4 Reasons Why AI Checkers Might Flag Your Writing Even If You Don't Use ChatGPT

The accuracy of AI writing detection tools is worrying. They often label human-written articles as AI-generated for reasons such as grammar, vocabulary, and style, causing great distress to students and others who rely on writing.

So, if an AI writing detection tool keeps identifying your writing, why? How to solve it?

Too perfect grammar

One way that AI detectors identify AI-generated text is by checking the sophistication of grammar and whether the writing primarily uses standard or common sentence structures. In theory, AI cannot make grammatical errors, but even the best writers can make small mistakes when writing. Likewise, if your writing rarely ventures into different styles and your grammar is too generic and lacks a personal writing style, this will trigger the AI ​​writing detector.

For example, I input a piece of text in this article into ChatGPT to generate text, and then pasted it into GPTZero for detection. The results showed that the possibility of it being written by AI was extremely high, reaching 100%.

If I remove some commas, introduce a typo, and make some minor style adjustments, GPTZero's score drops significantly to 81%.

Use vocabulary commonly used by AI

When we read a piece of work, many people will intuitively judge whether it is written by AI, such as paragraphs with empty content and words often used by AI, such as "depth discussion", "focus", "emphasis", “Propose,” “the world,” “endeavour,” and countless other words and phrases. This is evidenced by a significant increase in the term “deep dive” in academic research papers in 2023, which coincides with the launch of ChatGPT.

Using the same text as in the previous example, if I make some small modifications to the parts where the AI ​​detector is most sensitive, GPTZero's score drops further to 49% - a score that is considered "human-written", Although it wouldn't pass academic scrutiny. Apparently, AI detectors can be fooled easily with just a few edits, and this is just one of many examples that underscores its ineffectiveness.

I've been tutoring writing for a long time, and I've noticed that my non-native English-speaking students complain more often about false positives from the AI ​​detector. When creating large language models, AI companies often outsource data annotation to countries where English is a second language. Therefore, some of the words we associate with AI writing, such as "deep dive," may have resulted from data annotation by English second language speakers who used these words.

This exacerbates the fact that many people who speak English as a second language tend to understand the rules of grammar better than native speakers who learn English in a more intuitive way. Therefore, English second language speakers may face the double risk of false positives on AI detectors due to their correct grammar and vocabulary choices.

Using the AI ​​writing assistant will also trigger detection

The above two problems may occur even without using generative AI tools. If you make a serious attempt at original writing and use a writing assistant tool like Grammarly, your writing is more likely to be marked as AI-generated. This is definitely a gray area in academia, as these tools are technically generative AI assistants, and students often use them as a substitute for learning rather than as helpful aids.

In my tutoring experience, I have found that those who rely too much on Grammarly to revise their writing are most afraid of AI detectors detecting their work when they are submitted. When a large portion of someone’s text is rewritten by an AI assistant’s suggestions, a large portion of the writing is actually AI-generated. So when using tools like Grammarly, be careful if you rely on them too much, and use Grammarly's suggestions as learning opportunities rather than blindly accepting them.

The result of copying ChatGPT is true positive

Finally, and most obviously, if you actually use ChatGPT and barely modify its output, the AI ​​detector will almost certainly flag you; this is not a false positive. However, even if you make a serious attempt to write a completely original, unaided piece of work, perfect grammar and certain vocabulary and phrasing choices can lead to false positives.

The above is the detailed content of 4 Reasons Why AI Checkers Might Flag Your Writing Even If You Don't Use ChatGPT. For more information, please follow other related articles on the PHP Chinese website!