Corrective Feedback Timing in Speaking Practice: Evidence Brief

There is no evidence-based rule that every speaking error should be corrected immediately or that all feedback should wait until the end. Timing interacts with the target, task, feedback move, learner, and outcome. Immediate feedback can make a form available during subsequent attempts; delayed feedback can preserve a communicative turn and create space for review. A defensible practice policy defines the purpose first, then tests whether correction helps on a later, unprompted task.

Scope and method

The search was conducted and sources were accessed on 29 July 2026. It covered peer-reviewed timelines, systematic reviews, and comparative empirical studies that explicitly manipulated or analysed the timing of second-language corrective feedback. Priority went to oral or communicative tasks, although grammar-learning measures were retained when they clarify what a timing result actually tested.

Included evidence had to define when feedback occurred relative to an error, task, instructional session, or later review. Exclusions were unsourced teacher tips, product marketing, studies that compared feedback with no feedback but not timing, and written-only results presented as direct proof about conversation. Search terms combined second language, corrective feedback, immediate, delayed, oral, speaking, task, and timing.

This is a rapid evidence brief. It did not use a preregistered protocol, duplicate screening, formal risk-of-bias instrument, or meta-analysis. The source set is illustrative rather than exhaustive.

“Immediate” and “delayed” are not stable treatments

The 2026 Language Teaching research timeline shows that timing labels refer to different timepoints across studies. Immediate may mean inside the task, directly after an erroneous turn, or during the same instructional session. Delayed may mean after a task, in a later session, or after intervening practice (timeline and definitions).

The feedback itself also varies. A teacher may recast the utterance, prompt self-repair, provide an explanation, display a transcript, or collect several examples for later discussion. Two studies labelled “immediate versus delayed” can therefore compare different learning events.

Design choice	Questions that must be reported
Trigger	Which errors were eligible and who selected them?
Timing point	During the turn, after the turn, after the task, or in a later session?
Feedback move	Recast, prompt, explicit rule, transcript, model, or combination?
Opportunity to act	Did the learner immediately retry or use the form in later items?
Target	New grammatical form, known recurring error, pronunciation, pragmatics, or task content?
Outcome	Judgement test, imitation, planned production, spontaneous speech, or transfer?
Delay	How long after treatment was the outcome measured?

Without these details, “immediate worked better” is too vague to implement.

What comparative studies show

Fu and Li assigned 145 adolescent EFL learners to immediate feedback, delayed feedback, task-only, and control conditions for focused communicative tasks on the English past tense. The study reported an immediate-feedback advantage on some measures, while the outcome pattern did not reduce to a universal benefit across all tests (design and results). The target was one grammatical form in a structured intervention. It was not a test of interrupting free conversation whenever an error occurs.

An earlier study of a new English past-passive structure compared immediate, delayed, task-only, and control conditions with 120 learners. Both feedback conditions improved a grammaticality-judgement outcome, no feedback effect appeared on the elicited-imitation measure, and the authors interpreted the immediate advantage in relation to opportunities to use feedback while producing later sentences (primary study record and accepted manuscript). The result illustrates outcome dependence: a timing effect on explicit judgement does not automatically establish spontaneous speaking transfer.

A 2023 systematic review found a small and heterogeneous timing literature and highlighted variation in modality, target, and definitions (review). A 2026 System study also reported that timing interacted with task complexity and working memory, with delayed feedback showing an advantage in that specific design (study abstract). Taken together, these findings argue against one fixed interruption rule.

A decision rule for practice

Correct now when the next attempt depends on it

Immediate feedback is plausible when a learner is drilling a selected feature, misunderstanding prevents the task from continuing, a safety-critical meaning is wrong, or a short prompt can enable an immediate retry. Limit the correction to the target. A pronunciation contrast drill is not improved by adding five unrelated grammar comments.

Note and return when communication is the target

Delay feedback when the learner is completing a narrative, negotiating, practising turn-taking, or building willingness to sustain a response. Capture only a few high-value examples. Revisit them soon enough that context is still recoverable, and provide a chance to reformulate.

Ask before correcting identity- or register-sensitive language

Accent, dialect, politeness, and pragmatic choices are not reducible to a single standard form. Ask what community and goal the learner has. High-stakes or culturally sensitive feedback should be checked by a qualified human with relevant language-variety knowledge.

The practical immediate-versus-delayed correction guide turns this evidence into a session decision tree. The companion guide on giving useful speaking feedback separates task success from one selected language target.

A small evaluation design

1. Choose one recurring feature and define an acceptable response.
2. Randomise comparable prompts to immediate or end-of-task feedback.
3. Keep the correction form and practice opportunity constant.
4. Record uptake on the next attempt, but do not call uptake learning.
5. Test the feature later in an unprompted, comparable speaking task.
6. Record communication cost: abandoned turns, time, confusion, and learner preference.
7. Inspect individual patterns before making a default policy.

The speaking-transfer evaluation protocol explains how to add an untrained task and delayed measure. The technology feedback map covers machine signals that require human interpretation before correction.

Learners may use the LinguaLive practice tools to repeat a task and review selected feedback. The system is not an official examiner, and a generated correction should be verified when meaning, register, or assessment stakes are high.

Limitations

Much timing research targets a particular grammatical structure, short instructional sequence, and school or university sample. Results may not transfer to open-ended voice conversation, other languages, different ages, or long-term interaction. Timing is confounded easily with feedback type, attention, number of learning opportunities, teacher behaviour, and task complexity. Published group averages can hide learners who respond differently.

The evidence does not license claims that interruption is always harmful, delayed summaries are always better for fluency, or immediate correction prevents fossilisation. It also does not validate a product's default timing without product-specific outcome and harm data.

Editorial disclosure

LinguaLive sells AI-supported speaking practice and can benefit commercially from claims about feedback. No proprietary LinguaLive result is included. Vlad Podoliako is identified as the company's founder, not as an academic corrective-feedback specialist. Independent research and practitioner review should precede an efficacy claim or institution-wide feedback policy.