“How can you tell if text was written by AI?” is a question for everyone from teachers to editors. The short answer may be surprising: there is no method that proves for certain that a text was written with a tool like ChatGPT. So-called AI detectors can give hints, but they are often wrong.
How do AI detectors work?
These tools measure how “predictable” a text is. Because language models tend to pick the most likely word at each step, plain, uniform sentences may be flagged as AI. The problem is that a person who writes simply and correctly can show the same features.
Why are AI detectors unreliable?
- False positives: a 2023 study by Stanford University researchers found that popular detectors flagged texts by non-native English writers as AI-generated at very high rates.
- Easy to fool: lightly rewriting a text or editing it with another tool changes the scores dramatically.
- Even the makers are cautious: OpenAI withdrew its own text classifier in 2023, citing its low accuracy.
Signs that a text may have been written by AI
None of these is proof on its own; they are only reasons to look more closely:
- General statements on the topic with no personal experience or original examples.
- Non-existent sources, invented quotes or wrong dates (AI “hallucinations”).
- A style clearly different from the student’s or writer’s earlier work.
- A generic structure that would fit almost any question rather than the one asked.
What can teachers and editors do?
- See the process: drafts, notes and version history are far more informative than the final text.
- Have a short conversation: asking the author to explain the content is one of the most reliable checks.
- Check the sources: invented references surface quickly.
- Set clear rules: state in advance when and how AI may be used.
Frequently asked questions
Can Turnitin detect AI?
Platforms such as Turnitin offer an AI indicator, but it is also a probability estimate. The makers themselves advise against using the score alone to make decisions.
Does Google penalise AI-written content?
Google says it looks less at how content is produced and more at whether it is helpful, accurate and original. Mass-produced content made only to manipulate search rankings violates its policies.
Sources
- Liang et al., “GPT detectors are biased against non-native English writers”, Patterns, 2023.
- OpenAI, “New AI classifier for indicating AI-written text” (July 2023 update).
- Google Search Central, guidance on AI-generated content.
Leave a comment
Your email address will not be published. Required fields are marked *