The exact opening words that prevent an accusation or interrogation

The opening words matter more than tone. Begin with a neutral observation and an explicit statement that no finding has been made; do not open with a question that already assumes an answer, and do not let the meeting slide into an interrogation. Adapt the details to the records and procedure you actually have.

I would like to understand how you developed this submission. A checker returned a result that raised a question, but that result is not proof, and I have not decided that a rule was broken. Please talk me through your work from the brief to the final version. Then we can look together at any drafts, notes, sources, feedback, or version history you still have. I will explain what happens next under the applicable procedure.

If a score did not prompt the review, name the actual observation instead: “The final argument differs from the outline saved in the course system. Please explain how that change happened.” Do not attach a conclusion such as “so I know you used AI.” The phrase “you still have” is deliberate: it requests existing material without demanding that the student rebuild an archive after submission.

1. Set the status. Name the observation and confirm that no finding has been made.
2. Hear the sequence. Let the student describe the path from brief to submission before you isolate individual sentences.
3. Examine decisions. Ask why a source, example, term, or structure was chosen.
4. Review existing records. Identify what each item shows and what remains unknown.
5. Explain the next step. State who is authorised to decide and when an update should follow.

A suitable close: “I will compare your explanation with the submission, the available records, and the applicable rule. The detector result will not be treated as a finding.” Do not promise a penalty or an exoneration before the authorised review is complete.

What should you establish before the academic integrity meeting?

Establish the applicable rule, the precise observation that triggered review, the type of assistance at issue, the authorised decision-maker, and the student's opportunity to respond. Mark unanswered points as unknown instead of resolving them by assumption.

An academic integrity meeting is a structured conversation in which a teacher and student examine how submitted work was produced under the applicable procedure — not a demand for a confession or notice of a decision already made. Treat duty of care as a practical editorial principle: neutral wording, consistent treatment, communication support, and a genuine opportunity to answer all reduce the risk of harm from a premature accusation. Burden of proof concerns who must establish that a rule was broken and to what standard; no universal standard was supplied for this article, so consult the governing policy and never tell a student to “prove you wrote it.”

When the assignment, course, or institution policy is silent or ambiguous about AI use, do not resolve that ambiguity against the student. Record the gap in writing, apply the most specific written rule that actually exists, and refer the interpretation to the authorised decision-maker. Reading a general integrity statement as an unwritten AI ban converts your uncertainty into the student's liability. A silent policy is a reason to refer, not a reason to conclude.

Draft evidence means retained outlines, notes, sources, comments, and working files that may connect the student's explanation to the final work. Version history means changes captured while tracking was active in a particular application; it is partial by design, because it cannot show activity in another file, work completed offline, or changes made before tracking began.

Five process questions: a low-recall route for an honest author to explain how the work developed

These five rewrites are specific to this guide. Each gives an honest author a practical, low-recall way to explain how the submission actually developed, asking for a sequence, a decision, or an available record instead of requiring acceptance of an allegation.

A calm voice cannot repair a question that presumes guilt. Follow an answer with a precise open request: “You said the counterargument changed after the second source. Show me where that change appears and what the source changed in your reasoning.”

Corroborating records and their limits

This matrix is specific to this guide, and it reads in both directions. Read both right-hand columns before relying on any single item.

Record limitations narrowly — “No earlier file or tracked history was available for comparison,” rather than “No history means no authorship.” For the separate task of interpreting the number itself, see the guide to reading an AI detection score.

Two worked meetings using the same illustrative score

The numbers here are invented solely to demonstrate reasoning. They are not research findings, detector measurements, accuracy statistics, or observations about real students.

Meeting one: records connect the account

Suppose a submission receives an illustrative score of 88%. The student says a 1,400-word draft became a 950-word submission after feedback flagged repetition. The available material contains an outline with the final central claim, three early examples, one feedback comment recommending a tighter argument, and 42 logged edits across five days, several of which shorten or move paragraphs. These records do not disprove the detector; they support a narrower conclusion, that the account, outline, feedback, and recorded revisions form a connected development history. The teacher still compares any disclosed assistance with the assignment rule. If that comparison leaves no material question, the concern can be resolved even though the illustrative score remains 88%.

Meeting two: no earlier file survives

Keep the illustrative score at 88%, but now suppose the student drafted in a temporary document and kept no earlier file. That makes one comparison unavailable; it does not establish the opposite conclusion. Ask about planning, source selection, feedback, and decisions visible in the final submission; existing handwritten notes or an earlier message may add limited context, each kept within its own evidential limits. If the remaining material supports no conclusion either way, record the status as unresolved. The score must not become a tie-breaker because other evidence is incomplete. The two scenarios differ because of what the account and surviving records support, not because of any calculation involving the number 88.

Language background, anxiety, and specialist terms

Do not infer authorship from accent, oral fluency, anxiety, concise prose, intensive editing, or the use of English as an additional language, all of which are flagged more often without indicating misconduct. Keep specialist terminology when it expresses the subject accurately: if “heteroskedasticity” appears in a statistics assignment, ask what it describes and how it affects the chosen method rather than treating a necessary term as a red flag. Instead of reading nerves as evidence, ask what translation, tutoring, grammar tools, dictionaries, or peer comments shaped the submission, then compare the assistance described with the applicable rule. The site's guide to AI detector false positives owns the separate question of why human writing gets flagged.

How can a student explain work they wrote?

A student can explain authored work by describing the sequence of decisions they made, linking each decision to a specific passage, naming any assistance honestly, and sharing records that already exist. The test is the connection between decision, passage, and record, not polished delivery or perfect recall.

For example: “My first outline argued X. Source B challenged the second premise, so I removed that paragraph and narrowed the conclusion to Y. The original claim is still in my outline, and the feedback comment identifies the same weakness.” Each claimed decision points to material that can be examined. If the student cannot recall a minor wording choice, move to a meaningful decision: why a source was trusted, how an example supports the claim, or why a counterargument was included. Do not turn the meeting into a memory or performance test, and encourage the student to say honestly when a record is missing rather than building a retrospective document and presenting it as contemporaneous.

Choosing resolved, unresolved, or referred: accurate outcome wording

Each outcome should state the current status, the reasoning behind it, any remaining uncertainty, and the next procedural step. Choosing the accurate label is part of the fairness: referral transfers a question to an authorised process; it does not establish a finding.

Resolved

Use resolved when the account and available records connect adequately with the work and no relevant question remains under the checked rule. Example: “The student connected the central claim to the retained outline, identified the feedback that prompted revision, and disclosed the editing support used. That support raises no remaining question under the assignment rule, so the concern is closed.”

Unresolved

Use unresolved when the available information supports no conclusion either way. Example: “No earlier file was available, and the remaining records neither materially connect with nor conflict with the account, so the status is unresolved.” Missing history, nervous delivery, specialist language, or an accent cannot fill that gap, and neither can the detector score.

Referred

Use referred when the conversation surfaces a policy question belonging to another authorised decision-maker. Example: “The student disclosed using a generative tool to restructure the draft. The assignment wording does not clearly state whether that assistance was permitted, so interpretation is referred with the student's account and records. No finding has been made.” This is the correct route whenever the policy is silent or ambiguous.

Questions teachers and students ask

Does a high AI detector score mean my student cheated?

No. Under FiftyGPT's supplied responsible-use standard, a detector result is a probability signal rather than proof of authorship, intent, or misconduct. Examine the account, the submission, the available records, and the applicable rule.

How can I weigh a detector result fairly?

Keep it as one limited input, hear the student's production account, and record what each supporting item can and cannot show. If the material supports no conclusion, use an unresolved outcome instead of letting the score decide.

What can a student do if they say the detector is wrong?

They can request the precise concern, explain their development decisions, connect those decisions to passages, and share existing drafts, notes, feedback, sources, or version history without inventing missing records.

What if the student has no drafts or version history?

Record the limitation and examine other available context, including planning, sources, feedback, and decisions visible in the work. A missing archive does not by itself establish prohibited AI use.