What can a school use instead of relying on an AI detector?

A school can combine clear AI-use rules, staged checkpoints, progressive drafts, a short student explanation and a reviewable writing-process record. These sources give a teacher more context, but the school still needs a fair human review and an opportunity for the student to respond.

Does process evidence prove that a student wrote the work?

No. A process record can show captured edits, revisions, timing and integrity checks inside the recorded environment. It cannot establish who was physically at the keyboard, where every idea originated or what happened outside the environment.

Does Loam decide whether a student used AI?

No. Loam does not issue an AI score or misconduct verdict. It records document activity inside the Loam editor and gives teachers a record to inspect alongside the finished work, school policy, relevant context and the student's explanation.

Should schools stop using every AI detector immediately?

That is a school policy and assessment-design decision. A detector may be used as a limited triage signal when only finished text is available, but its result should not become the sole basis for adverse action. Schools should decide the evidence and review process before a task begins.

Assessment design · Human review

What is a practical alternative to AI detectors?

A practical alternative to relying on AI detectors is to design assessment around observable learning: clear AI-use rules, staged checkpoints, progressive drafts, brief student explanations and a reviewable writing-process record. This does not prove authorship or eliminate misconduct. It gives teachers more direct evidence to consider with the finished work and student conversation.

Primary question: What can schools use instead of relying on AI detector scores?
Search intent: Find a fair, practical school workflow for reviewing writing when generative AI is available.

Editorial responsibility: Difinity Pty Ltd, operator of Loam

Published 30 July 2026

Last verified 30 July 2026

What can schools use instead of an AI detector?

Use several modest sources of evidence rather than asking one score to carry the whole decision. Start with explicit task rules, collect work at useful stages, and keep the finished response at the centre of assessment. When a concern arises, invite the student to explain their choices and consider only evidence relevant to the task.

Practical alternatives and the role each can play

Set clear rules before the task: Always; Students need to know which AI use, collaboration, research and accessibility support are permitted and how assistance must be acknowledged.
Use staged checkpoints or progressive drafts: Often; Plans, notes, drafts and feedback cycles make learning visible and give teachers evidence created before a concern arises.
Ask the student to explain the work: When relevant; A short, proportionate conversation can test understanding and surface context that a text classifier or edit log cannot observe.
Capture a writing-process record: For suitable tasks; A contemporaneous record can show how the document changed inside the capture boundary. It must be planned before writing starts.
Treat a detector score as a verdict: No; A classifier estimates patterns in finished text. Even the product owner warns that the result may be wrong and requires further scrutiny and human judgement.

Practical alternatives and the role each can play
Question	Answer	Evidence boundary
Set clear rules before the task	Always	Students need to know which AI use, collaboration, research and accessibility support are permitted and how assistance must be acknowledged.
Use staged checkpoints or progressive drafts	Often	Plans, notes, drafts and feedback cycles make learning visible and give teachers evidence created before a concern arises.
Ask the student to explain the work	When relevant	A short, proportionate conversation can test understanding and surface context that a text classifier or edit log cannot observe.
Capture a writing-process record	For suitable tasks	A contemporaneous record can show how the document changed inside the capture boundary. It must be planned before writing starts.
Treat a detector score as a verdict	No	A classifier estimates patterns in finished text. Even the product owner warns that the result may be wrong and requires further scrutiny and human judgement.

Why should an AI detector not decide the outcome?

A detector analyses the submitted text; it does not observe who wrote it or how the work developed. Turnitin's current guidance says its model may misidentify human-written, AI-generated and AI-paraphrased text and should not be used as the sole basis for adverse action against a student.

A detector can still be one triage signal when no process evidence exists. The result should lead to proportionate scrutiny, not an automatic conclusion. The final judgement remains with the school under its policy.

Primary source: Turnitin, Using the AI Writing Report.

What does a writing-process record add?

It adds a contemporaneous record of captured changes instead of another inference about the final prose. In Loam, a teacher can inspect edits, revisions, timing and checkpoints recorded inside the editor and replay how the document developed.

That record remains incomplete by design. It cannot see activity outside the editor or determine identity, originality, intent or policy compliance. It is useful because its capture boundary is inspectable—not because it removes the need for judgement.

See the writing and review flow Inspect the verifier

Claim boundary

What can this alternative show, and what remains unknown?

Can show

The task rules and permitted assistance communicated before work began.
Plans, drafts, checkpoints and feedback collected during the task.
Captured document changes, timing and revisions inside a process-record environment.
The student's explanation and demonstrated understanding during a human review.

Cannot prove

The physical or legal identity of the person who made every change.
That no outside assistance, second device, collaboration or unrecorded tool was used.
That a detector score or process signal establishes misconduct.
That one workflow is accessible or proportionate for every student and task.

The useful question is not which tool produces the strongest accusation. It is which combination of assessment design, evidence and human review supports a fair decision with the least unnecessary collection.

How does this fit Australian school guidance?

Australian guidance keeps responsibility with educators, schools and their policies. The national framework addresses responsible and ethical use of generative AI. NESA says schools should provide clear advice, apply academic-honesty rules and use varied assessment tasks that let students demonstrate learning in different ways.

Neither authority endorses Loam. Their guidance supports the broader principle used here: define the task, collect proportionate evidence, preserve human judgement and apply review controls fairly.

Read the official Australian Framework for Generative AI in Schools and NESA's student AI guidance.

Source ledger

Which primary sources support this page?

01
Turnitin — Using the AI Writing Report
Turnitin's current product guidance states that its AI model may misidentify text and should not be the sole basis for adverse action. Updated 6 March 2026.
02
Australian Department of Education — Framework for Generative AI in Schools
National framework for responsible and ethical generative-AI use in Australian school education. Last modified 17 June 2025.
03
NESA — Use of Artificial Intelligence by students
NSW school guidance on academic honesty, permitted use, acknowledgement, varied assessment and fair application of plagiarism controls.
04
Loam — AI detectors vs process evidence
Product-specific comparison of what a score-based classifier and a captured process record can reasonably show.
05
Loam — Security, verification and proof boundaries
Primary source for Loam's current capture boundary, verifier, synthetic sample and explicit cannot-prove statements.

Where should you go next?

Understand proof of processStart with the category definition, the evidence model and the boundary between a record and a verdict.Compare the evidence modelsSee a detailed side-by-side comparison of text classification and captured writing activity.Test each claim boundarySeparate what a process record can establish from identity, authorship and misconduct inferences.Review the Australian contextConnect national and NSW guidance with Loam's hosting, privacy and school-review boundaries.Plan a school reviewMove from the evidence model to governance, administration and a bounded school rollout.Inspect the technical recordUse the synthetic sample and local verifier instead of taking the product claims on trust.