Which free AI text detectors are actually worth trying for a quick check?

My honest answer: there is no free detector I would call validated for your kind of text. The useful ones publish how they were tested and on what, admit their error rates, and say what they cannot do. Most free pages do not. So use any of them as a smoke alarm, not as a witness.

On the practical question: a "human" result tells you very little. A "likely AI" result tells you someone could see a similar result and ask you about it. That is about all. The thread comparing different tools goes through the evidence on the better-known ones.

Before the question of which tool: where does your text go? A free web page that takes a client's unpublished copy is a transfer of that copy to a stranger. Read the privacy and terms pages, and check whether the tool keeps what you paste and whether it uses it for anything else. If you cannot find an answer, assume the worst.

If the client's NDA covers drafts, you may already be in breach by pasting them. For anything unpublished I would test on a harmless paragraph of your own, not the client's.

Teacher view: the tools students and parents most often bring me are the free ones, and the numbers they bring are not stable. The same essay can score differently on a different day or after a one-word change. I now ask for the document history instead of the score. It is much harder to argue with.

For your case, the equivalent is simple. Keep your outline, notes, and the drafts with dates. If a client questions you, show those, not a score from a free site.

if three different free detectors all say "human" then you are safe, and if all three say AI then it is AI. thats how i do it. the more of them agree the more you can trust it, since they are all built differently.

@Sam Whitlock that is the part I had not thought about. My contracts do not say anything about pasting drafts into third-party tools, which means I should ask before I do it. For now I will test on my own paragraphs only.

I run a small check on my own writing sometimes, just to know what a student might see. My rule is: one tool is enough, because a second one adds confusion, not certainty. And I never change the text to chase a lower score. The moment you do that you are optimising for the tool and not for the reader, and the writing gets worse.

Also worth knowing: scores on short passages are the least stable. If you paste a paragraph and get a result, I would not read anything into it.

@Jules I would not use agreement between tools as proof. Several of these tools may rely on similar signals, trained on overlapping kinds of data, so they can be wrong in the same direction on the same text. Agreement makes a result feel stronger without making it independent. Plain, formulaic human writing is a case where several detectors can all be wrong together. Treat three matching scores as one opinion repeated, not three confirmations.

As a worried student, I have used these as a warning only. If something is flagged I do not rewrite it blindly. I look at which sentences were highlighted, and usually it is the plain, factual ones where there is not much choice in how to say them. It helped me understand why it looked suspicious, without changing what I said.

@Hana Sato ok that makes sense, they probably share a lot of the same blind spots. i guess i was treating them like independent witnesses when really it is one opinion said three times. i will tell my group chat to stop doing the three-tab trick, it just makes everyone more sure than they should be.