Results should say whether the match was in the title, body, tags, or another field

What it is

A result ranks because the query hit some field: title, body, creator, tags, attachment name, comments. Field-level match attribution writes that field down. It is not the same as painting characters yellow in a snippet. Yellow answers “which words in this passage coincide.” Field attribution answers “which layer of metadata the system took seriously.” A title hit and a footnote hit are different evidence; people need the difference in view before they can decide whether to rewrite the query or switch to a field restriction.

It is not a disclosure of ranking weights (“title ×3”). It is naming the slot the hit landed in.

Why it happens

People hold different expectations per field: title as the object’s true name, body as content, tags as someone else’s filing, filename as a storage habit. Which layer the hit lands on decides which layer to repair next—wrong title, change words; right body, wrong title, add quotes or a field prefix; tags only, distrust the tagging scheme. Unlabeled, every hit collapses into one kind of “contains this word,” and rewriting has no fulcrum.

Multi-field indexes also manufacture false relevance: the query is a person, the hit is a namesake in comments; the query is an error code, the hit is an unrelated document’s attachment name. Snippet highlighting may never show that field. Attribution lifts “why is this row here” from character overlap to structural place, so working memory can treat the row as one kind of evidence rather than a generic “found it.”

Studying it

Manipulate presence of field attribution and watch whether query rewriting moves toward the right field.

• Paradigms: one result set; one condition highlights in a body snippet only, the other marks “title / body / tags / attachment” on the row; tasks include “only if this word is in the title” and “discussed in the body.” Enterprise search and mail are field-rich field settings.
• Independent variables: whether the hit field is shown, whether field names use user language, how multiple hitting fields are listed on one row.
• Dependent variables: rewrites into a field restriction (title: or an in-UI “title only”), rate of treating a comment hit as an object hit, task completion.
• Methodological note: if every hit is in the title, attribution adds no information and differences will not show. Mix in body-only and metadata-only hits on purpose. Split highlighting from field attribution; do not count “there is highlight” as field explanation already done.

Where it stops holding

A single-field collection (a title-only catalogue, a filename list) makes field labels noise. On a screen reader the field name has to enter the accessible name, or attribution exists only for sighted users. Internal code names (body_html, facet_3) make attribution more confusing; they must map to user language. Hits in hidden fields (ACL, internal ids) must not be shown as the matching field, or structure leaks; rewrite as “matched on permission or an internal number,” or do not use hidden fields as retrieval evidence.

Applying it

• Mark the hitting field on the row with a short label (“Title,” “Body,” “Tags,” “Filename”). List multiple hits; do not rely on one highlighted snippet.
• Use the same words as navigation and forms, not index schema names.
• Offer a follow-on narrowing (“in title only”) so attribution can become an action immediately.
• Verify by mixing in objects that hit only in tags, with titles that do not contain the query. People should be able to say “it appeared because of tags,” and choose whether to search titles only. If all they see is a title without the word plus an unrelated highlighted snippet, the field was not marked.

Related

• Within the group: G3.17.2 Personalized ranking must be labeled as customized, not as the general result · G3.17.3 Implicit filters need their source disclosed · G3.17.4 Raw relevance scores are not fit to show end users · G3.17.5 Explanations exist to help reformulate the query, not to prove the ranker is right
• Adjacent: G3.06 Result snippets · G1.05 Metadata · G3.05 Result ranking
• Search terms: match attribution · fielded search · why this result