humanflow

Originality report

Also called Similarity report

An originality report is the document a text-matching service produces about a submission — usually bundling several independent measurements, produced by different systems, into one page that reads as a single verdict.

Last reviewed 15 August 2026 · The HumanFlow team

In plain English

It is not one number. It is a folder of separate checks printed together, and the checks answer different questions with different reliability.

Most of the confusion in this area comes from reading the page as a whole rather than as a list of unrelated findings.

A worked example

The anatomy of a report, with what each element is and whether you can inspect the evidence behind it.

  element              what it is                    inspectable?
  ───────────────────────────────────────────────────────────────
  similarity 24%       share matching indexed text   yes — named sources
  matched sources      the documents it matched      yes — open and read
  AI writing 40%       share reading as machine      no — nothing to open
  AI writing  *        score withheld below 20%      n/a
  integrity flags      formatting or character odd-  yes — shown in text
                       ities, e.g. hidden characters

Three of the five rows point at something you can go and examine. One does not, and it is the one that generates most accusations.

The similarity figure and the AI figure are produced by entirely separate systems and are not components of a shared total. A document can be high on one and zero on the other, and frequently is.

Integrity flags are worth knowing about separately: they cover things like text in a hidden colour or unusual character substitutions, and unlike either score they describe an observable property of the file.

Why it matters for AI detection

Because the report is the artefact an accusation is built on, and reading it as one verdict rather than several findings is the single most common error in this area.

It matters practically in a meeting. "Which row are we discussing?" is a reasonable question, and for the similarity rows the follow-up — "can we open the matched source?" — usually resolves things quickly.

It also sets expectations correctly for the AI row. There is no source to open, no passage-level evidence unless the tool provides it separately, and the vendor states the figure is not intended as the sole basis for a finding.

Commonly confused with

Similarity score
The similarity score is one row on the report. The report is the whole document, including rows produced by unrelated systems.
AI writing score
Another single row, from a different system. Its adjacency to the similarity figure on the page implies a relationship that does not exist.

Read next

Part of the AI detection glossary.