humanflow

Plagiarism

Plagiarism is presenting someone else's work or ideas as your own without crediting them — a question about attribution, which is separate from the question of whether a machine produced the words.

Last reviewed 15 August 2026 · The HumanFlow team

In plain English

The test is not whether the words are original. It is whether the source got credit. Copying a sentence and citing it is scholarship; copying a sentence without citing it is plagiarism; rewording that sentence to avoid a similarity match and still not citing it is also plagiarism, and most integrity policies name that case explicitly.

This is why paraphrasing is not a defence. Changing the words changes what a matching tool sees, not what you owe the source.

A worked example

One paragraph, two different systems, two different findings. Neither answers the other's question.

Submitted:  "Detection tools measure predictability rather than authorship,
             which is why fluent formulaic writing scores badly."

  similarity report  → 0% match     (no source in the database says this)
  AI writing report  → flagged      (the sentence is highly predictable)

Submitted:  "Detection tools measure statistical predictability, not
             authorship."  [uncited, taken verbatim from a paper]

  similarity report  → 100% match   (the source is in the database)
  AI writing report  → not flagged  (a person wrote it, unusually)

The first submission is original work that a detector dislikes. It is not plagiarism by any definition, and no amount of AI flagging makes it so.

The second is textbook plagiarism that the AI detector has nothing to say about, because the sentence was written by a human — just not by this one.

The two systems are answering different questions, and a report that prints both figures side by side invites exactly the confusion above.

Why it matters for AI detection

Because AI detection and plagiarism detection get conflated constantly, including in the reports themselves, and the conflation harms people in both directions — original work treated as misconduct, and genuine misconduct missed.

For a writer using any rewriting tool, ours included, the rule does not move: if material came from a source, cite the source. Rewording an unattributed passage to lower a similarity score is the specific behaviour most academic integrity policies single out.

For an institution, they are separate evidentiary questions requiring separate evidence. A similarity match points at a document you can go and read. An AI score points at nothing you can inspect.

Commonly confused with

AI writing detection
Plagiarism is about attribution and can be checked against a named source. AI detection is about statistical style and has no source to check against. A text can be either, both, or neither.
Contract cheating
Plagiarism reuses existing work. Contract cheating commissions new work, so there is nothing for a similarity tool to match — which is why it is both more serious and harder to detect.

Read next

Part of the AI detection glossary.