Metacognitive Training Facilitates Optimal Cognitive Offloading

Synthesis: Ngai & Gilbert (2026) show that a brief metacognitive training intervention — just five practice trials pairing performance prediction with feedback — improves metacognitive calibration and makes people's Cognitive Offloading strategy choices measurably more optimal, resolving earlier mixed findings about whether metacognitive training translates into actual offloading behavior.

Ngai and Gilbert (2026) report two preregistered experiments testing whether a brief metacognitive intervention can reduce systematic biases in Cognitive Offloading. People are known to offload suboptimally — most commonly over-using external reminders even when internal memory would earn more reward — a bias linked to miscalibrated metacognitive confidence. Prior attempts to shift offloading via metacognitive interventions produced inconsistent results, so this study tested a tightly-specified training recipe under conditions designed to isolate metacognitive effects.

The Intervention

Participants completed a brief practice session (just five trials) in which they made metacognitive predictions about how well they would perform, then received feedback on their actual accuracy. This prediction-plus-feedback loop was designed to improve calibration — bringing confidence in line with true ability — rather than simply raising or lowering confidence wholesale. The training was delivered immediately before an "optimal offloading" task in which participants decided trial-by-trial between internal memory (max reward) and external reminders (reduced reward), allowing reminder bias to be quantified objectively against each individual's optimal strategy.

Findings

• Experiment 1 (N = 164): Participants who made predictions and received feedback showed significantly improved metacognitive calibration and more optimal reminder-setting than the no-feedback control.
• Experiment 2 (N = 416, four-group additive design): Replicated the benefit and isolated its mechanism:
  • Making predictions alone was ineffective (no significant reduction in bias).
  • Adding performance feedback drove the improvement — the key transition occurred between the prediction-only and prediction-plus-feedback groups.
  • Explicitly labeling over-/under-confidence added no further benefit beyond the performance feedback itself.
• Effects on absolute (not signed) bias: The intervention improved calibration by correcting individual miscalibration in both directions — increasing confidence in underconfident participants and decreasing it in overconfident ones. Signed (directional) bias effects were not significant, but absolute bias improved.
• Metacognitive hindsight bias persisted: Participants' memory of their earlier confidence was systematically distorted toward their actual performance, and the intervention did not reduce this.
• Confidence continued to explain significant variance in offloading choices even after controlling for objective optimal strategy.

Why It Differs From Prior Null Results

The authors attribute their success — versus Engeler & Gilbert (2020) and Grinschgl et al. (2020, 2025) — to two design choices: (1) financial incentive tied to the optimality of offloading strategy, which minimized non-metacognitive influences like effort-avoidance; and (2) veridical, immediate, trial-by-trial feedback explicitly linked to the prior prediction, rather than a single block-level or fake feedback statement. This supports the view that feedback is most effective when salient, explicit, and iteratively connected to one's own prediction.

Connections to the Knowledge Base

This study provides strong empirical grounding for the knowledge base's Cognitive Offloading and Metacognition concepts, converging with the metacognitive beliefs-vs-experiences framework from Guo & Ye (2026): experience-targeting feedback (immediate performance feedback) is what drives change, while beliefs alone (predictions without feedback) do not. It also bears on Over-Reliance (reminder bias is a laboratory analogue of over-reliance), Feedback (design principles for effective feedback), and Self-Regulated Learning (calibration as a trainable regulatory skill).

What this means for practice

• Learners. Before a study or retention task, predict how well you will do and then check that prediction against immediate results: five prediction-plus-feedback trials in this study improved calibration and made reminder-setting more optimal, while prediction without feedback did nothing.
• Instructors. Give feedback that corrects miscalibration in both directions — the intervention raised confidence for underconfident participants and lowered it for overconfident ones, and it improved absolute bias even though signed directional bias did not move.
• Instructors. Do not substitute generic confidence prompts ("how confident are you?") for performance feedback: explicitly labeling over- or under-confidence added no benefit beyond the feedback itself in the four-group Experiment 2 (N = 416).
• Designers. Tie feedback to the learner's own prior prediction and deliver it trial by trial, and consider attaching a consequence to strategy optimality — the authors attribute their success over earlier null results to exactly these two design choices.
• Instructors. Do not expect metacognitive training to remove hindsight bias about performance: participants' memory of their earlier confidence shifted toward their actual performance, and the intervention did not reduce this.

Limitations

• Both experiments sampled online Prolific participants (Experiment 1 N = 164; Experiment 2 N = 416) paid £2.75 plus a £1 bonus for scoring in the top half, performing a laboratory circle-matching task rather than course tasks, so the findings describe paid online adults rather than classroom learners.
• The effect depended on a financial incentive tied to the optimality of the offloading strategy; the authors credit this design choice for beating earlier null results, so the benefit may not transfer to settings without a comparable incentive.
• Experiment 1 was powered to detect d = 0.39 — half the effect size reported by Engeler and Gilbert (2020) — and the intervention moved absolute bias but not signed bias, so the effect is modest and directional improvement is unproven.
• Generalization beyond the reminder-setting paradigm is untested: reminder bias here is a laboratory measure of intention offloading, and the study reports no follow-up on whether more optimal reminder use persists or reaches everyday remembering.

Citation

Ngai, C., & Gilbert, S. J. (2026). Metacognitive training facilitates optimal cognitive offloading. Cognitive Research: Principles and Implications, 11(21).