Beyond AI detection
Treat a detector score as a reason to talk, not a finding.
The short answer
A detector score is a guess about the final text, and it can be wrong. Before you act on one, talk with the student and look at how the work was made. For the next assignment, ask for drafts and a short explanation, so you have evidence instead of a guess.
A guide from Edudojo, an AI that won’t think for students: its coach asks questions instead. The advice works with or without it.
On this page
Why a detector score is not enough
AI-writing detectors estimate how likely it is that a text was produced by AI. They can be wrong in both directions, and their accuracy can vary with the language, the amount of editing and the model used.
Detection treats integrity as sorting: this text is AI, that one is human. The sorting breaks down when AI use is allowed, when a student edits heavily, or when what you want to assess is the thinking. Before any score is used in an investigation, review the tool’s evidence and limits, and give students a way to challenge errors.
A detector reports
A likelihood score (a guess) that the final text is AI-written.
Looks at: the final text only.
A process record shows
- Draft 1 saved at a checkpoint
- Coach asks a question about one sentence
- Draft 2 the sentence, changed after the coach's question
- How I worked the student's optional declaration at Submit, picked from a short list
Looks at: how the work was made.
- Claim not true: This proves who wrote it
- Claim not true: This measures learning
Neither tool can. Both are reasons to talk with the student.
If a piece of work already worries you
- Talk before you decide. Ask the student to walk you through one paragraph and one source.
- Look at what exists. Drafts, version history, notes, earlier work.
- Ask for the reasoning in a new setting. A few questions in class, or a short rewrite of one section.
- Follow your institution’s process. Do not let a score stand in for evidence.
What integrity can mean when AI is allowed
A durable approach asks questions you can answer from evidence:
- Did the student work with the learning goals?
- Can they explain and defend their reasoning?
- Did they check what an AI tool told them before using it?
Integrity then follows from visible process rather than hidden detection.
Rules also have to fit the task: if a task can be done by pasting a prompt into a chatbot, “AI is forbidden” invites dishonesty without changing what students learn.
Alternatives that work together
None of these replaces careful assignment design on its own. Most colleges combine several.
- Redesign the task. In-class parts, local or unique data, and formats where the student explains.
- State the AI rules openly. Allowed uses and what to declare, task by task. See the AI-use policy guide.
- Ask for the process. Drafts and a short note on how the student worked. See process-based assessment.
- Teach checking. Give students a routine for testing AI claims before they use them. See checking AI answers.
- Keep supervised checks where they matter. Oral defences and in-class writing are direct, but they do not scale to every assignment.
What each kind of tool records
| Aspect | AI detector | Draft-history tool | Edudojo |
|---|---|---|---|
| Looks at | The final text | Edits to one document over time | Saved drafts and the coach chat, inside the assignment |
| Gives you | A likelihood score | A timeline of edits | A read-only replay for the assigned teacher, with no score |
| When AI is allowed | Conflicts with permitted use | Misses AI used in another tab | Built for AI use inside the assignment |
| Known limits | False positives and negatives | The thinking often happens outside the document | Sees AI use only inside the workspace; misses other tabs too |
You can use them together. For an ordinary essay written where AI is not allowed, a draft history may be enough. For AI-heavy tasks, the questions and answers around the draft matter more. Some colleges keep detectors for tasks where AI is banned; if you do, treat a score as a reason to talk, not as a finding.
Students write with the Socratic Coach, practise checking with the Red Team, and the teacher reads the Replay after submission.
What we do not claim.
- No authorship check
- Edudojo’s record does not prove who wrote the work. It is not a detector.
Try it this week
When a piece of work worries you, ask the student to walk you through one paragraph and one source before you decide.
Trying this with a college? See the pilot plan