Text-generation meta-analysis (Educational Psychology Review 2023) — Hedges g ≈ .41; LARGEST for 301-600 word texts; NO EFFECT beyond ~900 words

Summary

Claim: Text-generation meta-analysis (Educational Psychology Review 2023): Hedges g ≈ .41, not attributable to time-on-task; largest for 301-600 word texts and NO effect beyond ~900 words.

Source: Educational Psychology Review (2023).

Confidence: Verified (meta-analysis).

Caveat: Length ceiling is the critical practical limit.

Why this matters for Candid: Counsel against very-long-form input tools — the generation effect ceilings beyond ~900 words. Practical implication: keep input fields short and focused; if a tool needs many inputs, break into screens (also supports R2 — Design every multi-step tool for the goal gradient: visible progress + low interaction cost + start-state non-empty when possible from Brief E).