AI Checker (AI-generated text detector)

"AI checker" (or AI detector) is the umbrella term for tools that estimate whether a text was written by generative AI. Schools use them to screen assignments, publishers to check manuscripts, and employers to review application documents.

How detection works

There are two main approaches. The first looks at statistical features of the text itself, such as how predictable the word choices are and how much sentence length varies (human writing tends to vary more), and reports a probability that the text "looks like AI." GPTZero, Copyleaks, Originality.ai and Turnitin's tools for education belong to this group.

The second approach has the AI provider embed an invisible digital watermark in the output at generation time and detect it later. Google's SynthID is an example. It only works on output from services that applied the watermark.

Where detection goes wrong

Accuracy figures published by vendors come from their own evaluations, and independent tests have reported lower numbers. At least one generative AI developer withdrew its own detector because it was not accurate enough. Treating a detector score as proof of misconduct risks accusing innocent people.

Using detectors in an organization

Use detection only as a first-pass screen and combine it with other evidence, such as the writing history or a conversation with the author. Check the terms of service to see whether submitted text is stored or used for training.

For companies, it usually works better to define where AI may be used and which parts people remain accountable for than to ban AI writing outright. An approved tool list and usage rules also support AI governance and reduce shadow AI. We help organizations set AI usage rules and run training through our AI & DX services.