Zero results should offer nearby queries rather than stop

What it is

The query has already run and the set is empty. Do not leave only “0.” If emptiness likely comes from spelling or a near-miss wording, the interface should offer approximate queries: one or two phrasings, already probed to hit something, that can be searched with a tap. This appears when auto-correction did not rewrite the original, or after the user just reverted to it—the original was respected, the empty set remains, and the repair is the next shot, not an insistence on the blank page.

An approximate query is a new query suggestion, not fuzzy hits silently poured into the same query. A tap switches to that sentence; without a tap, the current search stays zero. It also does not explain why the count is zero or strip filters; those are other zero-result actions.

Why it happens

An empty set breaks the feedback loop of query formulation: people do not know which character to change or which word to swap. Spelling correction’s value here is not rewriting the submission already made; it is turning “it might be this” into a visible option so working memory need not regenerate from blank. Nearby queries on an empty page are safer than auto-correction on a non-empty one—no list currently being read is swapped underneath, and the choice is delayed until after the user has seen “0.”

Offering has to be active, because the default moves on an empty page are leave, or randomly change one character. Hiding candidates behind advanced syntax or “please check your spelling” returns the generation cost. Candidates must be submittable queries, not a diagnosis (“you may have misspelled something”): a diagnosis is not a next action. If the nearby phrasing is itself empty, offering it produces a second zero and trust drops, so probe before display.

Studying it

On spelling variants that truly go empty, test whether a tappable next query shortens recovery.

• Paradigms: known items misspelled into zero results; one condition shows only 0, another offers probed nearby queries; spelling-correction UX studies that split “suggestion after empty” from “auto-rewrite when results exist.” Query logs: whether the query after a zero equals the sentence the system offered.
• Independent variables: whether nearby queries are shown, how many, whether each candidate is guaranteed non-empty.
• Dependent variables: time from zero to hitting the target, uptake of the offer, rate of a second zero.
• Methodological note: do not test this leaf on queries that already return hits, or nearby queries never appear. If candidates were not probed, count those failures against the suggestion, not as “the user misspelled again.” When a proper noun is the nearby target, mark separately whether the entity changed, so “successful recovery” is not a wrong person.

Where it stops holding

When zero is caused by a filter that is too tight, a wrong scope, or missing permission, nearby spelling queries do not help, and random spelling candidates mis-attribute the cause. When the object truly is not in the collection, offering a related but different object is passing off a neighbor as a hit. Emptiness from speech-recognition errors may need homophones rather than lookalikes; mixed strategies from mixed sources fail. If the user has already seen an auto-correct banner and chosen to keep the original, pushing the same correction as a nearby query ignores the revert just made.

Applying it

• When results are zero and the query looks like a spelling or wording near-miss, put one or two probed, tappable queries in the empty state, worded as the queries themselves, not “please check spelling.”
• Do not show a candidate that failed its probe; fewer rows beat a chain of empty pages.
• After a tap, run an ordinary result page and allow return to the original zero, so help does not become an irreversible rewrite.
• Verify by forcing zero with a query whose target exists a character or two away. If the empty page has no tappable next sentence, the offer did not happen. If the next sentence is still zero, or lands on a different entity, the candidate is unfit.

Related

• Within the group: G3.04.1 Auto-correction must show the original query and allow reverting it · G3.04.2 Proper nouns carry the highest spelling-correction risk
• Adjacent: G3.07 Zero-result handling · G3.14 Fuzzy matching and approximate search · T2.05 Empty-state and zero-result copy
• Search terms: spelling correction · zero results · query suggestion