16 September 2026 · 8 min read
Most Nepali students meet a similarity score the same way: a number handed back by a department with no explanation and no obvious route to respond. The AI writing indicator has started arriving the same way, and it is harder to answer because it points at no source you can open. This post sets out what a proper report pack contains, what each figure can honestly be used for, and how to have your own work checked while there is still time to act on the result.
Every piece of work we deliver is checked before it reaches you, and the reports are issued with the file rather than on request. That covers the similarity report in full, and — where your institution's configuration produces one — the AI writing indicator read alongside it.
The checks are also available on their own, on work we had no part in writing. This is a service in its own right and a large share of what we do: you send the draft, you get the reports back with every flagged passage explained. You do not have to have bought writing from anyone to have a document checked before you submit it.
Similarity measures text overlap with a corpus. That is all it measures. It is not a plagiarism score, it does not distinguish a correctly quoted methodology from a lifted paragraph, and it says nothing about whether you understood what you wrote.
This is why two documents at 18% can be in completely different positions. One is spread thinly across dozens of correctly cited sources and a reference list that matches every other paper citing the same works — that is a healthy report. The other is concentrated in three paragraphs tracking one source with no citation — that is a referral. Only the report tells you which one you are holding, which is why we send the report rather than the number.
It also means the score moves with settings you do not control. A report configured to count the bibliography and the block quotations reads far higher than one set to exclude them, on the same submission. When a supervisor says your similarity is too high, the first useful question is what the report was set to exclude.
The AI writing indicator is a separate figure produced by a different mechanism, and Nepali departments have been switching it on steadily. It is not part of the similarity percentage and the two do not move together — a document can show 4% similarity and still be referred on the AI score, which is now one of the most common ways a student first encounters a conduct query.
The critical difference is evidentiary. Similarity points at specific text in a specific source you can open, read and compare; it is evidence and it can be argued with. The AI indicator is a probability estimate over your prose with nothing behind it to examine. That asymmetry is why a high AI score leaves a student with so much less to work with, and why the response has to be different.
It is also unreliable in a direction that matters in Nepal. Detectors respond to prose that is formal, structurally regular and vocabulary-conservative, which describes careful academic English written by someone who learned the language through formal instruction rather than immersion. Writing more carefully — drafting, revising, tightening — pushes prose in exactly that direction. Students find that hard to believe until it happens to them.
We are an independent academic support service. We are not a university, we are not affiliated with any institution, and we are not a reseller of any detection platform. We run checks and issue the resulting reports; we do not sell access to a detection tool and we do not sell a way around one.
We also do not promise a similarity figure or an AI score, on our work or on yours, and we will not rewrite work solely to move a detector. Any service quoting you a guaranteed number is describing an output it does not control — and a guarantee is a useful signal that the rest of what you are being told is also unreliable.
Paraphrasing tools deserve a specific warning. They are themselves generative, so their output carries its own detectable signature. Running a flagged passage through one typically raises the AI score rather than lowering it, and frequently produces text that no longer says what you meant. It is the single most common way a recoverable problem becomes a worse one.
A report that arrives the night before a deadline tells you about a problem you no longer have time to fix. The whole value of a pre-submission check is the interval between the result and the deadline, so the timing decision matters more than most students assume.
The evidence that you wrote your own work is not in the finished file; it is in everything around it. Keep your outline, your reading notes, your earlier drafts and your supervisor's comments, with file dates intact and nothing overwritten. Write in a tool that keeps version history — Google Docs and Word both do — so the document's own record shows the text accumulating rather than appearing.
Keep the messy artefacts specifically: notes to yourself, abandoned paragraphs, the section you moved twice and deleted, comments from your supervisor and what you did about them. A generated document does not have those, and they are far more persuasive than a clean draft. If you are ever asked about your work, this is what answers the question — and it cannot be assembled after the question is asked.
Where this is set out in more detail:
If you want a draft checked — ours or entirely your own — send it over on WhatsApp with your deadline and we will tell you what the reports show and what, if anything, actually needs doing about it.