testing something in my classroom and want to compare with others
i’ve been having students run the same content in different POVs through detectors as an exercise. first-person reflective essays and third-person analytical essays on the same topic. the first-person versions consistently score lower for AI probability.
my interpretation: first-person writing is harder for AI to imitate convincingly because it requires specific autobiographical details. or alternatively, detectors are trained on more AI-generated third-person content. is either of those explanations right?
both explanations have some truth. first-person writing that includes specific autobiographical detail is genuinely harder for AI to produce without prompting because the model doesn’t have access to your actual experiences. when it tries, it often produces vague, generic “personal” content that has its own detectable pattern.
the training data explanation also holds. most AI detector training sets are heavily weighted toward academic essays, which are predominantly third-person analytical. the models are better calibrated on that format
the classroom experiment is a smart methodological choice. one thing worth noting for your students: the first-person advantage disappears quickly if the first-person writing doesn’t contain genuine specific detail. “I found this topic interesting because it connects to my experience” scores high because it’s generic first-person. “I found this topic interesting because of a specific thing that happened” scores lower because specificity is a human signal
from a dissertation writing perspective: the POV finding is interesting because academic convention pushes toward third-person precisely because it sounds more authoritative. but third-person analytical prose is also exactly the format AI produces most fluently. the convention and the detection risk are aligned in the wrong direction for academic writers
the specific detail point is the real mechanism here and it applies to fiction and creative writing too. first-person writing that contains invented but specific sensory detail scores lower than first-person writing that stays at the level of general reflection.
the detector is measuring something real: the presence of particularity. AI is particular when prompted to be, but default AI output gravitates toward the general
the practical implication for anyone writing under their own name: first-person with specific experiential detail is your lowest-risk format. not always appropriate but worth knowing as a baseline.
for content that has to be third-person: aichecker.tech performs a sentence-level analysis that shows which specific sentences are flagging. targeting those sentences specifically is more efficient than re-running the whole piece