Feedback in learning is information given to a learner about the gap between where their performance currently is and where it needs to be, with the aim of closing that gap. Defined this way, feedback is not praise, not a grade, and not the marking itself, but the actionable information a learner can use to do better next time. That distinction does a lot of work, because much of what passes for feedback in classrooms carries little such information.
John Hattie and Helen Timperley's 2007 review The Power of Feedback gives the most used framework. Effective feedback, they argue, answers three questions for the learner. Where am I going? sets the goal and the standard (feed-up). How am I going? reports progress against that standard (feed-back). Where to next? points to the next move (feed-forward). The questions matter in combination: progress information is hard to interpret without a clear goal, and a clear goal without a next step leaves the learner knowing they fell short but not what to do.
The same review distinguishes four levels at which feedback can be aimed. At the task level it says whether work is correct ("the date is wrong"). At the process level it addresses the strategy behind the work ("you could check this against the source first"). At the self-regulation level it builds the learner's capacity to monitor and steer their own learning. At the self level it comments on the person ("you're so clever"). The first three generally help; feedback at the self level, including ability praise, tends not to, because it directs attention away from the task and toward the ego.
The most sobering evidence is Avraham Kluger and Angelo DeNisi's 1996 meta-analysis in Psychological Bulletin, covering 607 effect sizes. On average, feedback interventions did improve performance, with a mean effect of about d = 0.41, but the average hid enormous variability: in over a third of cases, feedback actually reduced subsequent performance. Their explanation, feedback intervention theory, is that feedback works by directing attention, and where it directs that attention decides whether it helps. Feedback that keeps attention on the task and on details of how to improve tends to raise performance; feedback that pushes attention up to the level of the self, whether through praise or through threat to self-image, tends to lower it. This is the empirical backbone of the "stay at the task and process levels" advice.
Information only closes a gap if the learner can act on it. D. Royce Sadler argued that for feedback to function formatively the learner must hold a concept of the standard being aimed at, be able to compare their current work against it, and be able to take action to close the difference. Feedback that assumes a standard the learner cannot yet picture, or asks for a move the learner cannot yet make, stays inert however accurate it is. Paul Black and Dylan Wiliam's work on formative assessment puts the same point at classroom scale: feedback earns its value only when it changes what teacher or learner does next, which is why a comment-only return that students must respond to outperforms a return carrying a mark, since the mark draws attention to rank rather than to the next action.
This concept sits above several more specific notes. The correction of language errors falls under corrective feedback and error correction techniques; feedback from classmates is treated under peer feedback; and the assessment cycle that generates and uses it is formative assessment. The principles here, gap-closing, the three questions, task-not-self focus, govern all of them.
The Hattie and Timperley framework is a synthesis, not a controlled experiment, and the broad averages behind "feedback is among the most powerful influences on achievement" conceal exactly the variability Kluger and DeNisi documented; an average effect of 0.41 with a third of cases negative is not a green light to give more feedback, but a warning to give it differently. The task-versus-self distinction is cleaner in theory than in practice, since real feedback episodes mix levels and the same words land differently depending on the relationship and the learner's confidence. Much of the underlying research also comes from short laboratory tasks, and how cleanly it transfers to the slow, identity-laden business of learning to write or speak a second language is not fully established.
Make the standard visible before commenting, so the learner has something to compare against, and end every piece of feedback with a usable next step rather than a verdict. Keep the language at the task and process levels and resist person-level praise, however well meant. Build in time and structure for learners to act on feedback, since unacted feedback changes nothing; returning comments without marks, then requiring a revision, is a low-cost way to force that action. Given that a substantial share of feedback episodes harm performance, the goal is not more feedback but better-targeted, actionable, gap-closing feedback that keeps attention on the work.