lists explain where a match lives

Documents groups its query results with page references and matching snippets, with Text and Thumbnails modes available. PDF Editor presents thumbnail-backed snippets and a total of 13 matches. These views help when a document contains several similar occurrences and stepping through them one by one would hide the broader structure.

Make the row identify enough context to choose deliberately: the page or section, the matching passage and its relationship to the document. Avoid showing only repeated copies of the query as row titles. If several occurrences belong to one page, distinguish the page group from the number of matches inside it. A collapsed group should not make those two quantities look interchangeable.

Preserve list position when a person opens a result and returns. Use a stable result identity tied to the document revision, not an index that can point somewhere else after recalculation. For long snippets, support readable wrapping and large text without cutting off the only words that distinguish neighboring results. The list exists to reduce uncertainty, not simply to add another navigation layer.

Keep matching options close to their consequences

WPS exposes Match Case and Whole Words as distinct options beneath the query. PDF Expert offers Text and Thumbnails and displays different result-context icons beside matching snippets. These examples show that search behavior and result presentation can both vary; the interface should not collapse every variation into an unexplained settings symbol.

Show active matching restrictions where they can explain an unexpected result count. A person searching for a short word may get no results because whole-word matching is enabled, not because the term is absent. Explain supported scope precisely, particularly when comments, notes, handwriting or extracted text may be treated differently. Do not imply that every kind of visible content is searchable unless the implementation supports it.

Recompute results when a matching option changes and retire stale counts visibly. Keep the query so the person can compare behavior rather than retyping it. Use language appropriate to the audience: a technical option can remain available without requiring every reader to understand pattern matching. Unsupported syntax should not quietly produce a reassuring zero-results state that looks like an authoritative absence.

Label the quantity instead of displaying bare fractions

PDF Hero displays more than one fraction in the viewer while a query is active. MyScript Notes also combines a search-related count with a page count in the surrounding interface. The stills show the risk of competing numbers; they do not establish that either app calculates results incorrectly.

Prefer labels such as Match 2 of 8 and Page 3 of 12 when both quantities are relevant. A naked 1/2 near one toolbar and 1/4 near another asks the reader to infer the denominator’s meaning. If one number refers to selected pages or a loading range instead, name that scope rather than styling every fraction as if it belonged to the search task.

Handle partial result calculation explicitly. Searching a large document can produce results incrementally, so an early count may not be final. If the app cannot know the total yet, say that results are still being found instead of presenting a provisional denominator as complete. Announce meaningful changes accessibly without reading every intermediate count while the person types.

Do not mistake viewer context for search state

The Claw AI transcript viewer displays a file identity, page context and a search entry point, but no active query in this recorded state. MobiPDF shows an active query alongside a page carousel and navigation controls. Comparing them helps separate entering find mode from viewing results after a query has been submitted.

Keep a document’s page indicator meaningful before, during and after search. Opening find should not repurpose the same unlabeled number from page navigation into a match counter. If the app uses a different result view, make the transition legible and preserve the original reading anchor. Closing find should have a defined destination rather than jumping to the first page simply because the search panel disappeared.

Offer a clear response when text is unavailable for search. An image-only scan, a document still loading and a completed query with no matches are different states. Where OCR is available, describe it as a separate supported operation with its own processing and accuracy limits. Do not manufacture matches from a preview or imply that a missing text layer means the visible words do not exist.

Tie results to text and revision identity

Represent a result as a document revision, searchable-content scope, stable range or anchor and display context. Keep query options in the search request identity. A late response for an older query must not replace the result list for a newer one, and a changed document should not retain ranges that now point at unrelated text.

Plan normalization deliberately for the actual languages and formats supported. Test capitalization, punctuation, line breaks, repeated words and composed characters rather than assuming a simple substring operation matches reader expectations. A highlighted visual range needs to agree with the selected textual occurrence. If layout changes after rotation or text resizing, resolve the same semantic anchor instead of reusing an obsolete screen rectangle.

Separate content search from navigation focus. Arrow controls need accessible names, the active occurrence needs an understandable announcement, and leaving find should not strand focus in a removed field. Keep a useful reading position when the keyboard closes. Avoid logging document passages or full queries by default when aggregate result and navigation events can evaluate the task without exposing private content.

Test the passage someone actually needs

Ask a tester to find the second occurrence of a phrase in a long document, read the following paragraph and then return to the results. Repeat with a query that spans a line break and a word that appears in several contexts. Check whether the person can explain the current match and page without relying on a designer’s knowledge of the toolbar.

Include no matches, image-only pages, pending extraction, a document edit during search and rapidly replaced queries. Test large text, a hardware keyboard, screen-reader navigation and rotation with the software keyboard open. Confirm that exiting find restores a sensible reading anchor and does not silently discard a deliberate navigation choice.

Measure successful passage retrieval, wrong-result openings, repeated query reformulation and lost-position recovery. Do not treat the first matching highlight as task completion. A useful find experience helps someone use the passage they located; the number of search-field taps or a short query session cannot establish that outcome by itself.

Review query, matches and place

Use these checks on a working build, not only on the polished reference state.

• ✓Match counts and page counts are labelled separately.
• ✓The active occurrence differs from other highlights.
• ✓Matching options remain visible when they explain results.
• ✓Late queries cannot overwrite newer result state.
• ✓Unsearchable content is not called a completed no-match search.
• ✓Keyboard, rotation and large text preserve the selected passage.
• ✓Closing find returns to an understandable reading location.

Related design decisions

The app search examples cover discovery across a collection. Keep that scope distinct from finding a passage inside the currently open document.

Use the AI note-taking examples when a found passage needs to retain its relationship to a source transcript or note rather than becoming an isolated snippet.