What Should Faculty See When Students Talk to the AI?
Every AI tutoring product on the market makes a quiet decision about how much of a student's conversation a faculty member can see. Most of them ship one of two defaults: total transparency or total privacy. Both are wrong.
The argument for full transparency is intuitive. If students are using AI on their coursework, instructors need to know what's happening. The argument for full privacy is just as intuitive. Tutoring conversations have always been confidential — we don't bug human tutors, and we shouldn't bug the AI ones.
Both arguments are missing something.
The two failure modes
Total transparency turns the AI into a surveillance device. The student stops asking the questions they actually need to ask, because they know everything is being read. The conversation becomes a performance of competence rather than the messy, halting, productive work of learning. We've seen what this does — it's the same dynamic that makes some students sit silent in office hours rather than admit they don't understand the prompt.
Total privacy creates a different problem. The instructor has no idea what's happening in their own course. They can't tell whether their lecture on equilibrium worked, whether the homework set is too hard, whether half the class is stuck on the same misconception, or whether the AI is leading students somewhere unhelpful. They lose the most valuable signal a teacher has ever had access to: a real-time map of how their class is actually thinking.
The instinct to default to one extreme or the other is a policy instinct. The interesting work is a design instinct — figuring out what kind of visibility is useful for teaching without being corrosive to learning.
Aggregate signal, not transcripts
The shift we've made in our own product is to treat the conversation log as raw data, not as something faculty should ever read directly.
What instructors should see is the layer above it: which concepts students are getting stuck on, which prerequisites the class is missing, where misconceptions are clustering, where the AI is being asked the same question twenty different ways. That's the signal that helps a teacher teach. None of it requires reading what any individual student typed.
When we build instructor views, we ask one question of every piece of information we're about to surface: does this help the teacher teach better, or does it only help them evaluate students? The former is pedagogy. The latter is surveillance with a gradebook attached.
When individual visibility actually matters
There are two cases where individual-level visibility is the right call, and both are narrow.
The first is when a student opts in. A struggling student should be able to share a conversation with their instructor — the same way they might bring a worked problem to office hours. That's a feature, not a setting. It belongs in the student's hands.
The second is when there's a credible academic-integrity concern, and it should look like every other integrity process: a documented request, a defined scope, and a human in the loop. Not a dashboard the instructor browses on Sunday night.
Anything outside those two cases is a design failure dressed up as oversight.
Why this is a design problem
You can write a privacy policy that says all the right things and still build a product that does the wrong one, because the things faculty actually look at are determined by what's on the screen. Every default is a position. Every dashboard is a claim about what teaching is.
The schools we work with have figured this out. They're not asking for transcript access. They're asking, "Help me see where my students are stuck." That's the right ask. It's the design question we should be answering — not the policy one we keep arguing about.
What we built
LabNotes shows instructors a class-level concept map: which topics students are converging on, which ones they're working through, and which ones the AI keeps getting pulled back to explain. No transcripts. No "what did Sarah ask the bot." Just a faithful picture of where the class is, drawn from thousands of conversations the students get to keep private.
If you teach chemistry and you want to see what that looks like in your own course, reach out about a fall pilot. We'd rather show you than describe it.