Student feedback loops: the curriculum your students are already writing.

The best curriculum research you will ever get arrives one email at a time.

Students tell you what the course is missing constantly, in the ordinary business of getting stuck. Almost all of it is thrown away within the hour, because answering a question feels like finishing it.

That is the whole problem, and it is a filing problem rather than a listening one. A question answered in a reply has helped one person; the same question written down beside the lesson it belongs to eventually helps everybody who ever takes the course. The difference between those two outcomes is about eight seconds of typing at the moment the question arrives — and nothing else in course maintenance has that return.

1. 01The live room, if you have one. The richest surface, because people ask in front of each other and the second person says “I had that too”. The whole value leaks away unless somebody writes questions down during the session rather than afterwards — office-hours structure and group call structure both build the queue that makes this automatic.
2. 02Email and direct messages. The highest-signal channel and the one most completely lost. Somebody describes exactly where they got stuck, you write a thoughtful reply, and the whole exchange evaporates into a thread nobody will read again. One rule fixes it: never answer a student question without logging it first.
3. 03The support or billing inbox. Mixed traffic, and the curriculum signal hides inside tickets that look administrative. “How do I get back to the part about pricing” is not a support issue; it is a navigation problem in your course, arriving disguised as a question about logging in.
4. 04A question field inside the lesson itself. The closest thing to the point of confusion, and the most under-used surface on this list. A lesson with nowhere to ask does not generate fewer questions — it generates the same number, silently, and you never see any of them.
5. 05Replies to the onboarding sequence. Ask one question in the welcome email and you collect what people wanted before they started: the prerequisites you assumed, the outcome they think they bought, and the words they use for the problem. Those words are also the best sales copy you will ever be handed.
6. 06Refund, cancellation and offboarding replies. The hardest to read and the most honest. One open question in the message that confirms a cancellation will tell you more about the course than a year of praise from the people who finished it.

If you do not run a community

Most course businesses do not, and the advice to “watch what people post” is useless to them. Without a community every question arrives privately and one at a time, so nobody ever says “I had that too” in public and no thread accumulates. The consequence is not that you get less feedback — it is that the pattern is invisible unless you write each one down. Logging is therefore the entire system for a solo course, rather than a supplement to a forum.

One thing worth adding on purpose: a question field inside each lesson. It costs nothing, it collects confusion at the exact point it occurs, and it converts a private email you might never receive into a logged entry attached to the right lesson automatically.

Five columns. Nothing more elaborate survives a busy month.

Every course maintenance system that fails fails the same way: it was too well designed to keep up with. A spreadsheet with five columns is not a compromise — it is the thing that still exists in November.

Three types, three destinations

The Type column does more work than the other four combined, because it stops the log turning into an undifferentiated pile that is easier to ignore than to read.

• →Curriculum — the teaching did not land. The person watched the lesson and still cannot do the thing. This is the only type that becomes a curriculum change, and it is the one worth counting carefully.
• →Administrative — the course is hard to use. Access, billing, where to find the recording, when the deadline is. Real problems, but they belong in the onboarding sequence or a help page, not in a lesson. Adding them to the curriculum is how a course grows a second, worse course underneath itself.
• →Scope — “do you cover X?” Not a gap in the teaching; a gap between what they expected and what you promised. A handful of these means the sales page needs a line. A lot of them, all naming the same X, means you have found your next module or your next product.

Read the log in one sitting once a month. The single sitting is the point: three occurrences met one at a time across nine weeks feel like three unrelated conversations, and the same three read together are obviously one broken lesson.

A repeated question is a missing lesson. Decide where “repeated” starts.

Without a stated threshold, curriculum changes get made on whichever question annoyed you most recently. With one, the log makes the decision and you only have to do the work.

1. 01Asked once — answer it well, log it, and wait. One person is not a curriculum signal. It may be them, it may be their situation, it may be a bad afternoon. Attach the answer to the lesson it belongs to and let the count decide whether it is anything more.
2. 02Asked three times — the lesson is unclear. Three separate people hitting the same wall is the lesson failing, not the students failing. Rewrite or re-record the lesson. Adding another entry to an FAQ instead is how the real answer ends up somewhere only the most diligent student will ever find it.
3. 03Asked by nearly everyone who reaches it — the lesson is missing. Something the curriculum assumes and never taught. Build it, and place it before the step where people get stuck rather than after it, which is where the FAQ instinct would have put it.
4. 04Clustered in one module — check sequencing before difficulty. Either the genuinely hard part of the subject, which is fine and expected, or an ordering problem. The questions themselves usually tell you which: “I do not understand this” is difficulty, “what am I supposed to do now” is sequence.
5. 05Never asked about — check who reached it first. Nobody asks questions about a module they never opened, and silence from an untouched module is indistinguishable from silence from a perfect one until you look at how many people got there.

Why three, and why not wait for ten

Three is a working number rather than a discovered one, and the reasoning behind it matters more than the figure. Most people who are confused never ask — asking costs effort, exposes you slightly, and requires believing an answer will come. So every question in your log stands in for some unknown number of people who quietly closed the tab. Since you cannot know that multiplier, waiting for a large sample before acting is not rigour; it is a way of never acting. Three separate people, inside one window, is early enough to be useful and late enough not to be noise.

Scale the window rather than the count. In a twelve-person cohort three questions in six weeks is a strong signal about that module. In a self-paced course with hundreds of students moving at different speeds, use a longer window and pay more attention to which lesson the questions attach to than to the raw total.

The feedback from the people who never wrote to you.

Questions come from engaged students. Drop-off comes from everybody, including the ones who decided the course was not working and said nothing — which is most of them.

Before reading any of it, separate two populations that look identical in a completion figure: people who reached a lesson and stopped, and people who never reached it at all. A module with no activity is not a bad module; it may be a module nobody has arrived at yet. Conflating those two is the most common way course data gets misread, and it usually leads to rewriting something that was never the problem.

1. 01The first-lesson cliff. Bought, never started, or started and never returned. This is almost never a curriculum problem — it is the gap between purchase and first win, and it is fixed in the onboarding sequence rather than in module three.
2. 02The mid-course cliff. People reach a specific lesson and stop. Cross-reference it against the question log: questions clustering at the same point means difficulty, and the fix is teaching. Silence at the same point means the lesson is unclear about why it matters, or it is simply boring, and the fix is framing or length.
3. 03The last-mile stall. Most of the way through and stopped. Usually there is nothing at the end worth walking to — no deadline, no assessment, no artifact, no event. Completion systems and a real assessment are what supply one.

One more signal that is easy to miss: lessons people re-watch. Repeated views of the same lesson can mean it is a reference people keep coming back to, which is good, or that one pass is not enough to follow it, which is not. The question log usually tells you which, and if it does not, that lesson is the obvious candidate for the next section.

Two questions, asked at the right moment, beat any survey.

End-of-course surveys are answered by finishers. Finishers are the people for whom the course worked, which makes them the least informative group you could possibly ask about what is broken.

That is not an argument against asking — it is an argument about timing and audience. Ask inside the course at the point of friction, and ask the people who stopped, by email, one question at a time. Both produce answers from the population a completion survey structurally excludes.

The two questions worth asking

1. 01“What were you trying to do when you got stuck?” It asks about their work rather than about your lesson, which is the only version most people can answer accurately. It also produces the specific situation you need in order to write a scenario question, a worked example, or the missing lesson.
2. 02“What did you expect this lesson to cover that it did not?” This finds the gap between the promise and the delivery — sometimes a curriculum hole, and just as often a sales-page problem where the course is fine and the expectation was set wrong upstream.

Both are open questions, deliberately. At the numbers most course businesses have, rating scales produce an average of a dozen people that moves for no discernible reason; a sentence in a student’s own words is directly usable and cannot be confused for a measurement. Save the scale for when you have enough responses that the average means something, and even then keep one open box.

The email to somebody who stopped

Send it manually while numbers are small, and only to people who genuinely started and then stopped — not to everybody who has been quiet for a fortnight. One question, no pitch, and a genuine reason for asking.

Subject: One question about [course name]

[First name] — you started [module or lesson name] on [date]
and have not been back since. That is completely fine, and this
is not a nudge.

I am rewriting that part of the course this month and I would
rather fix the real problem than guess at it. If you have thirty
seconds: what were you trying to do when you stopped?

Anything is useful — “it was too long”, “I got busy”, “I could
not work out what to put in step two”. One line is plenty.

[Your name]

The reason it works is in the third paragraph: you are rewriting that part of the course and you would rather not guess. That is true, it is not a nudge to come back, and it makes replying feel like helping rather than confessing.

Ninety minutes, four times a year, with an agenda.

Capture is daily and nearly free. Reading the log is monthly and takes twenty minutes. The quarterly review is where it stops being an observation exercise and starts changing the course.

Everybody who answers student questions should be in it, which in most businesses means one person and occasionally two. Book it as a recurring appointment rather than intending to do it, because a review that depends on remembering is a review that happens once.

The cut is the item people skip, and it is the one that keeps the course teachable. A curriculum that only ever grows becomes longer, slower and harder to finish, and every quarter of additions without a subtraction moves it further from the promise on the sales page. If nothing deserves cutting, cut the lesson with the lowest completion that generates no questions at all — it is doing nothing for anybody.

Fix the module. Do not stage a relaunch.

The most common reason a known-broken lesson stays broken for a year is that fixing it has been mentally attached to relaunching the course. Separate those two and the loop starts turning.

A version is a number and a date, not an event. v2026.1 is enough — the year plus the revision — and it goes at the top of the syllabus, on any certificate you issue and in a dated changelog inside the course. That changelog is worth more to past students than any re-engagement campaign you could write, because it is evidence that the thing they bought is still maintained.

Match the announcement to the size of the change. A re-recorded lesson needs a changelog line and nothing else. A new module that changes what the course delivers is worth one email to current students. A change to the promise itself is not a version — it is a different course, and the sales page and the price have to move with it.