A writer scans the same unedited essay through the same plagiarism checker twice, five minutes apart, expecting an identical result. The first run comes back at 14 percent. The second comes back at 16 percent. Nothing about the essay changed between the two scans.

This piece explains why identical text can produce slightly different scores on repeated scans, which parts of the process introduce that variation, and how much of a difference should actually raise a concern versus simply reflect normal noise in the system.

The index itself is not frozen in time

The single biggest source of run-to-run variation is that a plagiarism checker's index of source material is not a fixed, unchanging snapshot.

In practice, this effect is usually tiny over a five-minute window, but it compounds over longer periods. A scan run today and the same scan run next month can show a more noticeably different result, not because your writing changed, but because the index it is being compared against has grown or shifted in the meantime.

Small variance in matching thresholds and processing

Some of the variation comes from how the matching process itself handles borderline cases. A phrase that sits right at the edge of a similarity threshold, close enough to a source to register as a partial match but not clearly over the line, can occasionally get classified slightly differently between runs, particularly if the system uses any randomized or probabilistic component in how it samples.

Settings differences that look like inconsistency but are not

A more common source of larger score differences is not randomness at all, but a difference in scan settings between the two runs that the user did not notice. Whether quotations are excluded from the calculation, whether the bibliography or reference section is included, and whether a specific comparison database is toggled on or off can all change the resulting score meaningfully.

Before assuming a score difference reflects genuine inconsistency in a plagiarism checker, it is worth checking that both scans used identical settings. A difference caused by an excluded bibliography in one run and an included one in the next looks like inconsistency but is actually a straightforward, explainable settings difference once you know to check for it.

How much variation should actually concern you

A one to three percentage point difference between repeated scans of unchanged text, with identical settings, reflects normal system behavior and is not worth worrying about. It does not mean the tool is unreliable, and it does not mean either number is the wrong one.

A useful habit for anyone who needs to compare scores across multiple scans, whether checking a document before and after revision or comparing results across different tools, is to note the exact settings used at the time of each scan alongside the resulting score.

The Repeatability Question

Small differences between repeated scans of the same unchanged text are a normal consequence of how these tools work: a continuously updated index, borderline matches that occasionally classify differently, and processing that is not perfectly deterministic at the margins.

For more detail on how plagiarism checking systems handle their indexes and what factors influence scan consistency, further reading on how plagiarism detection systems actually work walks through the underlying mechanics in more depth for anyone who wants a fuller technical picture.