In practice
How to Collect Real-Time Student Feedback During a Lecture
End-of-semester evaluations arrive months after the lecture they describe, from a self-selected minority, about a cohort you will never teach again. Real-time feedback is a different instrument entirely, and the difference that matters is not speed. It is whether you can still act on it.
By Alae Belaich9 min read6 sourcesLast modification 13 August 2026
What counts as real-time feedback?
| Instrument | Arrives | Can you act on it? |
|---|---|---|
| End-of-semester evaluation | Months later | For a future cohort |
| Mid-semester survey | Weeks later | Later in the same course |
| In-lecture check | Seconds later | In this lecture |
Only the third changes what happens to the students who gave it. That is the entire argument for doing it.
What to ask
The quality of real-time feedback is set almost entirely by the question. Four formats, in ascending order of usefulness.
1 · The comprehension check (diagnostic)
A multiple-choice question with plausible wrong answers, tied to the concept you just taught.
"Which of these two methods applies to case B, A or B?"
This is the highest-value format because the wrong answers are informative. If 40% choose the distractor built on the classic misconception, you have learned something specific and can reteach it immediately.
Aim for 30 to 70% correct. Everyone correct means the question was too easy to be diagnostic, a point Crouch and Mazur make about peer instruction for the same reason.
The confidence check (calibration). Ask for the answer and how confident they are. The dangerous cell is confident-and-wrong, which is a misconception rather than a gap, and it will not fix itself.
Which open questions are worth asking?
The muddiest point (open, diagnostic). > "In one line: what was least clear in the last twenty minutes?"
Free text, so you learn things you did not think to ask about. Slower to read, and the only format that surfaces the unknown unknowns.
The mood check (sentiment). > "Are you following?" or a reaction button
The weakest format, and the most commonly used. It produces a sentiment, not a diagnosis: you learn the room feels lost without learning where or how many. Fine as ambient signal. Not a substitute for a question with an answer.
What should you not ask?
"Any questions?" is structurally broken. To answer, a student must decide their confusion is legitimate, articulate it and accept the room's judgement, in seconds and in public. Over half decline all semester.
"Does that make sense?" is answerable by a nod from a room that does not understand. It measures politeness.
"Put your hand up if you're lost." asks students to publicly self-identify as the ones who didn't follow. You will get the bravest few, and the interview research suggests they are not the ones most in need.
When should you ask?
- After each concept, before moving on. The point of real-time feedback is to act on it; asking after you have already built the next idea on top wastes it.
- At transitions. Szpunar, Khan and Schacter (2013) inserted brief tests through a 21-minute lecture: students tested at intervals were half as likely to report mind-wandering and took three times as many notes. The checks buy attention as well as information. Caveat: video lecture, laboratory setting.
- Before the last third in a long block, while there is still time to change course.
- Not only at the end. A muddiest-point at the end informs next year. That is worth doing, and it is not real-time feedback.
Why is anonymity the mechanism?
Cooper et al. (2021) surveyed 417 undergraduates in large science classes: over half reported never asking or answering a question in a semester, with fear of negative evaluation by peers dominating.
So a feedback channel that identifies the respondent is a channel the majority will not use. It will still return data, from the confident third, and that data will look like a room in better shape than it is.
Anonymity does not encourage participation. It removes the specific cost that prevents it.
The trade-off, stated honestly: anonymous feedback makes it harder to follow up with an individual student who is struggling, a limitation the comparative work on anonymous feedback systems treats as the real cost of the design. For diagnosing a room mid-lecture, the trade is worth it.
What decides whether this works?
Act on it visibly, or stop collecting it.
A room that answers a comprehension check and watches you move on regardless learns, correctly, that answering changes nothing. The second check will get less honest answers and the third will get fewer.
Acting visibly can be small:
- "About a third chose B, so let's go back to why that's tempting."
- "That's split, so I'll do another example."
- "Nobody picked C, so I'll skip the caveat I was going to give."
Ninety seconds of visible response makes every future check more informative.
What the evidence does not show
- That real-time feedback raises grades. Hunsu et al. (2016), 53 studies and more than 26,000 participants, found a small effect on cognitive outcomes and a near-medium effect on engagement and participation.
- That gains persist. The 2025 systematic review found no included study used a delayed post-test.
- That the tool matters. A well-chosen question on paper beats a bad question on a dashboard.
Frequently asked questions
How do you collect real-time feedback during a lecture?
Ask a question with an answer, take a response from every seat rather than from volunteers, and read the result while you can still act on it. The four useful question types are a comprehension check, a confidence check, the muddiest point, and a mood check, and only the first two tell you what to reteach.
What is the best question to ask mid-lecture?
A comprehension check tied to one concept, with plausible wrong answers rather than obvious ones. Aim for roughly 30 to 70% correct: a question everyone answers correctly measures nothing. The value is in the distribution, because a concentration on one distractor names the misconception and tells you what to address.
Why does the feedback have to be anonymous?
Because the barrier is peer judgement rather than effort. Over half of students in large science courses never ask or answer a question all semester, driven by fear of negative evaluation by peers (Cooper et al., 2021). A named response system collects data from the confident minority, so the measurement stops representing the room you are trying to read.
Is "does everyone understand?" useful feedback?
No. It can be answered with a nod by a room that does not understand, and it asks students to assess something they are often least able to judge when they understand least. Replace it with a question that produces a distribution, such as which of two methods applies to the case on the slide.
What should you do with the answers?
Act on them visibly, at least sometimes. Students work out very quickly whether their responses change anything, and a check that is collected and never mentioned teaches the room that the exercise is decorative. Reteaching the weak point immediately, or opening the next session with the top two muddiest points, is what keeps response rates up.