Keyword highlighting shows where the match occurred

What it is

Bolding or coloring query words inside a snippet—keyword highlighting / KWIC—points at which characters produced the match. People need not finish the sentence to see whether “reimburse” sits inside “travel reimbursement” or inside “reimbursement deadline.” Highlighting supplies location, not extra explanatory prose, and not a substitute for the sentence as decision evidence. Evidence is what the sentence says; location is which span lined up with the query.

Without highlighting, the match relation has to be found in the literal by the reader; with it, scanning can hop by bright patches. The reading paths differ.

Why it happens

Visual search along a result row is pulled by salient patches. Once query terms are bold or colored, saccades land on the match, and working memory immediately gets “this local span is what lined up.” The same word in the title, the URL, or the body snippet implies different match types: a title hit is more known-item, a body hit more topical. Highlighting turns type into spatial distribution without a separate caption “matched in body.”

Highlighting also exposes false matches. Query “apple” lighting up inside “apple pie recipe” when the user wanted a brand is itself evidence to skip. Without highlighting, that error waits until the whole sentence is read, or until after the click. Over-highlighting (synonyms, stems, almost every token lit) makes patches lose pointing; location signal drowns in noise and people fall back to reading the whole sentence.

Studying it

Compare time to find the matching span, and ability to say where a match went wrong—not whether highlighting looks nice.

• Paradigms: snippet studies with highlighting on/off, recording first-fixation landing and time to judge whether the match sits in the expected slot; KWIC versus an unmarked full sentence. Hearst lists query-term highlighting as a standard result-row part.
• Independent variables: exact terms versus expanded terms, highlight in title versus snippet, contrast.
• Dependent variables: time to locate the match, detection of false matches, rate of ignoring real evidence because of highlight noise.
• Methodological note: in color-blind and forced-colors modes, color-only highlighting vanishes; include a bold or underline condition. Very short, very common queries light up everywhere and ceiling-out the location signal; pick queries with enough discrimination.

Where it stops holding

Logographic queries in Chinese are often one or two characters, so highlight density is naturally higher and expansions must be more conservative than in English (do not light every near-synonym). Code, error messages, and identifiers should highlight the literal; stemming lights the wrong token. Yellow-on-cream highlighting dies in dark mode and low-contrast themes; highlighting is a contrast problem, not a palette accent. Screen readers do not “see” color; match location needs to be speakable as “the query term appears in…,” or highlighting only serves sighted scanning.

Applying it

• Highlight query terms exactly in title and snippet (bold before color-only) so the matching span is first on the scan path.
• Mark expanded matches (synonyms, post-correction terms) more weakly, and avoid lighting most of a sentence.
• If a false match is visible in the highlight, do not rewrite the snippet to hide it; part of the location signal’s value is “see that you should not click.”
• Verify with a polysemous query: without opening, can people point at “which sense is lit.” If they cannot name a location, highlighting is not supplying match information. If every content word in the sentence is lit, location has stopped being a signal.

Related

• Within the group: G3.06.1 A snippet must contain the evidence needed to decide whether to click · G3.06.3 Uniform snippet templates erase differences among results
• Adjacent: G3.17 Search result explainability · G3.04 Spelling correction · G3.02 Query formulation
• Search terms: keyword highlighting · KWIC · query term