Decision guide · Academic integrity
AI detectors vs process evidence
AI detectors estimate whether patterns in finished text resemble machine-generated writing. Process evidence records captured activity while a document was built. They answer different questions: neither proves authorship or misconduct on its own, but a contemporaneous process record can give a teacher inspectable evidence to review alongside the final work, policy and student conversation.
Primary question
How is writing-process evidence different from an AI detector?
Search intent
Compare evidence models before choosing a school review workflow.
Different evidence
What question does each approach answer?
AI detector
“Does this finished text resemble writing produced by a model?”
The answer is an inference from the submitted text. It is not an observation of how the document was made.
Process evidence
“What recorded changes occurred while this document was being built?”
The answer is a record of captured activity. Interpretation still belongs to a person.
How should a teacher interpret an AI detector result?
Treat it as a signal for further review, not as a finding. Turnitin’s own current guide says its model may misidentify human-written, AI-generated and AI-paraphrased text, and that the result should not be the sole basis for adverse action against a student.
A detector can still be convenient when the only available artefact is a completed document. Its usefulness is triage: it can prompt a teacher to look more closely. The school still needs its policy, the submitted work, relevant context and a fair opportunity for the student to respond.
Primary source: Turnitin, Using the AI Writing Report (accessed 29 July 2026).
What does process evidence add?
It adds a contemporaneous record of the document’s construction. In Loam, that means captured edits, timing, revisions and evidence checkpoints from inside the editor. A teacher can inspect the record and replay the sequence rather than infer a whole process from the final prose.
The record is deliberately bounded. It does not see activity outside the editor, identify the person at the keyboard, establish where an idea originated or decide whether school policy was breached. Loam’s verification establishes facts about the captured record; the academic judgement remains human.
Side-by-side
How do the two evidence models compare?
Neither column is a verdict. The table describes the job each approach can reasonably do.
| Decision point | AI detector | Loam process evidence |
|---|---|---|
| What is examined? | The submitted text, after it has been written. | The document changes and checkpoints captured during a defined writing session. |
| What is the output? | A model-generated indicator about text that may resemble AI-generated writing. | A record a teacher can inspect, verify and replay in the context of the finished work. |
| Where is it strongest? | Low-friction triage when only the finished text is available. | Contemporaneous evidence for assignments planned around a recorded writing process. |
| What is the main limitation? | The indicator can be wrong and does not establish misconduct by itself. | It sees only the activity captured in the writing environment and requires the workflow to begin before submission. |
| What role should it play? | A prompt for further scrutiny under the school’s policy. | One evidence source for a teacher-led review and conversation. |
What can a writing-process record show, and what stays unknown?
It can show recorded document activity; it cannot turn that activity into an authorship verdict. The boundary matters most when the outcome could affect a student.
What the Loam record can show
- • The sequence of captured edits and revisions.
- • Timing and checkpoints recorded during the session.
- • Whether the recorded chain passes Loam’s integrity checks.
- • The final document rebuilt from the captured changes.
What it cannot prove
- • Who was physically at the keyboard.
- • Where every idea, source or phrase originated.
- • What happened outside the Loam editor.
- • Whether the evidence amounts to misconduct under policy.
School decision
How should a school choose?
Start with the decision the evidence must support. Then design the assessment and review process before choosing a tool.
- 01
Define permitted assistance
Tell students what AI, collaboration, quotations and accessibility supports are allowed before work begins.
- 02
Choose evidence before the task
If the process was not captured, it cannot be recreated reliably after submission. Decide whether the task warrants contemporaneous capture.
- 03
Minimise collection
Collect evidence that is relevant to the assessment, and test the privacy, accessibility and retention consequences.
- 04
Keep a human review
Use any signal or record alongside the work, policy, context and a fair student conversation.
Policy context: the Australian Framework for Generative AI in Schools centres responsible and ethical use for schools. A local school policy and applicable education authority requirements still govern each assessment.
Related questions
What else do schools ask?
- Does process evidence prove that a particular student wrote the work?
- No. A process record can show what happened inside the recorded writing session and whether that record remains intact. It cannot by itself establish the physical or legal identity of the person at the keyboard.
- Does Loam detect whether AI wrote a sentence?
- No. Loam does not issue an AI score or determine whether AI was used. It records document changes made inside Loam so a teacher can inspect the process and make their own decision under school policy.
- Should a school replace every AI detector with process evidence?
- Not automatically. The right approach depends on the assessment, the evidence already available, privacy and accessibility requirements, and the school’s review policy. Process evidence is most useful when it is designed into the task before students begin.
What should you read next?
What can schools use instead of detector scores?
Turn the comparison into a practical workflow using rules, staged evidence and human review.
What is proof-of-process assessment?
Start with the category, its evidence boundary and the role of human judgement.
What can a writing-process record prove?
Test each factual claim against the record's capture boundary before drawing a conclusion.
When is Google Docs history enough?
Compare the familiar classroom record with a controlled writing process.
How can a school test the workflow?
Plan a bounded pilot around one faculty, one task and one review rubric.
Which primary sources support this guide?
The comparison uses product owners’ and public authorities’ own descriptions. Sources were reviewed on 29 July 2026.
- Turnitin: Using the AI Writing Report — output, scope and stated limitations.
- Australian Department of Education: Australian Framework for Generative AI in Schools — national school context.
- Loam: verification method and claim boundary — the primary product description for Loam’s process record.
- Loam Privacy Policy — captured data, access and retention commitments.
Useful next step
What should you test before changing school policy?
Run one bounded assignment with an agreed review rubric. Check whether the record helps the teacher ask better questions without turning evidence into an automated verdict.