AI detector: An AI detector is a tool that tries to estimate whether a piece of text, an image, or audio was generated by AI. Detectors for text in particular are unreliable, with false positives and easy evasion, and should not be used as sole evidence.
Text detectors look for statistical patterns typical of model output, such as unusually predictable word choices. Image and audio detectors look for generation artifacts or embedded watermarks. Products like GPTZero and Turnitin's AI writing indicator are widely used in education and hiring, which is exactly where their limits matter most.
The evidence on text detection is not encouraging. OpenAI retired its own AI text classifier in 2023, citing low accuracy. A 2023 study by Stanford researchers led by Weixin Liang found that popular detectors frequently misclassified essays by non-native English writers as AI-generated. Light editing or paraphrasing lowers detection rates further, while polished human writing can score as likely AI. A detector's percentage is not a probability you can act on.
Image and audio detection is somewhat better when a watermark is present, but a missing watermark proves nothing. If you must evaluate whether work is AI-generated, use process rather than a score: ask for drafts and revision history, discuss the work with the person, compare against their prior writing, and treat any detector output as one weak signal. Never make a disciplinary, academic, or hiring decision on a detector result alone.
Example at work
A hiring manager runs a candidate's writing sample through a detector and gets "87 percent AI." Instead of rejecting the candidate, he asks her to walk through her reasoning in the interview and write a short response live. She clearly wrote the original. The detector had flagged clean, formal prose.
Why it matters
Detectors promise a certainty they cannot deliver. Knowing that protects you from accusing someone falsely, from building a policy that will not hold up, and from a false sense of security when the thing you are checking really was machine-made.
Related terms
- AI watermarkingAI watermarking embeds a hidden or attached signal in AI-generated content, such as an invisible pattern in an image or metadata in a file, so that its origin can be identified later. It helps with provenance but is far from foolproof.
- DeepfakeA deepfake is synthetic audio, video, or imagery, generated or altered by AI, that convincingly depicts a real person saying or doing something they never did.
- Generative AIGenerative AI is a class of AI models that create new content, including text, images, code, audio, and video, in response to a prompt. Chat assistants like ChatGPT, Claude, and Gemini and image tools like Midjourney are generative AI.
- Large language model (LLM)A large language model (LLM) is a neural network trained on enormous amounts of text to predict the next word, then tuned to follow instructions and hold a conversation. LLMs power ChatGPT, Claude, Gemini, and Microsoft Copilot.
- Workplace AI policyA workplace AI policy is a written set of rules that tells employees which AI tools they may use, what data they may put into them, how to verify and disclose AI-assisted work, and who to ask when unsure.