AI has gained popularity across the globe. With the introduction of ChatGPT and Bard, AI has reached a new level that was never expected before. With AI generators, you can now create content of almost any type in seconds. If you're a blogger, you can write 1500 words posts in minutes. If you're a digital marketer, you can write ad copies with just a few simple clicks. AI tools have changed the world already and this is just the beginning!
Image by Joseph Mucira from Pixabay. Modified using Canva.
Ai tools have been out there for years but a breakthrough like ChatGPT was never witnessed before. Now comes the important question, are AI tools legal to use? The answer to this question is not very simple. If used within limits and not with the purpose of manipulation or cheating tactics, AI tools are a great way to boost productivity.
However, many platforms forbid or discourage AI content and to make sure that a person is not using AI generators, multiple AI content detectors have been launched on the internet. Some are free, and some are paid. If you're a blogger planning to monetize your content with Google Adsense, we highly recommend you avoid AI generators to create automated content that is not unique at all.
Even if you're a school student, AI content can help you solve your assignments but I don't think any school would allow that. The same is the case for university students who are planning to apply abroad for higher studies and are using AI generators to sound excellent for the selection committee.
So to overcome this problem and catch AI content, AI content detectors were developed. As I explained two scenarios in which AI content can prove to be troublesome, AI content detectors aim to differentiate between human-written content and AI-generated content to resolve this problem.
But the question is, are AI content generators really accurate enough to be trusted?
Understanding How AI Content Detectors Work?
AI content detectors analyze the textual inputs to give an outcome that will allow the user to decide whether the content is bot-generated or human-written. In simple words, when you write a piece of content and put it into an AI detector, the detector will go through the text and analyze its perplexity, sentence structure, word patterns, and personal touch to judge whether the text was written by an AI blog writer or was it actually written by a human.
When a human writes an article, the article will have some grammatical mistakes or a weak sentence structure in some cases. Even if you're a very professional writer, you still can't be 100% awesome while writing something. At least that's how I feel. An AI content writer analyzes this concept and decides whether the content is written down by a bot or a human.
Moreover, normally when a person writes an article, there's a personal touch to it. For example, when I'm writing this article, I'm planning on sharing my personal experience with AI content detectors. However, ChatGPT or any other AI tool is unlikely to do so with a simple command of writing an article on some topic. This gives a chance to detectors to identify AI-generated content with accuracy.
Having explained how an AI content detector works, now let's move on to my experience with AI content detectors.
My Experience with AI: Can We Rely on AI Content Detectors?
When I first heard about ChatGPT, I thought it was just another tool that was being exaggerated to gain some visitors. But then I decided to give it a try and I was amazed at how accurate it was and the kind of possibilities it brought with it. Since I've been blogging for a long time, the first question that came into my mind was whether AI generators were legal to use for blogging or not?
I got the answer to my first question that for now, it was better not to use AI tools for automated content as it might be a violation of the platforms that I currently use. Now the second question was even trickier, how can I detect AI content since I've hired writers to write content for me. This was the question that made me go nuts since I had no answer at that time.
Fortunately, soon after the launch of AI generators such as Jasper and Writersonic, AI content detectors were introduced. I was kind of relieved that now I'd be able to detect AI content and won't pay for bot content when my goal was to publish original content based on human experiences.
AI Detector Tools That I Used
To begin my journey with AI detector tools, I started exploring different options and I selected the following tools to begin with:
- GPTZero
- ZeroGPT
- CopyLeaks
- Content at Scale
Whenever my writers wrote a new blog post, I simply put it in the AI detector tools and tested it for AI content. My experience was that all AI content detectors gave a very similar output so I was glad that I found no discrepancy there!
However, after checking a few articles, I found out that these detectors were not as accurate as they seem to be. And that's when the reality hit me.
Inaccuracy of AI Content Detectors
I remember one of my writers made a complaint to me that she had written all article herself but the AI detector showed it as an AI-generated article. I first found it difficult to believe her but then I decided to experiment with it myself. By using different writing patterns and different tones, I tested multiple AI detectors to analyze the outputs they gave me. To my surprise, AI detectors were not as accurate as I thought they were!
Consider the following screenshot:
Screenshot of ZeroGPT Result Box. Reuse is allowed with proper attribution and a link back to us.
This was an article that I wrote myself and is still published on this website. The article was based solely on my own experience and not even a single word was generated by any AI tool. I used ZeroGPT to analyze the text you can see in the image above and it showed a 42.75% chance of it being an AI-generated text.
Have a look at how ChatGPT responded when I asked whether this text was written by AI:
Screenshot of ChatGPT response in reply to me. Reuse is allowed with proper attribution and a link back to us.
Why AI Detectors Are Not Detecting Accurately?
To be honest, I don't have a valid answer to that right now. Maybe the Ai detectors are programmed on certain criteria which is not as strong as it should be. Or maybe, some patterns of words can confuse the tools into thinking that the content was written by bots instead of humans. To understand why AI content detectors were not working properly, I decided to use different combinations of text to experiment.
Since I didn't have any official words from the developers to support my claim with a piece of evidence, I decided to do my own research, and here are my findings:
- The more creative you get, the higher chance of your content being labeled as AI-generated.
- If you write search engine optimized content with good keyword usage, your content is likely to be confused by AI tools.
- Using content improvement tools like Grammarly can cause the AI detectors to give out a false output.
- If you make spelling mistakes, your content will be labeled as AI (in my experience).
- If you're using a definition from the internet and even citing it to the original author, get ready to be labeled a bot!
- The article that had 0% plagiarism were labeled ai AI (some or most parts of them) by most tools in my case.
These are some scenarios that I personally tested to see how AI content detectors really worked. What I feel is that AI detectors are not very advanced yet and based on the qualities that a human may lack, AI tools make a decision. If someone is good at SEO, that doesn't mean he used AI tools to generate a 3000 words post. Sadly, AI might think otherwise.
How to Detect AI Content Accurately?
Well, as you would've known already by now, there's no accurate way to detect AI content and you pretty much have to design your own guidelines to make a better guess. As far as my strategies are concerned, I've set a percentage for acceptable AI content as shown by the tools, however, the original policy that I have established for my websites has a 0 AI content policy for my writers.
Since AI was not a problem before the introduction of AI content writing tools, I didn't have to change or modify my policy but after the introduction of so many automated tools, the policy had to be modified. You can use the AI detectors mentioned in this article to analyze the likelihood of content being generated by a bot but the final decision might not be 100% accurate for sure.
It is better to use software that captures the activity of your writers while they're writing, as done by many companies like Upwork. This is the only accurate way of detecting any suspicious activities being carried out by the writers. It is important to note that you'll need the full consent of the writers to use any such tools.
This was all for today. If you found this case study helpful, do share it with your friends and family. Keep visiting us for more interesting content!