What are AI Detectors & Do They Work?

Many AI Detectors claim that they can let you know, with a certain level of accuracy, whether or not text has been written by a human, or by an artificial intelligence software.

However, these detectors are rather controversial. Some people claim they don’t work, and others even go so far as to call them a scam. 

Still, school presidents, marketing managers, teachers, and more, have penalized people for writing content that gets flagged by AI detectors. So how useful are they really?

What is an AI Detector, Exactly?

An AI detector is a software tool designed to identify content that is generated by artificial intelligence (AI) systems. Essentially, it distinguishes between human-created content and content that is generated by AI algorithms like ChatGPT. 

How Do AI Detectors Work?

AI detectors look very similar to plagiarism checkers. They assign a percentage score: 0% means that there is a 0% chance that the text was written by AI. 100% means that the detector is 100% sure that the text was written by AI.

So, when it comes to AI detectors, we want low scores.

This percentage model can be misleading. A score of 40% could make it sound like 40% of the article was written by AI, but really it means that the program is only 40% sure that the text was written by AI. (And what does that mean? Nothing, really.) Some AI detectors even highlight certain portions of text as if those were the parts of the text that were written by AI. 

For example, a teacher who has used plagiarism checkers before might be worried about these percentages, because if a paper is 40% plagiarized, that would be grounds for disciplinary action. But a 40% score on an AI detector can be unclear and subjective. 

AI Detectors Examples

A few examples of websites that claim to be able to detect AI-written text are:

• Originality.ai
• Undetectable.ai
• GPTZero
• Quillbot

Are AI Detectors Reliable?

Here is a screenshot of an AI detection result from Originality.ai, which claims on its home page to have “99% AI Detection Accuracy.”

The program is about 70% certain that the text was written by AI, which means that there is a very good chance that the text was written by AI software. A person who wrote this text could get in trouble at school or at work for such a high AI score.

There’s just one problem: The text that Originality.ai scanned “with 99% accuracy” in the photo is the first 100 words of The Art of War by Sun Tzu, which was written in the fifth century B.C. So, I think we can be 100% certain that it was not written by ChatGPT.

This leads us to the principle question: Are AI detectors accurate? No. 

OpenAI confirmed that AI detectors do not work. Cornell University published a paper pointing out that AI detectors are biased against non-native English speakers. Still, university professors have failed entire classes of students because ChatGPT claimed credit for their work when asked. 

That said, some people claim that there is credibility to AI detectors, they just don’t offer answers that are as clear-cut as they might appear. 

These programs tend to have a low rate of false negatives but a high rate of false positives. In other words, if an AI detection software says that something is not AI, then you can be pretty sure that it’s not AI. But if it says something is AI, then that often doesn’t mean much at all. And if someone accuses you of using artificial intelligence because you got a high percentage result from an AI detector, then ask them if they apply the same logic to Sun Tzu.

So, how can I know if a writer or student is using AI to write?

Look at the Google Docs Time Stamps

If you really want to know if someone is using artificial intelligence to write their work, then you can request that they write in Google Docs and provide you with the link. That way, you can go into the document history and see the time stamps to check how long it took them to write it.

Use the Originality.ai Plugin (But Ignore the AI Score)

Originality.ai, although its AI-checking results are inaccurate, has a Chrome plugin that allows you to view a recording of the writing process of any Google Doc. That way, you can check if someone pasted in a chunk of text at any point from an external source. 

Check for Plagiarism

That said, you should primarily check writing for accuracy and plagiarism before checking for AI. While ChatGPT is what brought AI into the public eye, it still makes a lot of mistakes, such as copying large chunks of text from many sources across the web and hallucinating. There are other businesses (like TypeTone) that offer a higher quality product and don’t plagiarize.

While a person might plagiarize by taking chunks of text from a few different sources, lower-quality AI bots tend to plagiarize by copying small chunks of text from many different sources. But when those little plagiarized bits add up, a big portion of the writing can still be plagiarized. And while we’ve established that a high AI score doesn’t mean much, a high plagiarism score is a serious problem in most academic and work environments. 

Remember: A Low Score Doesn’t Mean Good Writing

Remember the saying: “To err is human.” Humans make mistakes, and AI detectors know this. That means that if a paper is riddled with grammatical errors and awkward text, then that might mean that it gets a better score from the AI detectors. Sometimes, when text is grammatically accurate, it is more likely to get flagged as AI-written.

AI was made by humans, and sometimes that means that it has human biases. For example, AI detectors have been accused of being harsher judges when scanning content written by people who speak English as a second language. Nonnative English speakers might write in a more international, neutral English that relies less on slang and local sayings. AI detectors might flag this as more likely to be written by ChatGPT.

If you use software to help you proofread and check your grammar, like Grammarly, then your writing will be more likely to get flagged as AI-written as well, for many of the same reasons mentioned above.

In this sense, it can seem like AI detectors might actually punish writing for being more grammatically correct. So make this mental note: A low AI score does not equal good writing, and a high AI score does not equal bad writing.

Not All AI Programs Are Created Equal

At TypeTone, we recognize the importance of expressing the uniqueness of your tone and voice. Artificial intelligence does not have to produce uncreative, plagiarized, inaccurate content. Our digital workers and chatbot are completely aligned with the most modern EU tech rules, and we show a dedication to quality that goes beyond shoddy promises. 

Typetone provides AI employees and an AI chat tool that is perfect for modern marketing teams. Our AI software is just better–it’s safe, follows the rules (GDPR compliance, check!), and matches your brand perfectly. 

Give Typetone Chat a try and see how it can supercharge your marketing objectives. Schedule a demo today!

‍