The rapid development of AI writing tools has led to many positive changes. Businesses with limited resources can produce more content, writers suffering from writer's block can find solutions, and those who struggle to get their thoughts down on paper can finally explain their thoughts.
But there have also been some side effects, one of which is deception.
While we here at Originality.ai have no issue with the usage of AI-generated content and AI-writing tools like ChatGPT, we have made it our mission to become a world leader in AI transparency, ensuring that AI-generated text is identified and not passed off as human-written using our accurate AI detector. We don't think you should be able to generate text in Chat-GPT or another AI writing tool and then pass it off as your own work. That is why we have made our AI writing detector.
One such tool that claims to bypass AI content detectors is Undetectable AI, which claims to generate undetectable AI content.
In our test the output was able to pass detection but when the grammar mistakes were fixed it became detectable again.
An alternative AI writing aid that produces grammatical correct and undetectable AI content is Duey.ai
Note: although Duey.ai seems to be the best undetectable AI writer and student writing aid (citations, tutor etc) the content is still detectable by Originality.ai's AI Detector.
Although no tool is fully effective at bypassing AI detectors a 100% free tool that makes writing that is not able to be detected by most AI detectors is Undetectable AI although the quality significantly drop off.
So, we decided to put it to the test.
Here, we will discuss the following:
Undetectable AI is an AI rephrasing tool that claims to turn your 100% AI-generated content more human-like so that tools like Originality.ai and other AI content detectors believe it is a human-written article.
For content creators seeking a shortcut around the limitations of human scale speed, Undetectable AI may seem like an attractive offering
Some people see the recent AI advancements as an opportunity for positive change, speeding up the research process for creating high-quality content and helping plan articles,
However, there are also those who see it as an opportunity to cut corners, passing off AI-generated texts as human-written. That is why AI writing detection is so important, especially as artificial intelligence continues to advance.
Whether people are trying to create content at scale or using the AI paraphrasing tool to pass off ideas as their own, an AI detector is vital to ensure that work practices remain ethical and audiences know what they are consuming, whether that be AI or human-written content.
But the question remains: Does Undetectable AI actually get past some of the market-leading AI content detection tools, or does it fall short like other AI alternatives?
We decided to put the tool to the test by generating content across three different formats and then running it through Undetectable AI to see what happens.
For the AI content generation, we will use ChatGPT and focus on very common types of content for a broad understanding of the tool's features and shortcomings.
For this test, we imagined we were running a sustainable clothing company, and we wanted to create an article that talks about the value of natural fiber clothing over synthetic fabrics. This type of content is common for brands as it shows off their expertise and helps build trust with their audience.
However, if the audience were to realize the content was AI-generated, they may believe the brand is no longer trustworthy and question how sustainable it really is, giving them an incentive to possibly try an AI tool like Undetectable AI.
Here is our content:
Now, let's get started with the test.
Firstly, let's run our current content through Originality.ai to see if it gets detected as 100% AI-generated.
https://app.originality.ai/share/xylkj3bq60m2rhzg
And now let's run that content through Undetectable.ai:
And run that new content back through Originality.ai:
Now, let's do the same using the GPTZero Score tool.
Here's the score for the original ChatGPT content:
And for the Undetectable AI output:
And the same for Writer.
Original content:
Undetectable AI content:
Original content:
Undetectable AI content:
Original Content:
Undetectable AI content:
As you can see from the data, only Originality.ai was able to detect the original content, which was 100% AI-generated.
While Originality.ai immediately identified that the original content was 100% AI-generated, ContentAtScale and CopyLeaks gave it a 69% and 73% chance of being AI-generated, GPTZero said it was 50/50, and Writer even was as far as to determine the content was 100% human written.
Next, the results of the Undetectable AI content rewrites.
The closest any of the tools got to determining that the content is AI-generated is GPTZero, but the score is still an underwhelming 19%.
Originality.ai is the only other tool to identify that the content may be AI-generated, but only give it a 5% score. The three other tools determined the content is 100% human-written.
At first glance, the results are not great for content detection tools. While it's concerning that some of the tools couldn't detect that the original content was AI-generated, it is even more alarming that none of the tools identified the Undetectable AI content was AI-generated.
So, we did a little more digging.
One of the first things we noticed is that the content has many grammatical errors. In fact, here's the GrammarlyPro score:
97 grammatical errors and a 54/100 content score? Now, those results are starting to make a little more sense.
This is a common tactic that AI content detection tools use to "trick" AI detection tools, but ultimately, they are just tricking their clients.
By creating content that is grammatically incorrect, the Undetectable AI content does not seem AI-generated, as an IAI tool would not make so many mistakes.
But bad grammar is not the only issue with this content.
As you can see, the content itself is also poorly written, and the content is often wordy and unnecessarily complex, and in some cases, completely incorrect.
The content itself would still need a complete makeover to get it to a standard where it could be published, but let's see how it scores now that it is a 99/100 on Grammarly.
As you can see, even from this simple tidy-up, the results have changed significantly, except for the Writer score. All of the other tools now identify that the content is likely AI-generated, with Originality.ai immediately noticing this content is not human-written.
If we were to tidy up the language throughout, there are no doubts that these scores would continue to rise.
This information further proves the importance of transparency in your work, or you run the risk of getting left with poorly written content that may get past AI detection tools, but it certainly won't get past your target audience.
Undetectable AI allows you to test out the AI tool and its key features for free but then requires you to purchase a membership once you hit your word limit.
Undetectable AI is a tool that claims to turn your AI-generated text from detectable to undetectable. However, as part of our Undetectable AI review, we found that to not always be the case.
At the time of writing this article, there are no sure-fire ways to get AI-generated content past high-quality AI detectors like Originality.ai. The only way to do so is to use deliberate bad language, poor grammar, or combine AI tools with human-written elements.