It sometimes amazes me how little we think of consequences when it comes to new innovations because we’re too excited about the upside.
Case in point: the rise of LLMs like ChatGPT has completely changed the way we do work. It’s come to the point where a lot of people even use it for less-than-ideal purposes such as cheating. Because of that, AI detection tools were created, then innocent students started paying the price for false positive AI detection. And finally, AI humanizers were created to protect students from getting falsely accused of using LLMs.
And so the cycle continues.
For anyone who’s using AI humanizers, the biggest question is simple: does it actually pass AI detection?
Today, we’re going to find out. This is a head-to-head of Undetectable AI and Writer.com’s AI detector.
What is Undetectable AI?
As a writer and a student, I’ve grown accustomed to the fear of receiving a false positive AI detection. This is the reason why I consider Undetectable AI to be a godsend.
Undetectable AI is an AI paraphrasing tool that claims to make any writing avoid AI detection — a claim that they’ve backed up well. We tested Undetectable AI before against the most popular detection tools in the market today, and guess what, it passed. Undetectable AI also has a unique feature called output customization that lets you pick between different readability levels and use cases.
So, let’s address the elephant in the room: it makes sense that most people would consider Undetectable AI as cheating. I disagree, and so does Undetectable AI. There’s actually an ethical way of using this tool, which you can read here. If you want to know more about Undetectable AI, I highly suggest our review of the product.
What is Writer.com’s AI Detection Tool?
Before talking about their AI detector, we need to talk about Writer.com as a whole first.
To say that Writer.com is an AI writing assistant minimizes it a little, because it’s so much more than that. This platform aims to help businesses to write high-quality content at scale. Whether it’s recaps, FAQs, emails — anything. Writer.com can write it. Aside from that, they also provide support for automating business processes and integration with productivity apps like Slack.
Now, as for their AI detection tool, there’s not much to say. It’s free (well, up to 5,000 words) and doesn’t offer much in terms of features. So, since AI detection is the only thing it offers, it’s reasonable to assume that it’s going to be accurate.
… right?
How Does Undetectable AI Even Work?
AI humanizers like Undetectable AI have two methods that allow them to get past AI detection tools. Number one is an in-depth rearrangement of the original text and removal of certain AI telltale signs like common AI words (examples include delve, revolutionize, utilize) and lack of flow.
The second is adding intentional errors. Yep: misspellings, grammatical errors, missed punctuations — you name it. It’s true that this makes for a lower AI likelihood score, but it also makes your output so much more unreadable.
So let’s test it out against Writer.com’s AI detector …
Undetectable AI vs. Writer.com’s AI Detection: Accuracy
Test #1
Undetectable AI identified as human.
AI Likelihood Score: 0%
Test #2
Undetectable AI identified as human.
AI Likelihood Score: 0%
Test #3
Undetectable AI identified as human.
AI Likelihood Score: 0%
Test #4
Undetectable AI identified as human.
AI Likelihood Score: 0%
Test #5
Undetectable AI identified as human.
AI Likelihood Score: 2%
Test #6
Undetectable AI identified as human.
AI Likelihood Score: 0%
Test #7
Undetectable AI identified as human.
AI Likelihood Score: 0%
Test #8
Undetectable AI identified as human.
AI Likelihood Score: 0%
Test #9
Undetectable AI identified as human.
AI Likelihood Score: 0%
Overall Tally
Undetectable AI’s Likelihood Score |
|
So, What Now?
This is a bad look for Writer.com, and here’s the clincher: it’s not even the first time this has happened.
We already tested Writer.com’s AI detection algorithm earlier this year. The results? It’s at the very bottom of our true positive tests with only 18% accuracy. That’s almost half of the second least accurate AI detector we tested, which was ZeroGPT. And in that test, we included raw outputs from ChatGPT, Claude, and Gemini.
So, it really came as no surprise for me that Writer.com has low AI accuracy when it comes to only Undetectable AI outputs. I just didn’t expect it to be this low. That said, we can’t discount the effort of Undetectable AI in this win. It proved, once again, that when it comes to AI humanizers, nothing else comes close.