VIEW THIS AS

Auto mode follows the Route Engine until you choose a viewpoint.

YOU ARE HERE

ROUTE CHECK

CONNECTED TO

WHAT NEXT

Use the canonical route for this room, or HELP if you are unsure.

AI and Feedback | How Learners Use Advice to Improve

Three students sit around open books and worksheets at a classroom table, reading, writing and discussing the work together.

AI feedback becomes useful when a learner can explain what the advice means, check whether it fits the work, and choose a change for a reason. A longer comment or a smoother second draft does not establish that this has happened. Teachers and parents need to see the learner’s decision, the resulting revision, and, later, what the learner can do when the same advice is no longer supplied.

Start with one clear learning goal and a small piece of work. Ask for a specific observation, check it against the task and its evidence, then let the student make and explain a revision. This article develops that approach through a complete fictional writing lesson, including useful feedback, an unsupported suggestion the learner rejects, an improved paragraph, and a later check of independent understanding. The routines are proposed teaching designs, not an official assessment policy or a guaranteed route to higher marks.

  • Separate the quality of the feedback, the quality of the revision, and the learner’s later independent performance.
  • Check advice against the actual task, source material, subject knowledge and relevant criteria.
  • Give learners room to accept, adapt, question or reject a suggestion with reasons.
  • Keep the revision record brief enough to serve learning rather than become another assignment.
  • Preserve access supports, protect teacher time, and help parents discuss the work without taking over authorship.

CHAPTER 1 OF 19 · Understand and check advice

1. Start with the decision the learner needs to make

Back to contents

A student receives the comment “develop your explanation”. Before changing anything, the student needs to locate the explanation, identify what is missing, and understand what a fuller explanation would accomplish. Adding three longer sentences might leave the original gap untouched. A useful next question is: “Which connection would a reader still have to work out?” The answer directs attention to reasoning rather than length.

Treat a feedback comment as information offered for consideration. The educational process includes interpreting that information and deciding what to do with it. An accurate observation may remain unused; an inaccurate observation may be followed enthusiastically. These are different problems, requiring different responses. A teacher might clarify a misunderstood comment, teach a missing concept, correct the advice itself, or provide time to revise.

For this article, the working sequence is: locate the issue, check the advice, choose a response, make the change, explain it, and revisit the underlying idea. It is a proposed routine that teachers can shorten or adapt. A child who needs help reading the comment can still make the intellectual decision. A fluent reader who copies a replacement paragraph may have made very few decisions.

This extends the approach to purposeful work in AI-powered assignments: ask what the learner is becoming able to judge, not simply how much feedback the system can produce.

Contents · Next chapter

CHAPTER 2 OF 19 · Understand and check advice

2. Read feedback research at the level it supports

Back to contents

Carless and Boud’s 2018 paper develops a conceptual account of feedback literacy focused on undergraduates in higher education. It connects recognising the value of feedback, judging quality, handling emotional responses and taking action. That framing helps explain why receiving comments and using them are different matters. It is not an experiment establishing the effectiveness of an AI routine for all children, and the school activities here are adaptations proposed by this article.

The Education Endowment Foundation’s 2021 guidance concerns teacher-delivered feedback for pupils aged five to eighteen. It emphasises instruction, learning goals, appropriate timing and opportunities to use advice; it also asks schools to consider workload. Its scope supports attention to the teaching surrounding feedback. It does not establish that an AI comment stream reproduces those conditions or that one fixed frequency suits every lesson.

Steiss and colleagues’ 2024 study compared human and ChatGPT feedback on secondary students’ writing. Trained human evaluators produced higher-rated feedback on four of five assessed dimensions; the exception concerned alignment with criteria. The study assessed feedback quality. It does not by itself establish students’ subsequent uptake, independent retention, or the performance of every later model. This article therefore avoids treating a convincing comment as evidence of learning.

These sources motivate questions for practice, not a universal recipe. The lesson below is an original illustration. Its invented numbers, dialogue and revisions are not research results, and no effect size is attached to the proposed routine.

Contents · Previous chapter · Next chapter

CHAPTER 3 OF 19 · Understand and check advice

3. Make the learning goal visible before requesting advice

Back to contents

Define the current goal in terms a learner can apply. “Improve this writing” is broad. “Make a recommendation that uses the supplied evidence without claiming more than it supports” gives the reader something to inspect. Other lessons may target paragraph sequence, accurate explanation of a mechanism, or a particular sentence construction. Choose deliberately; they require different feedback.

Show what the criterion means through a small contrast. “Two interviewees preferred the later session” describes the available evidence. “Everyone prefers later sessions” extends beyond it. Ask students what changed between the claims before asking them to assess their own sentences. This establishes a shared reference point and reveals whether the prerequisite idea needs teaching.

The task instructions should also distinguish requirements from options. A required reason supported by evidence is different from a suggestion to vary sentence length. Learners should know which features are central to this assignment, which are preferences, and which can wait. A model should not turn its preferred style into an invented marking rule.

If the student cannot identify a claim or distinguish it from supporting evidence, begin there. Repeated comments about “stronger justification” will not supply that foundation automatically. The curriculum progression article develops this readiness principle. Readiness is specific to the task: a student may assess factual support confidently while still needing direct teaching about how to organise a paragraph.

Contents · Previous chapter · Next chapter

CHAPTER 4 OF 19 · Understand and check advice

4. Read each comment as a claim that can be checked

Back to contents

Ask four practical questions. What exact feature does the comment identify? Where can I see that feature in my work? Why does it matter for this task? What evidence supports the proposed change? An observation that cannot point to the relevant sentence may be too vague to use. A recommendation that depends on an absent source needs verification before inclusion.

Different comments require different checks. A calculation can be recomputed. A statement about a document can be compared with the document. Advice about audience or tone needs to be judged against the communication purpose. A claim about an examination requirement needs an applicable authoritative source, not an AI assurance. Students should be taught these distinctions through examples.

Ask the tool to separate observations from suggestions. “The paragraph says ‘every Friday’, while the source offers one additional Friday” is a checkable observation. “Begin with a conditional recommendation” is a possible response. The student may accept the observation while choosing a different effective revision. This preserves room for authorship without making factual accuracy optional.

An unresolved comment can be marked “question” rather than forced into agreement or rejection. The learner might ask the teacher whether a phrase changes the strength of a claim. That question is useful evidence of judgement. Repeating the prompt until the AI agrees with a preferred answer would provide much weaker evidence.

Contents · Previous chapter · Next chapter

CHAPTER 5 OF 19 · Understand and check advice

5. Select changes that matter and can be completed

Back to contents

Prioritise a change that affects the meaning or purpose of the work. If the main recommendation rests on a false reading of the evidence, decorative vocabulary should wait. If the reasoning is sound but an ambiguous pronoun obscures who did what, resolving the reference may be the most useful move. Priority depends on the actual draft and learning goal.

A teacher can propose one substantial revision and one optional refinement, then adjust the amount according to the learner’s capacity and the task. This is a workload choice for the lesson, not a scientifically fixed maximum. Explain the priority: “First repair the evidence claim, because your conclusion currently depends on it.” Students should understand why one change comes before another.

Before revising, ask for a short intention: “I will replace the claim about everyone with a statement about the respondents.” After revising, compare the result with that intention. If the paragraph now adds a different unsupported claim, the revision has not achieved its purpose, even if the original sentence disappeared.

Students can also identify what should remain. A clear opening, a well-chosen example or an accurate qualification need not be rewritten simply because feedback was requested. A sensible decision record includes justified preservation. The aim is purposeful improvement, not a quota of visible edits or a requirement that every draft become longer.

Contents · Previous chapter · Next chapter

CHAPTER 6 OF 19 · Follow the complete revision

6. Set up the worked lesson: a repeat lunchtime performance

Back to contents

Everything in this case is fictional, including the school event, records, student writing and sample comments. The teacher asks learners to recommend whether the school should offer one repeat lunchtime performance of a short student play. The paragraph must use at least two facts from a common source pack and identify a practical condition. It may recommend proceeding, postponing, or gathering more information.

The pack contains five short records:

  • An attendance record says 48 students attended the first performance in a room with 60 seats.
  • Seventeen audience members voluntarily returned an exit card: twelve said they would attend again and five were unsure.
  • Four of those seventeen respondents reported missing the opening because of the canteen queue.
  • The director says the cast could perform on one additional Friday if a stage is available; no weekly commitment has been offered.
  • The hall manager says the 12:50–13:10 slot is currently free on that Friday, but a booking still needs confirmation.

The pack does not give the play’s running time, opinions from the entire school, or evidence about examination results. Those absences are intentional. Students can identify information they would need without inventing it. Twelve positive responses plus five uncertain responses account for all seventeen cards; neither group represents everyone who attended.

The learner’s job is to make a bounded recommendation. The teacher should make clear that a thoughtful request for missing information can meet the goal. There is no requirement to recommend an event simply because the sources describe some interest.

Contents · Previous chapter · Next chapter

CHAPTER 7 OF 19 · Follow the complete revision

7. Inspect the first draft before adding feedback

Back to contents

Before: the learner’s initial paragraph

Our school should put on the play every Friday because almost everyone wants to see it again. There were 48 students at the first performance and 12 people said they would return, so a weekly show will certainly fill the hall. The cast is available on Fridays. Moving the start later will solve the queue problem and give everyone enough time for lunch. A regular performance would make school more enjoyable. The organisers should announce it now.

The draft has a clear position and attempts to use evidence. Those are useful starting points. However, it turns a proposal for one repeat into a weekly commitment, combines counts from different groups, and promises outcomes the records cannot establish. The room was not full at the first performance: sixty seats minus forty-eight attendees leaves twelve unused seats. That observation still does not predict the audience for a repeat.

Before seeing comments, the learner underlines the strongest claims: “every Friday”, “almost everyone”, “certainly” and “solve”. The teacher asks which record supports each phrase. This initial attempt matters because it shows what the student can already notice. Feedback can then address a remaining gap instead of repeating a correction the learner has independently made.

The classroom discussion should also identify the valid core: an event took place, students attended, and some respondents expressed interest in another performance. Repairing the paragraph does not require abandoning its position. It requires making that position proportionate to the evidence.

Contents · Previous chapter · Next chapter

CHAPTER 8 OF 19 · Follow the complete revision

8. Examine useful, questionable and unsupported suggestions

Back to contents

Constructed feedback for the exercise

These comments are written for this teaching example. They are not a transcript from a named product or evidence of how often any model behaves this way.

  • “Check ‘every Friday’. The director offers one additional performance, so your recommendation currently promises more than the source supports.”
  • “Replace ‘almost everyone’ with a description of the seventeen respondents. Twelve positive cards cannot establish the views of the whole school.”
  • “State what must be confirmed before an announcement. The hall slot is available provisionally, and the pack does not state the running time.”
  • “Strengthen your case with this research fact: a national education study found that weekly performances raise examination scores by eighteen per cent.”
  • “Replace ‘show’ with ‘performative enrichment intervention’ to make the paragraph sound more sophisticated.”

The supposed research fact in the fourth comment is deliberately invented and unsupported. It must not be presented as a real finding. The learner rejects it because the pack supplies no study, population, method or traceable source, and the task does not require an examination-score argument. In an actual assignment, missing support is sufficient reason to withhold a claim; it does not prove that no related research exists anywhere.

The fifth suggestion is a stylistic preference with no demonstrated benefit for this audience. The student can keep “performance” or “show” because either communicates the subject clearly. This is a different reason for rejection from the unsupported research claim. Evaluating advice includes distinguishing an evidence problem from an unnecessary change in voice.

Contents · Previous chapter · Next chapter

CHAPTER 9 OF 19 · Follow the complete revision

9. Record the learner’s decisions briefly

Back to contents

The learner now chooses responses rather than accepting the whole batch. The decision record below is an illustrative completed example. It is deliberately selective: a classroom version might record only the most consequential change and one questioned suggestion. The teacher needs enough evidence to understand the decision, not an essay about every edit.

Advice consideredDecisionReason checkedVisible response in the revision
Limit the proposal to one additional performance.Accept.The director offers one additional Friday. The source does not promise a weekly programme.Recommend a single repeat and remove the weekly commitment.
Describe the seventeen respondents instead of almost everyone.Accept.Twelve positive cards and five uncertain cards describe the respondents. They cannot establish whole-school demand.Name the responding group, retain both counts and qualify the conclusion.
Confirm timing before announcing the event.Adapt.The hall slot is provisional. The play’s duration is missing, so a later start cannot yet be guaranteed to fit.Suggest considering 12:50 while explicitly requiring a duration and booking check.
Add the invented eighteen-per-cent examination claim.Reject.No supporting study is supplied or verified. The claim is unnecessary for the assigned recommendation.Leave the supposed finding out and use the actual source pack.
Replace simple wording with performative enrichment intervention.Reject.The change adds no useful meaning for the student audience and is not a task requirement.Keep clear wording such as performance or show.
Illustrative learner decision record for the fictional performance recommendation

For the timing advice, the learner accepts the need for a condition but does not invent a running time. The revision can identify the missing check directly. This is an adaptation: the student acts on the concern while choosing the wording and retaining the uncertainty. “Question” remains available if the source itself is ambiguous.

The record should not reward disagreement for its own sake. Accepting a correct comment can show good judgement; rejecting one requires an appropriate reason. Nor should disagreement be treated as disrespect. A learner who can point to a relevant sentence, criterion or source has given the teacher something specific to discuss.

After a short conference, the student should do the writing. If an adult or tool supplies the final paragraph, record that support honestly and provide another opportunity for the learner to practise the target decision. A completed table cannot substitute for understanding the change it describes.

Contents · Previous chapter · Next chapter

CHAPTER 10 OF 19 · Follow the complete revision

10. Read the revised paragraph and trace its improvements

Back to contents

After: a possible learner revision

I recommend one repeat lunchtime performance, subject to confirming the hall and the cast. The first show attracted 48 students, so there is some evidence of interest. Of the seventeen voluntary exit-card respondents, twelve said they would attend again and five were unsure; this does not show what the whole school wants. Four respondents reported missing the opening because of the canteen queue. A 12:50 start is worth considering within the hall manager’s currently free 12:50–13:10 slot, but we still need to check that it leaves enough time for the full performance. The director has offered one additional Friday, not a weekly commitment. After the repeat, organisers could record attendance and ask whether the timing worked before proposing further shows.

The revision preserves a supportive recommendation while changing its scope. It identifies whose views were recorded, removes the guarantee of a full hall, and treats scheduling as something to check. The invented research claim is absent. The paragraph is longer because it now explains relevant distinctions; length itself is not the achievement.

Another good revision could recommend waiting for the running time and hall booking before deciding. Students should compare the reasoning in different acceptable answers. The task is not to reproduce this paragraph’s wording. The learner should be able to explain why the sources permit more than one responsible recommendation and why some claims remain unjustified whichever recommendation is chosen.

Contents · Previous chapter · Next chapter

CHAPTER 11 OF 19 · Check learning and independence

11. Separate accurate advice, learner action and understanding

Back to contents

Three outcomes need separate attention. First, was the comment correct and appropriate? Second, did the learner use it to improve the work? Third, can the learner explain and later apply the relevant idea? A successful result on one question does not settle the other two. The following table offers a practical way to inspect the distinctions.

Outcome to examineEvidence worth looking forWhat that evidence does not establish
Accurate, relevant feedbackThe comment identifies a real issue and can be checked against the draft, source pack and learning goal.The learner has understood or acted on the comment.
A justified revisionThe learner changes a consequential feature and explains how the change improves this draft.The learner can make the same judgement later without the comment or supplied wording.
Independent retention and applicationOn a later task, the learner makes and explains the relevant judgement without the earlier reasoning prompts; access supports are recorded.The skill transfers to every subject, persists indefinitely or was caused by AI alone.
Different evidence is needed for feedback quality, revision and later understanding

Ask the learner to explain one meaningful change in ordinary language. “I changed everyone to respondents because only those people returned cards” is more informative than “I made it more precise”. If the student repeats a technical phrase without explaining it, ask for a small contrast: what could be claimed if every audience member had returned a card? Even then, the audience would not become the whole school.

Do not require a polished spoken defence from every learner. A brief written note, a selection between contrasting claims, or an explanation supported by an appropriate communication aid may show the same target understanding. Decide what intellectual evidence is needed before selecting the response format. The explanation should illuminate the learner’s thinking, not become an additional test of speed or public confidence.

Contents · Previous chapter · Next chapter

CHAPTER 12 OF 19 · Check learning and independence

12. Revisit the idea when the original comments are absent

Back to contents

A later task can check whether the learner remembers how to limit a claim. In an original fictional history exercise, three diary extracts from one journey describe crowded conditions. A sample paragraph concludes that every journey on the route was overcrowded. Ask the learner to revise that conclusion using only the extracts and explain the limit of the evidence.

The new task changes the subject, evidence form and wording while retaining a relevant reasoning demand. There are no exit cards or performances to copy. A learner who qualifies the conclusion has shown something beyond reproducing the earlier paragraph. However, difficulty might also arise from unfamiliar diary language or historical context. Inspect those demands before concluding that the earlier learning disappeared.

Use a short interval appropriate to the teaching sequence, and record what support remains available. An initial attempt without the previous feedback reveals different evidence from a second attempt after a reminder. Both can be useful. Keep them distinct rather than collapsing them into one score. Independence here concerns the targeted reasoning; necessary reading or communication accommodations may remain.

The learning transfer article explores these changed conditions in greater depth. One successful later task supports a bounded conclusion about that performance. A stronger claim about durable, flexible understanding would need repeated evidence across time and sufficiently varied tasks.

Contents · Previous chapter · Next chapter

CHAPTER 13 OF 19 · Check learning and independence

13. Adjust timing and assistance to the learning problem

Back to contents

Some feedback needs to interrupt an error before the learner builds on it. Other feedback is more informative after the student has completed an attempt that reveals their reasoning. In the performance lesson, the teacher might stop an invented source from entering a submitted paragraph immediately while allowing a first draft to reveal how the student interprets the exit cards.

Choose the level of assistance deliberately. A direct explanation may be appropriate when the distinction is new. A question may be enough when the learner already understands it but has overlooked a phrase. A marked location without a solution can invite a more independent repair. These are teaching choices to try and review, not a universal ladder that every student must climb in order.

If the student cannot act after a hint, reduce the size of the problem or teach the missing concept. Repeatedly withholding the explanation can turn an intended learning opportunity into guessing. Conversely, supplying a complete replacement every time removes opportunities to practise the decision.

Fade the relevant reasoning support when the learner provides evidence of readiness, and restore it when the task changes substantially. Keep access support where it is still needed. A student can become more independent in judging evidence while continuing to use text-to-speech or an adapted input method.

Contents · Previous chapter · Next chapter

CHAPTER 14 OF 19 · Support teachers and families

14. Protect teacher time by reviewing the right things

Back to contents

Before using AI feedback with a class, test it on a few teacher-created or appropriately approved examples with known strengths and errors. Include a sound paragraph as well as a flawed one. Check whether the system invents problems, misses a consequential error, misreads the sources or recommends a change outside the learning goal. A polished response to one sample is insufficient evidence of dependable behaviour.

For a manageable first lesson, use a shared task and a short set of teacher-checked comments. Learners can select the comment that applies to their work and show why. This reduces variation while the class learns to interpret feedback. It also makes the useful and unsupported comments in the worked example available without requiring every child to hold a live tool account.

Where individual AI comments are used, define the review route. A small sample can help a teacher monitor a low-stakes practice activity, but sampling does not certify every unseen comment. If errors could materially misdirect learning, narrow the task or check all advice before use. Decisions about formal grades need the school’s applicable assessment arrangements, not an unchecked model judgement.

Keep a small reusable bank of verified feedback patterns, then connect them to actual passages. Update or discard a pattern when it does not fit. Record preparation, checking, student clarification and follow-up time, including time spent correcting bad advice. The relevant workload comparison concerns the whole feedback process. Faster comment generation alone does not establish a saving for the teacher or the class.

Use the school’s approved arrangements for tools and student work. A common fictional extract can support the initial lesson without uploading real drafts. When individual work is needed, include only the material relevant to the feedback task and follow the agreed access and retention arrangements. This keeps the activity focused and avoids creating unnecessary copies of students’ work.

Contents · Previous chapter · Next chapter

CHAPTER 15 OF 19 · Support teachers and families

15. Preserve access while keeping the learner’s reasoning visible

Back to contents

Offer comments in a form the learner can use: clearly formatted text, an accessible audio version, a teacher explanation, or an appropriate translation where this supports the task. Confirm that a simplified or translated comment preserves the intended distinction. “Some respondents” should not quietly become “most students”. Access work should make the information available without changing its meaning.

Keep the common intellectual goal visible across routes. A student receiving spoken feedback can still identify the unsupported claim and choose a revision. A student using a prepared comment sheet can still compare advice with the source pack. Neither route, by itself, demonstrates skill in operating a live AI interface; that is a separate objective if the course includes it.

For a learner using a scribe or speech-to-text, clarify who selects the claim and wording relevant to the assessed goal. Mechanical recording support need not remove the learner’s authorship. Supplying the argument would change the evidence available about independent reasoning. Formal accommodations should follow the relevant school or assessment arrangements.

The educational access article develops practical participation routes. In this feedback lesson, the teacher can record a short note about substantive assistance. The purpose is to interpret the work fairly and plan a next opportunity, not to rank students by how little support they use.

Contents · Previous chapter · Next chapter

CHAPTER 16 OF 19 · Support teachers and families

16. Give parents a useful role without rewriting the work

Back to contents

A parent can help by asking the child to choose one comment and explain it. “Show me where this applies” and “What will you change?” leave the decision with the learner. If the child is unsure, the parent can help locate the task instructions or write down a question for the teacher. Neither requires turning the paragraph into adult prose.

Agree on the kind of help being offered. Reading a difficult comment aloud differs from supplying the argument. Helping the child organise a short revision period differs from repeatedly prompting until the expected answer appears. A simple note about substantial help allows the teacher to interpret the resulting work more accurately.

Parents can also ask what the child chose to retain and why. This makes room for a clear sentence to stay clear, even when a tool suggests longer vocabulary. Avoid treating every model preference as a requirement. When a comment appears to conflict with the teacher’s instructions, keep the relevant wording and seek clarification rather than commissioning another complete rewrite.

Finish with one small explanation of the change, then allow the task to end. A child does not need an endless cycle of comments to demonstrate care. If independent reasoning remains difficult, the useful message to the teacher is the specific point of difficulty, not a flawless paragraph produced through unrecorded adult intervention.

Contents · Previous chapter · Next chapter

CHAPTER 17 OF 19 · Support teachers and families

17. Handle disagreement and emotion through specific dialogue

Back to contents

Feedback can concern work the learner values. Make the discussion about a changeable feature rather than an identity judgement. “This sentence extends beyond the exit cards” gives a clear focus. “You are careless” assigns a personal label and offers no repair. Invite the learner to ask for clarification without having to defend their general ability.

If two comments conflict, return to the task and evidence. One reviewer might prefer a stronger opening; another might want greater caution. The student can preserve a clear recommendation while qualifying its scope. A teacher can model this reconciliation and acknowledge when their own earlier comment was too broad. Authority should not make a factual error immune from correction.

A student may need a pause before deciding what to do. Build a calm opportunity to return to the comment and choose a first action. Do not equate immediate agreement with understanding or hesitation with unwillingness. Ask what is unclear: the language, the purpose, the evidence, or how to carry out the change.

AI tone can be friendly while the advice remains wrong. Equally, a correct point may be expressed poorly. Teach learners to examine both the substance and the delivery. They can request a clearer explanation while still checking the underlying claim. Teachers remain responsible for making the learning relationship workable.

Contents · Previous chapter · Next chapter

CHAPTER 18 OF 19 · Review and build the routine

18. Evaluate the practice without confusing its measures

Back to contents

Consider an invented review of a class activity involving twenty-four learners. Eighteen make a justified revision during the supported lesson. On a later task, fourteen demonstrate the target reasoning independently, four do so after a reminder, and six still need teaching. These are illustrative counts, not evidence that the routine produces those results.

The two totals answer different questions. Do not subtract fourteen from eighteen and declare that four learners lost the skill: the identities and earlier performances must be linked before making an individual claim. A student who struggled during revision might understand after subsequent instruction. Attendance, available time and task demands also need consideration.

Review the comments separately. Record whether a sample is accurate, relevant to the criterion, understandable and usable. Then inspect student decisions and revisions. A high rate of accepting suggestions is not automatically desirable if some advice was unsound. A well-reasoned rejection should be recognised as a different kind of successful response.

Keep the evidence proportionate: a brief initial attempt, one consequential revision, a short explanation and a later task may be enough for a small classroom inquiry. Compare workload and access alongside learning. If introducing AI changes the task, support and time simultaneously, an improvement cannot confidently be attributed to AI alone. Use the observations to adjust practice, and avoid presenting a classroom trial as a controlled study.

Contents · Previous chapter · Next chapter

CHAPTER 19 OF 19 · Review and build the routine

19. Build a feedback routine learners can gradually own

Back to contents

Begin with a task whose evidence and criteria the teacher can check. Model one sound comment and one problematic suggestion, then let students explain the difference. Give time for a focused revision and a short account of the decision. Schedule another opportunity to use the underlying idea before treating the lesson as complete.

In subsequent lessons, let learners identify the part they want reviewed and make a first judgement before receiving advice. Their request can become more precise: “Check whether my conclusion goes beyond these sources” is useful when the learner understands what that means. The request itself becomes something the teacher can discuss and improve.

Stop generating comments when the next useful step is already clear. If the learner needs subject instruction, provide it. If the current draft meets the intended goal, move to a fresh application or a different goal. Continuing to polish the same paragraph can conceal the absence of a plan for further learning.

The durable aim is a learner who can notice a problem, evaluate advice, make a justified change and recognise when help is needed. AI can participate in that process through checked observations and questions. Teachers and parents make its value visible by watching what the learner chooses, understands and later carries into new work.

Contents · Previous chapter · Continue the series

Sources and scope

Continue the AI, Education and Civilisation series

Follow the argument from human learning into the institutions that shape shared life. Each article develops its own questions through practical examples and clear evidence limits.

Learning and human capability

Culture and shared life

Knowledge and public institutions

Economics, organisations and work

Learning design and participation

Assessment, feedback and school practice

Return to How Super Intelligence Works.