Update, October 2026: OpenAI withdrew the AI Text Classifier in July 2023 because of its low rate of accuracy, and the tool is no longer available.
Introduction
OpenAI, the American tech firm, has recently announced the release of an AI-text detection tool – the “AI Text Classifier” to help users detect AI-generated texts. The tool can detect AI-generated texts from different tools including ChatGPT, although it currently lacks the necessary precision.
Threat to Human Writing Capabilities
OpenAI’s ChatGPT has disrupted the industry as an AI text generator and has even threatened the survival of human copywriters, as seen from Microsoft’s recent layoff of thousands of employees while simultaneously investing $10 billion in ChatGPT’s maker, OpenAI. The tool’s precision has made it a one-stop solution for creating texts.
Performance of the AI Text Classifier
According to OpenAI’s blog post, the latest AI text classifier tool correctly identified 26% of AI-written text as “likely AI-written” in OpenAI’s tests (its true-positive rate). It is currently in its early stages and has many limitations, generating up to 9% of false alarms (detecting human-written text as AI-generated). However, the classifier works relatively well with texts longer than 1000 characters but produces unreliable results for shorter texts.
Language Limitations and Availability
OpenAI recommends using the tool for the English language only (at present) as it is unreliable for other languages yet. The tool is now available to the public, and OpenAI seeks feedback from different user groups, especially educators. The firm is also keen to improve the tool to better serve users like journalists and information/disinformation groups.
In conclusion, OpenAI’s “AI Text Classifier” is a step towards creating a balance between human writing and AI-generated texts. While the tool still has many limitations, it is a work in progress and can be improved with user feedback.
Correction, 26 October 2026: this article said the classifier detected AI-generated text with 26% precision; 26% was the share of AI-written text it correctly identified (its true-positive rate), not its precision.