The question this answers
Not "did an AI write this". Nobody can answer that, and any tool that claims to is guessing at your expense. The question here is narrower and completely answerable: what characters are actually in this file?
That matters because the answer is checkable by anyone. Invisible characters survive copy and paste, they survive a Word round-trip, and they are visible to anybody who opens the file in a hex editor or runs it through a Unicode checker. If they are in your document, they are in your document whether or not the person reading it goes looking.
So the report is a factual inventory rather than a verdict. Here is what is in there, here is what each one is called, here is where it sits. What you do with that is your business.
What we find, and what we cannot
Hidden characters come out completely. Zero-width spaces, word joiners, variation selectors, tag characters, private-use characters, odd spaces, letters borrowed from another alphabet to stand in for Latin ones. These are data sitting in your file, we find all of it, and once removed you can run the check again and see nothing left.
Word-choice watermarks do not come out, because there is nothing to remove. Google's SynthID for text works by nudging which words the model picks, so the watermark is the word choice itself. There is no character to delete, and no public tool can read it either — that needs Google's key. Anyone selling you SynthID removal for text is selling you nothing.
The writing-style score is a third thing again, and we are careful about what we claim for it. It measures the surface habits a human grader notices: dash density, stock phrasing, sentence rhythm that never varies, the rule of three used four times a page. It is a heuristic, it is shown with the exact words that triggered it, and it is not evidence of anything.
Why the whole report is free
Because nobody should have to pay to find out whether they have a problem. You can run the check, read every finding, see each one highlighted in your own text and close the tab without giving us anything.
What a pass buys is the clean copy and the rewrite — the output, not the information. That split is deliberate and it is the same one on the image side.