The AI Detector Flagged Her Thesis. It Also Flagged the Professor's Syllabus.
Manila universities run student essays through tools that mistake second-language English for machine writing. The appeals process is a Google Form that nobody answers.
A working student in Sampaloc spends her LRT ride drafting an essay on her phone. She writes in English, her third language after Bikol and Tagalog. By the time she gets to campus, she has 800 words. By the time her professor uploads it to GPTZero, she has a high AI score and a meeting with the dean.
This is the year Manila universities decided that AI detectors were the answer to ChatGPT. The detectors don't work. The professors know. The students know. The appeals process pretends nobody knows.
The tools were never built for second-language writers
GPTZero, Turnitin's AI checker, Copyleaks, and the rest were trained mostly on English written by native speakers. Peer-reviewed studies in the US have shown these tools disproportionately flag non-native English writing as machine-generated. The patterns that trigger a high AI score, simpler vocabulary, predictable sentence structure, fewer idioms, are the exact patterns of someone writing carefully in a language they learned in school.
Filipino students writing in English have been getting flagged at rates that would be funny if the consequences were not failing grades and academic dishonesty cases. A jeepney driver's kid at a state university who works nights and writes her essays between shifts produces clean, careful prose. The detector reads clean and careful as suspicious.
The professors are using the same tools that flag their own work
Faculty members across Manila campuses have started running their own lecture notes, syllabi, and published papers through the detectors out of curiosity. Many come back flagged as likely AI. Academics abroad have reported the same thing with dissertations and journal articles written years before ChatGPT existed.
Nobody is pulling the tools. Department heads cite "academic integrity policy." The integrity policy was written by an office that bought a subscription and called it a solution.
The appeals process is a Google Form
A student flagged for AI use gets a notice, a meeting, and a chance to appeal. The appeal usually means submitting drafts, version history, and a written explanation. Working students who draft on Notes app, on a borrowed phone, or in a shared computer at the barangay hall do not have version history. Google Docs revision logs require a Google Doc. The kids being accused are often the ones least likely to have the digital paper trail to clear themselves.
When the appeal goes up, it lands in a queue that takes weeks. Semesters end before decisions come down. Students take incompletes. Some lose scholarships tied to GPA. A few have been quietly told to retake the course, which costs another semester of tuition the family does not have.
Meanwhile, on the bus
The students who actually use ChatGPT, the ones who paste the prompt and copy the output without reading it, mostly pass. They know to run their draft through a humanizer first. They know which detectors their professor uses. They know to add typos, drop articles, shorten sentences. The cheating is happening. It is just not the kids getting caught.
The student flagged for her LRT essay paid for this semester out of a closing shift at a milk tea kiosk in Recto. Her appeal has been pending long enough that the semester is over. Her scholarship coordinator has stopped replying to her messages. The detector software stays on the procurement list for next school year.
