Reading results
Why can writing from before ChatGPT receive an AI flag?
The short answer
Because the detector is judging the wording you give it. It isn't checking when the work was published. An old text can therefore come back with a high score. That score does not prove it was generated: compare the scanned passage with the dated original.
Which text are you holding?
The age of the work and the age of the passage you're scanning can be different. A summary written today doesn't inherit the age of the book it describes; a web transcription may contain later edits. Find an original edition, archived document or dated manuscript and compare the sentences with your input.
A file's last-modified date is a different clue. On its own, it gives you little evidence of when the prose was written. A library record, publication scan or earlier saved draft may tell you more.
The flag may be wrong
Weber-Wulff and colleagues documented classification errors in the tools they tested in 2023. We cite that study here as a reason to seek evidence beyond a flag. It isn't a test of today's AiChecker, and it doesn't decide the origin of the paragraph in front of you.
Source: Weber-Wulff et al. (2023): detection errors in tested tools
Keep the disagreement in the record
Keep the date and location of the original beside the exact input you scanned. If this is also a test of your detector, set the label from the passage's documented origin. A score that disagrees with the record is a result to examine, not a reason to relabel the passage.