16 September 2026 · 8 min read
Two numbers now come back with a submission in a growing number of Nepali departments, and students treat them as versions of the same thing. They are not. Similarity is a measurement: this text matches that text, here is the source, open it and look. The AI writing indicator is an estimate about the prose itself, with no source behind it and nothing to compare against. Everything awkward about it follows from that one difference.
Where an institution has the feature enabled, the report typically gives an overall figure — the share of the document the system judges likely to be machine-generated — and a marked-up view showing which passages drove it. That is roughly all. There is no list of sources, because there are none, and no passage-level explanation of why a sentence was marked.
AI detectors respond to prose that is formal, structurally regular and conservative in vocabulary. That is a fair description of careful academic English written by someone who learned the language through formal instruction rather than immersion — which is most students writing a thesis in Nepal. The pattern the detector associates with generated text is also the pattern good second-language academic writing tends to have.
Heavy editing pushes the same way. A student who drafts, revises and tightens repeatedly produces smoother, more uniform prose than one who writes once and submits, and uniformity is part of what the indicator reacts to. Working harder on the writing can raise the score, which students find hard to believe until it happens to them.
Certain chapters attract flags for structural reasons too. A methodology chapter is formulaic by design — standard procedural language, standard instrument descriptions — and formulaic is precisely what the detector is sensitive to. A flag concentrated in your methodology is telling you something about the genre more than about you.
Running the text through a paraphrasing tool is the most common response and the worst one. Those tools are themselves generative; their output carries its own signature, so the score typically goes up rather than down, and the prose frequently stops saying what you meant. Students turn a recoverable question into a much harder one this way every year.
There is also no honest service that will lower an AI score for you. Any figure guaranteed to you is being promised by someone who does not control the system producing it. We do not make that promise about our work or yours, and we will not rewrite a document for the sole purpose of moving a detector.
Because the score has no source behind it, the only real answer to it is evidence that you wrote the document. That evidence is built while you write and cannot be assembled afterwards.
Do not rewrite in a panic — that usually makes the position worse and it looks like a reaction. Gather the evidence instead, and approach your supervisor with it rather than waiting to be approached. A high AI score is a prompt for a conversation, not a finding of misconduct, and turning up with dated drafts and notes changes the character of that conversation completely.
Ask, politely, what the department's process is and what weight the indicator carries in it. Practice varies widely between institutions and often between supervisors, and knowing whether you are in an informal discussion or a formal procedure changes what you should be doing next.
Related:
Send the document and the report you were given. You will get an honest read on whether the flag looks like a detector artefact, and a clear list of the evidence that would answer it.