How to Detect Text Written by ChatGPT and Other AI Tools

30 November 2022. The day when ChatGPT was launched.

Everyone was growing crazy about it. Because it was something that none of us had used before.

Within just a year, 350 million people were using ChatGPT to write emails, website content, landing page copies, and a thesis.

But it raised a big question!

Since ChatGPT was trained on huge academics and other data, it had limited knowledge.

It was important to detect ChatGPT-generated text to see what’s unique and what’s generated from GPT.

In this article, we are going to discuss exactly the same thing. From this article, you will understand

• How can you detect the ChatGPT text?
• What are the challenges in ChatGPT text detection?
• How can you spot misinformation?
• Tools to detect ChatGPT text.

So, let's start with what is ChatGPT text and how you can detect it.

What is ChatGPT Text, and how can you detect it?

In simple terms, ChatGPT text means the content generated by ChatGPT. It was trained on human written content from the internet. So, it becomes hard to differentiate between human content and ChatGPT-generated content. People even ask ‘can anyone recognize if ChatGPT was used?’

But after all, it's still a machine that is trained by humans. So, there are few factors you can use to detect the content generated by ChatGPT.

1. Perplexity

This is a way to see how well a computer model can guess what comes next in a sentence. If a piece of writing is easy for the computer to predict, it might be AI-written because AI tends to follow predictable patterns.

2. Burstiness

This checks if the lengths of sentences vary a lot. In natural human writing, some sentences are short, and some are long. AI-written text often has sentences that are more similar in length, making it less "bursty".

3. Repetitiveness and predictability

AI often uses the same phrases over and over and uses words that are commonly put together. It makes the content sound less creative and unique. If you notice a lot of repetition or very common phrases, it might be a clue that the text was written by an AI.

4. Consistency and Coherence

AI can make text that sounds right the first time it is read. But if you look closer, sometimes the ideas don’t fit together perfectly, or the details aren’t quite right. This lack of deep understanding or small mistakes in logic can help you find the content generated by ChatGPT.

Detecting ChatGPT Text: The Caveats

False Positives and Negatives

Even the best tools can get it wrong sometimes. Tools aren’t perfect, and sometimes they can show human-written text even when the content is AI-generated and vice versa. It can happen because sometimes the text is tricky.

People Can Trick the Tools

Those who regularly write content know what are the loopholes. They can tweak the content in a certain format so that AI tools can’t detect it. If it happens, you will never get an answer to the most common question that everyone has been asking lately- “Did ChatGPT write this?”

Ethical and Privacy Concerns

Using these tools, especially in schools, can lead to big decisions like grades or accusations of cheating. This raises questions about whether it's fair or right to use these tools in these ways.

Technological Limitations

AI is getting better every day. So, the AI detectors might not be able to keep up with it. It can lead to results that are not accurate.

How to Spot Misinformation Within ChatGPT?

Double-check the facts

Always check the content for facts before submitting or publishing it. You can even ask ChatGPT for resources that it uses to provide content. If it doesn’t provide resources, try Google. If the resources found match the content in your text, that means it is correct.

Check for the latest content

Remember, ChatGPT has limited access to data. It doesn’t have real-time access to data. So, the text generated by ChatGPT might be outdated. It usually happens when you are researching content based on statistics.

Look for Logical Consistency

This is something you might have noticed if you use ChatGPT regularly. If you ask GPT the same question but with a little twist, it will provide different answers. This is something to look for.

Be skeptical of overly detailed answers

AI can generate stories with so many details, that are enough to convince humans that it is true. So, if you feel that this is too good to be true, always fact-check.

Use fact checking tools

You can use tools like Google fact-check tools to verify information provided by the ChatGPT. Again, it’s a tool, so you shouldn’t rely on it completely, but it’s worth trying.

Read more, How do AI Detectors Work

AI Tools to Detect Chat GPT Text

There are many tools available to detect ChatGPT text, but it's hard to say which one you can rely on. So, we took upon a task to check them ourselves. What we will do is provide the same content to the tools and see which one detects the content with the most accuracy.

So, we asked ChatGPT to generate terms and conditions for our website. Here is the screenshot of the response that we received from ChatGPT.

Now, we are going to test these AI detector tools with the same text.

1. AIchecker pro

As you can see, it shows that 100% of content is generated by AI, which is true. So, AI checker pro is something you can rely on. It is used by teachers, students, and marketers because of its ability to detect AI content accurately.

AI CheckerPRO is trained on 1000s of data sets which includes human text, academic thesis, and more. It uses advanced algorithms to detect perplexity, burstiness, repetitiveness and coherence.

Check Aichecker.pro and start using it for free today to detect AI text accurately.

2. GPTZero

GPTZero shows that 4% of text is generated by humans, which is not true, because the whole text is generated by AI. 4% might not seem like a lot, but we can not ignore the fact that it has a slight margin of error when detecting the AI text.

3. ZeroGPT

ZeroGPT shows 30% human content, which makes it the 3rd most inaccurate among all 4 tools that we have tested for AI content detection. Though it lets you text up to 15,000 characters for free, its AI detection capabilities don’t seem to do the justice.

4. Writer AI Content Detector

Writer AI detector shows 91% human-generated content. So, to put it in simple words, it's not something you can rely on. Though it offers 5000 words to check for free, which is the highest among all four tools, the results are unreliable. The whole content was ChatGPT generated, and if it shows 91% human-generated content, there is no point of using it.

Humans Are Still the Best AI Detectors

Tools are helpful, but we would suggest you shouldn’t completely rely on them. Ask yourself questions, whether you are a teacher, student, or marketer when reading content, ask if you were to write this content. Will you write it like this?

As we discussed, no tool is perfect and can generate incorrect results sometimes. So, using a mix of both the tools and logic can be more helpful.

After all, AI is trained on human content, and it is going to use the same writing tone that humans use. It has its own limitations, so look for them and decide which content you want to go with.

Conclusion

All tools have their pros and cons. Some offer more words to check for free, and some detect AI content more accurately. However, when we talk about accurate AI detection, AIChecker Pro is the clear winner. Other tools did detect the AI content, but it was not as accurate as it should have been.

It's because AI Checker Pro is regularly updated with the latest advancements in AI and is trained on large sets of data. It enables users to accurately track the AI content.

Don’t take our word! Try it yourself for free!

FAQs