An AI-writing score is an indicator, not evidence. A school that treats it as proof will, sooner or later, accuse a student of something they did not do.
Since generative AI arrived in classrooms, many institutions have added an AI-writing percentage to their originality reports. The number is easy to read and easy to act on. That is exactly the problem. It describes a probability, and a probability is not a finding.
Use an AI-writing score to decide which submissions deserve a closer look, never to decide an outcome. Write that rule into your academic integrity policy, require corroborating evidence before any penalty, and give every student a fair way to show how their work was produced. The judgement belongs to the instructor and the integrity process, not to the software.
A similarity finding points to something concrete: a passage in the submission and a source it matches. The instructor can put the two side by side and see for themselves.
An AI-writing indicator has no source to point to. It estimates how closely a piece of writing resembles the statistical patterns of machine-generated text. Human writing can share those patterns, and machine writing that has been edited can lose them. The result is an estimate with a margin of error in both directions: some genuine student work is flagged, and some generated text is not.
That does not make the indicator useless. It makes it one input, to be weighed alongside everything else an instructor knows about the student and the work.
False positives do not fall evenly. A 2023 Stanford study published in the journal Patterns tested several widely used AI detectors and found they consistently misclassified essays written by non-native English speakers as AI-generated, while essays by native speakers were identified correctly.
The reason is uncomfortable. Careful second-language writers tend to use common vocabulary and regular sentence structures, which is also what machine-generated text looks like. The more carefully a student has written in a language that is not their first, the more likely a detector is to doubt them.
In the Philippines, where most students write assessed work in English as a second or third language, that finding should shape policy directly. A rule that treats a high AI score as proof would place the heaviest risk on exactly the students who have worked hardest at their English.
Most school handbooks were written before AI-writing indicators existed, and many are silent on how to use them. That silence gets filled by whatever the individual instructor assumes the number means. A few clauses remove the ambiguity:
When an indicator raises a concern, a short, consistent routine protects both the student and the instructor:
The most reliable protection against misuse of AI is assessment that shows the work as it develops. Staged submissions, an outline, a first draft and a final version, give an instructor a record of how the student's thinking developed. Short in-class writing gives a baseline of each student's own voice. Brief oral follow-ups on major papers settle most doubts in a few minutes.
Letting students check their drafts against an originality tool before final submission helps too. It moves the conversation from penalty to correction, while there is still time to fix a citation or rewrite a passage.
It should not. An AI-writing score is a probability with known false positives, and relying on it alone risks penalising honest students. A fair process requires corroborating evidence and gives the student an opportunity to respond before any decision is made.
Research published in 2023 found that detectors misread the simpler, more regular vocabulary common in second-language writing as a sign of machine generation. For Philippine institutions, where most students write in English as a second language, this makes careful policy especially important.
Not necessarily. Used as a prompt for a closer look, an indicator can help an instructor decide where to spend limited review time. The harm comes from treating it as a verdict, not from having it on the report.
Evidence the instructor can explain to the student: a clear mismatch with the student's other writing, missing or inconsistent drafts, fabricated or unverifiable citations, or an inability to discuss the submitted work. The indicator can start the inquiry; this kind of evidence decides it.
Veritext presents an AI-writing signal as a supporting indicator alongside the similarity findings, never as proof. The public demo shows the similarity report only; the rest of the platform is covered in a full walkthrough for your faculty.
Read your current integrity policy and check what it says an AI-writing score means. If the answer is nothing, that is the first clause to write. Then see how an originality report should look by running a paper you already know through the Veritext live demo, and contact us to arrange a full walkthrough.
Veritext is built and supported by Onprem2Cloud IT Solutions Co., a Microsoft CSP partner based in Muntinlupa City, Metro Manila, serving schools, colleges and universities across the Philippines with peso invoicing and official receipts. For more on reading originality reports, see what a similarity score does and doesn't tell you.
Our latest news and trending topics