UsewrightA brief for instructors · usewright.org

Students practice judging AI on your real assignments

AI can now produce most of what an assignment asks for, so students can finish the work without building the skill it was meant to build, often without noticing. Rules alone do not fix this, and detectors make mistakes and breed suspicion. Usewright takes a different path: students practice deciding, task by task, when AI helps their learning and when it gets in the way.

What it is

Not a detector and not a chatbot. It never watches or scores AI use.
Works with any AI tool, and with courses that allow no AI at all.
Generates nothing for students. Its value is what they can explain afterwards.
Free while in testing.

What students do, in about three minutes

Plan. Before the work, the student chooses to use, limit, or decline AI, and says why.
Try first. They name the part they will attempt on their own before opening AI.
Do the work. Anywhere, with any AI tool or none. Usewright never sees the work.
Explain without AI. They explain it in their own words and see whether their confidence held.
Revisit. A week later, a one-minute check: can they still explain it?

Your setup, in about five minutes

Paste the assignment. Usewright fills in the title, due dates, stages, and AI rule for you to check. The pasted text stays in your browser.
Use stages for projects. Up to four, such as design, build, and demo, each with its own due date and AI level.
Set the AI level for each task, from No AI to AI encouraged, and name the one skill that must stay the student's own.
Paste one link into Canvas. Each task has a link that signs students in, joins them to your course, and opens that task.
Adopt a syllabus statement. Four model policies, written in plain language, free to use or edit.
Link your institution's policy once, and students see it on every task.

Where it fits

Any discipline: lab reports, problem sets, code, design projects, essays, and clinical communication. Beyond courses, it also supports research supervision, project teams, and co-authored papers, with AI statements in the formats journals ask for.

What you see

Class patterns, never individuals. Counts appear once at least 10 students respond to a task: how many planned, followed their plan, checked AI output, disclosed it, and could explain the work.
Plain suggestions, such as “most students kept everything AI gave them; ask them to test two suggestions against the rubric.”
An oral check kit: the same short questions for every student, with a shared rubric, if you want to hear their reasoning.

If you prefer no AI

Set tasks to No AI. Declining AI for a stated reason is a full choice in Usewright, not a gap. Students commit before the work, keep the named skill their own, and explain it in their own words, all without detectors.

What students get

A ready first message that sets an AI tool's role from their own plan.
A three-part check on their ownership of the work: who set the direction, what they questioned, and whose reasoning it is.
A disclosure statement written for them, and practice questions to answer out loud.
A private record of their AI judgment they can share with a program or employer.

Trust and privacy

No work is uploaded; Usewright stores only what people type into it.
Privacy rules are enforced in the database, not just on screen.
Leads agree not to use entries as evidence in integrity cases.
Nothing is sold or used to train AI. Anyone can delete their data.

The evidence behind it

Unrestricted AI help raised practice scores but lowered later exam results in a large field experiment, while guardrails largely prevented the harm (Bastani et al., 2025). Explaining in your own words deepens understanding (Chi et al., 1994), people often overrate how well they understand until they try to explain (Rozenblit & Keil, 2002), and recalling after a delay strengthens memory (Roediger & Karpicke, 2006). In early testing, nine testers rated how likely they were to use Usewright again at 4.0 out of 5; a larger round is under way.

Try it

Open usewright.org, create a course space, and use “Try it yourself” to see the student side. Questions, pilots, or a security review: hello@usewright.org.

Bastani, H., et al. (2025). Generative AI without guardrails can harm learning. PNAS, 122(26), e2422633122. · Chi, M. T. H., et al. (1994). Eliciting self-explanations improves understanding. Cognitive Science, 18(3), 439–477. · Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning. Psychological Science, 17(3), 249–255. · Rozenblit, L., & Keil, F. (2002). The misunderstood limits of folk science. Cognitive Science, 26(5), 521–562.