Query formulation: teaching children to ask machines good questions

Evidence grade: PROMISING-to-STRONG. How children search has a real research base (query logs, task studies of young searchers), with consistent findings: natural-language questions, single attempts, first-result dependence. That instruction improves query behavior has support; the evaluation half of search (lateral reading, ad-spotting) has the stronger trials and its own pages.

Before a child can evaluate results, they need results worth evaluating — and the research on how young searchers actually behave shows the gap: children (like novice adults) type whole natural-language questions, spell them approximately, issue one query, and treat its first screen as the answer (position ≠ endorsement). Modern engines are forgiving enough to make this sort of work, which is exactly why the underlying skill goes untaught — until the question is hard, and the gap appears.

What the research shows about young searchers

The children's-search literature (Druin, Foss and colleagues' studies of young searchers; Bilal's task studies; query-log analyses) documents recurring patterns: query-as-question ("what do dragons eat in winter") rather than term selection; vocabulary lock — reusing the same failing words rather than reformulating; first-page dependence; and difficulty with the translation step — knowing what you want to know, but not which words the answer would contain. That last one is the teachable core, because it names the actual cognitive move of searching: imagine the page you're hoping for, and type the words that would be on it.

The skills, in teaching order

Keyword thinking — the translation move above. The classroom exercise that teaches it: take a question ("which dragon is fastest?"), ask "what words would the answer page contain?" ("fastest dragon" — maybe "speed"), and search those. Children find the reframe genuinely satisfying — it's a puzzle, not a rule.
Reformulation as the norm — the single-attempt habit is the one to break: no results ≠ no answer; it means different words. Synonym-generation ("what else might they call it?") is vocabulary work with a search payoff, and the metacognitive frame — "the first query is a draft" — normalizes iteration.
The operators that earn their keep: quotes for exact phrases (the one operator with big, teachable payoff — words that belong together), and minus-terms for the disambiguation cases (jaguar -car). The exotic operators can wait; two well-automatized beats ten half-known.
Scan before you click — click restraint, the bridge into the evaluation curriculum: the results page read as a menu with labels, not a verdict.

What the evidence doesn't say

It doesn't condemn natural-language queries outright — modern engines (and now AI answers) handle them increasingly well; the skill matters when the easy path fails, which is precisely when children currently give up.
It doesn't separate cleanly from evaluation — query skill gets you candidates; judgment picks among them; teach as one arc.
It doesn't age the AI-search question away — asking machines good questions is more load-bearing in the chatbot era, not less (the well-read guesser rewards precise asks and iteration identically).

In the classroom

Run the answer-page game weekly for a term: question on the board, class proposes the words the answer would contain, compare result quality live.
Celebrate reformulations: "that's your second query — that's what real researchers do" flips the failure frame.
Teach quotes with a designed contrast — a search that floods without quotes and lands with them makes the operator unforgettable.
Keep a class synonym wall for search vocabulary — the translation move, made visible.

How Wiz Kids applies this

The Oracle realm teaches the arc in order: asking with the right words (keyword thinking against our authored web, where query quality visibly changes results), judging what came back (sponsored entries, second sources), exact-phrase quotes with a task built on the flood-then-land contrast, and multi-page research trails — with the searched-goals engineered generously so any reasonable phrasing counts, because the lesson is thinking, not incantation.

References

Druin, A., Foss, E., et al. (2010). Children's roles using keyword search interfaces at home. CHI 2010 — and the successor studies of young searchers.
Bilal, D. (2000–2002). Children's use of the Yahooligans! search engine (I–III). JASIS(T) — the task-study classics.
Wineburg, S., & McGrew, S. (2019). Lateral reading. Teachers College Record, 121(11) — click restraint, the bridge skill.

© Glu IO Pty. Ltd. — Wiz Kids (wiz.kids). Link freely; republication requires permission — see terms. Found an error in our reading of the research? We correct fast: tell any teacher piloting Wiz Kids.