Lesson Feedback: What Humans Can Learn From AI
- Last updated on: 20th July, 2026
- IRIS Connect
Teachers describe the feedback they receive from our AI Insights tool in words that, if we are honest, should give us pause. “No pressure, calm, naturally me.” “Reduction in anxiety.” “Supportive.” These are lovely things to hear. But there is an uncomfortable question hiding inside the praise.
AI Insights is automated. A teacher records a lesson, clicks a button, and software returns an analysis. If that feels more supportive than what teachers usually get from the people around them, that should make us wonder about the feedback they normally receive. We think it tells us something useful, and that humans have a good deal to learn from how the AI delivers it.
Why Feedback So Often Misses
The research here is sobering, and it has been for a long time. Kluger and DeNisi’s 1996 meta-analysis of more than 600 effects found that feedback improved performance on average, but that over a third of feedback interventions actually made performance worse. Nearly thirty years on, the broad lesson still holds: a 2020 meta-analysis of 435 studies concluded that feedback cannot be treated as one consistent thing, and that what decides its effect is the information it carries. Feedback became less effective the more it drew attention towards the self and away from the task. Criticism that lands on the person, rather than the work, is the kind people stop listening to.
The lesson observation tradition has often pulled in the wrong direction too. Professor Robert Coe’s analysis found that even highly trained observers could tell an above-average teacher from a below-average one only around 60 per cent of the time, against 50 per cent for a coin toss. That unreliability helped push Ofsted to drop graded observations, and it has confirmed they will not return. Where a grade is attached, teachers fixate on it rather than the development points, and a teacher told they are “good” assumes there is nothing left to do.
This is not the fault of those giving feedback. It is a problem of skill and conditions, and skill can be taught.
The Basics Everyone Should Learn
The structure behind AI Insights is plain. It observes, stating what happened in factual terms. It affirms, naming a genuine strength. Then it offers an “I wonder”, a prompt to consider rather than an instruction to obey. As one teacher put it: “It stated minute by minute what I was doing. It’s very factual.”
Each move maps onto what the evidence says works. Factual observation keeps attention on the task, where feedback helps. Affirmation lowers the defensiveness that makes feedback bounce off, and teachers are most receptive when it feels nonthreatening and respectful. The “I wonder” preserves agency, inviting reflection instead of dictating a fix.
None of this requires artificial intelligence. We regularly hear from teachers that it can feel intimidating to ask a colleague for feedback in an area where you feel vulnerable. The structure above is simply a disciplined way to remove that fear. Observe before you judge. Affirm something real. Replace the verdict with a question. These are learnable basics, and every school could adopt them tomorrow.
The Bigger Story: Reflection, Coaching, and Back Again
Reflection has always been recognised as powerful. Dewey’s observation that we learn not from experience but from reflecting on experience has shaped professional learning for decades. Yet reflection on its own has a poor track record, because simply asking teachers to reflect does not produce results, and the instinct for it is far from universal. Research on trainee teachers points the same way: reflection deepens only with the right scaffolding. It is knowing how to reflect, and against what criteria, that turns looking back into genuine learning. The practical barriers compounded it: rarely time to trawl through a whole recording, if one existed, and rarely a clear sense of the next step or what better looked like.
Instructional coaching rose to solve exactly these problems. In place of the daylong seminars that research found had little effect, coaching offered a steady stream of feedback from an expert who could model new techniques and suggest concrete next steps. It supplied the focus, criteria, and direction that solo reflection lacked, and the evidence for it is strong.
But coaching carries a hard ceiling of cost and scale. A major meta-analysis of coaching by Kraft et Al. found that the average effectiveness of a programme declines as more teachers are involved, because quality coaches are hard to recruit and retain, and financial pressure thins the support until much of its power is lost. Not every teacher in every trust can have regular, high-quality coaching. For many, the honest alternative to a coach is not a slightly worse coach. It is nothing.
Why Reflection Works Again Now
Technology has quietly removed the barriers that held reflection back. Through video, a teacher can see the lesson as it actually unfolded, not as memory reshapes it. AI Insights then highlights the key moments and pairs them with reflective prompts tailored to that teacher’s chosen development focus, so reflection finally has both a clear focus and a clear set of criteria. Our Pathways supply the rest: bite-size research summaries and technique guides that show what better looks like.
The evidence is not just theoretical. A 2025 randomised controlled trial across 224 maths and science teachers found that automated feedback increased their use of focusing questions, which press students to explain and reflect, by 20 per cent. It was the first such experimental evidence from ordinary classrooms. The effect depended on how often teachers engaged and did not transform everything at once: automated feedback is a strong input, not a finished answer.
Because the teacher chooses the focus, reviews their own lesson, and decides what to change, they stay in the driver’s seat. Hatano and Inagaki drew the distinction that matters here: routine experts perform familiar procedures quickly and accurately but struggle to adapt to the unfamiliar, while adaptive experts understand why their methods work and when to flex them. Development done to a teacher risks building the former. Reflection done by the teacher builds the conceptual understanding and self-regulation behind the latter, which is the goal.
A Better Feedback Culture, and an Honest Alternative
We usually say that our AI Insights support the human conversation rather than replacing it. Where skilled coaching exists, that is exactly right. A trusted colleague who knows the class and the teacher will always offer something software cannot, and one of our schools described AI Insights as a springboard for deeper conversations with their coaches. That partnership remains the gold standard.
But for a great many schools, that gold standard is out of reach: regular high-quality feedback for every member of staff is neither affordable nor available. In that reality, AI-guided reflection is not a poor substitute. It is a genuinely strong developmental input, and far better than feedback that is rare, generic, or absent.
So our call is twofold. First, let us raise the standard of human feedback in every school. The basics are learnable, and they cost nothing: observe before you judge, affirm something real, and trade the verdict for a reflective question. Second, let us be clear-eyed about what is now possible. Where you cannot yet give every teacher regular, high-quality feedback from a person, AI-guided reflection can put them in the driver’s seat of their own development today.
To see how AI Insights and Pathways support teacher-led reflection across a school or trust,
Get blog notifications
Keep up to date with our latest professional learning blogs.
By clicking Subscribe you consent to us contacting you regarding our products and services, as well as other content that may be of interest to you. You also consent to us storing your personal data to provide you with the content you are requesting. You can unsubscribe at any time. To find out more about how we manage your data, please review our Privacy Policy.