A fair AI-writing review checklist for essays and assignments

An essay receives a high AI-writing score. The tempting next step is to treat the number as the answer. A fairer next step is to examine what the system analysed and give the student a chance to explain the work.

This is a suggested workflow, not a finding about a student or a substitute for institutional policy. It applies to essays, reflective assignments and academic reports, but does not assume equal detector reliability on every format.

1. Check the assignment and input

What assistance was permitted, what disclosure was required, and what work was assessed? Do not treat permitted spelling assistance and undisclosed generation as automatically equivalent.

Inspect the submitted text. Did PDF extraction repeat page headers? Were references folded into a paragraph? Are slides mostly labels rather than continuous prose? Resolve damaged extraction before interpreting output. Our PDF preparation guide explains a simple check.

2. Review passages rather than accusing from a colour

Read flagged sections with neighbouring sentences. Note quotations, assignment instructions and required terminology. A fluent sentence or repeated transition is not by itself evidence of AI use. An unflagged passage does not certify human authorship either.

Turnitin's guidance warns that its model can misidentify text and should not be the sole basis for adverse action against a student. Apply the same cautious review principle to Certaffy output.

3. Ask about the writing process

Useful questions include: How did you choose your argument? Which sources changed your view? Can you explain this example? Do you have notes or drafts you can share under the institution's policy?

Draft history can add context, but is not a perfect authorship test. Students use different tools, may draft offline and may not retain every version. Avoid treating polished style, a student's first language or missing history as automatic grounds for suspicion.

4. Record uncertainty and allow a response

Keep the requirements, submitted version, report date and student's explanation together. Distinguish observed facts, such as missing citations, from model inferences. Record the reason for follow-up and provide the normal review or appeal route.

For a departmental pilot, use consented examples with known provenance. Include genuine student writing, different subjects, relevant lengths and permitted assistance. Report false positives alongside detected AI examples; one overall accuracy figure is not enough.

The academic workspace is Certaffy's education entry point. Check account features and limits before uploading. Do not submit sensitive student material without the necessary permission and privacy review. Start with a non-sensitive sample and confirm the report refers to the right text.

Read what an AI detector score means for the difference between probabilities, highlights and estimated shares.

Prepared with AI assistance for Certaffy. Source links checked on September 26, 2026.