OpenAI launches new tool to detect AI-generated text

OpenAI, the creator of ChatGPT, today released an AI Text Classifier that might be able to help you determine whether the text you’re reading was written by AI or a human.

But there’s a but.

OpenAI notes it’s “impossible to reliably detect all AI-written text” and it has “not thoroughly assessed the effectiveness of the classifier in detecting content written in collaboration with human authors.”

Why we care. There are already a variety of AI content detectors. If these tools are able to do a fairly effective job of identifying AI-generated text, it’s safe to assume search engines can. Think: Google’s helpful content system.

Also, if you are publishing content, it may be smart of you to run any content submitted by your agency or freelance writers through these tools. We’re entering a new era – and you don’t want to publish anything that could potentially get your site flagged as publishing automatically generated content, which is against Google’s guidelines. We have already started using tools like this for all contributed content here at Search Engine Land.

What is the AI Text Classifier? OpenAI calls its new tool a “fine-tuned GPT model that predicts how likely it is that a piece of text was generated by AI from a variety of sources, such as ChatGPT.”

It has some limitations – most notably the results may be inaccurate. OpenAI primarily trained the tool on English content written by adults so it’s entirely likely there will be false positives. Also, as OpenAI pointed out, it’s easy to edit AI-generated text to pass as human-generated. As noted in its FAQ section:

• “The results may help, but should not be the sole piece of evidence, when deciding whether a document was generated with AI. The model is trained on human-written text from a variety of sources, which may not be representative of all kinds of human-written text.”

The results of the classifier. After you paste your text into tool (Open AI says you need a minimum of 1,000 characters, which is roughly 150-250 words) the AI Text Classifier will evaluate whether the text was generated using AI. You will see one of five labels:

• Very unlikely.
• Unlikely.
• Unclear if it is.
• Possibly.
• Likely.

Here were the results after I pasted in the text of this article you’re reading (human written, I swear):

And thankfully, the classifier thought it was “very unlikely” that this article was AI-generated.

Try out the tool for yourself. OpenAI’s AI Text Classifier. It’s free.

More coverage. You can read OpenAI’s announcement here: New AI classifier for indicating AI-written text. You can see more coverage on Techmeme.