Students can now generate a plausible paragraph on any topic in seconds. The paragraph looks like knowledge. It is not. Source hygiene is the classroom practice of asking, every time, where a claim came from and how much weight it can carry.
Start with a definition students can actually use
A source is the specific place a claim came from. Not "the internet," not "a chatbot," not "an article I read." The actual URL, book with page number, interview with a named person, or dataset with a version. If a student cannot name the source, the claim is not yet ready to be used in an argument.
This is not new. Historians, scientists, and journalists have taught this for a century. What is new is that the surface layer of student writing can now be produced without ever touching a source at all. So we have to make the tracing step visible again.
Teach four evidence classes as the shared vocabulary
Give the class one shared vocabulary for how much a claim is worth. We use a small set adapted from research practice (Class E):
- Class A: the student observed it themselves in this session (a measurement, a lab result, a lived experience with a date).
- Class B: it is in the code, the primary text, or the raw document, and the student can point to the line.
- Class C: it is in a configured system or integration, verifiable by inspection.
- Class E: an expert cited it in a named, findable publication.
- Class F: the student states what would prove the claim wrong, and looked.
- Class U: unverified. The student believes it but has not checked. Class U is allowed. It is not allowed to hide.
When a student writes a paper, every non-obvious claim gets a tag. A student who writes "sharks have existed for over 400 million years (Class E: Klimley 2013, p. 4)" is doing history-of-science work. A student who writes "sharks are misunderstood (Class U)" is being honest about what they are still checking.
The chatbot changes what needs tracing, not whether
A generative model can produce a sentence that reads like Class E and is actually Class U. That is the whole hazard. The practice that survives this is simple: a chatbot output is never itself a source. It is a lead. The student has to follow the lead to the actual paper, dataset, or person, then cite that. If the lead does not resolve to a real source, the claim goes back to Class U or gets cut.
We pair this with the AI-authorship fence: if a generative tool touched the draft, the student declares it, names the tool, and describes what it did. That declaration is part of the assignment, not a confession (Class C: this is wired into our assignment templates and the workbench sign-off gate).
Three concrete classroom moves
One: the trace-back drill. Give students a paragraph that looks authoritative. Ask them to find the underlying source for each factual claim. Some will resolve to a real paper. Some will resolve to nothing. That is the lesson.
Two: the class-tagging pass. Before a paper is turned in, students annotate their own draft with the evidence class of every claim. A paper that is mostly Class U is not failing. It is a paper that has not finished its work yet.
Three: the falsifier line. For the central claim of any argument, the student writes one sentence: "This claim would be wrong if..." That is a Class F move. It teaches that a claim you cannot imagine being wrong is not yet a claim you understand.
What this is not
This is not a way to catch cheaters. It is a way to teach the underlying skill that the chatbot exposes: knowing what you actually know, and being able to show your work. A student who uses a generative tool and then does the tracing work honestly is doing the assignment. A student who copies without tracing is not, whether the copy came from a chatbot, a sibling, or a website in 2004.
Source hygiene is a habit, and habits are built by doing the small move every day, in every subject, until it becomes the natural shape of the thought.
Next steps
- Give your class the declaration language: Student declarations, a template
- Put the vocabulary on the wall: Evidence classes as classroom vocabulary
- Handle the live case: When a student uses a chatbot, what to do
- Bring the practice into your school: /workshop