Is It Possible to Bypass an AI Detector? | Tips and Prompts
The idea of using AI tools for writing assignments is tempting to most students, but is it safe?
These days, Turnitin and other similar software that schools use to check student submissions can detect not just unoriginal content but AI writing, too. It’s not without errors, and false positives aren’t rare. Still, submitting an AI-generated essay is a huge risk, at least without substantial editing.
If you do practice AI-assisted writing for schoolwork, please use the most accurate free AI detector you can find. Edit every flagged passage and triple-check your work afterward. We can help you with a few helpful prompts and strategies for bypassing AI detection.
How do tools detect AI text?
AI detection isn’t an exact science, largely because the creators of detection tools aren’t transparent. This makes sense, as the more the public and creators of AI writing tools know about detection strategies, the easier it will be to come up with solutions for bypassing them.
That said, there are a few key tell-tale signs of AI-generated writing that basically any paid or free AI detector tool is programmed to look out for:
These are only some of the giveaways of AI-generated content. Most detectors use them to decide whether to flag the text, but the list isn’t complete. Advanced detection software can have additional strategies for identifying AI-generated writing that the general public has no idea of.
Does AI detect AI-generated content without errors?
No, AI detection is never 100% accurate. Detectors can both miss AI-generated passages and flag human-written text as AI-generated, which is referred to as “a false positive.” Even Turnitin, which is universally recognized as the gold standard for plagiarism and AI detection, can be wrong.
According to Annie Chechitelli, Turnitin’s Chief Product Officer, the sentence-level false positive rate for Turnitin’s AI detection is about 4%. Keep in mind, though, that the actual number might be way higher. There are no official research results released for now.
If you’ve ever tried to check your work with more than one AI detector online, you know that it’s common for different detectors to show different results. You can run the same text through one detector and get a 95% human-written score, and then recheck it somewhere else and get a 55% instead. This is the best proof that AI detectors aren’t always correct, at least for now.
What about ESL writers?
Another problem with the reliability of most AI detectors is that they are known to be biased against non-native English speakers. In a 2023 study conducted by Stanford researchers, 89% of TOEFL essays were flagged as AI-generated by at least one of the tools used in the research.
A few years have passed since then, and the developers of AI detection software have been actively working on minimizing the bias against ESL writing. They’ve achieved some success, but you still need to be extra-cautious about AI detection if English isn’t your first language.
How to make any AI detector tool believe that your text is human-written
The laziest way to bypass AI detectors is to use a humanizer, but it’s not the best idea. Most humanizers make text weird, unnatural, and hard to read. The person reading the result will either think that your command of English is questionable or instantly realize that you’ve used a humanizer.
Instead, if you use AI writing tools for school assignments, we recommend editing them heavily and triple-checking them with several high-quality detectors to make sure the results are consistent. This step-by-step guide will help you edit AI-generated work.
Read the text out loud to catch monotonous passages and repetitive syntax
Change sentence structures to be less uniform and predictable, and make sure that you have a good balance in terms of sentence length.
Review all transitional words and phrases
AI writing tools overuse transitional words, such as furthermore, moreover, in addition, etc. Of course, you can’t remove all of them, especially in academic writing, where they are expected. But try to substitute at least some of them with more organic, logical transitions.
Remove excessive hedging
Again, a certain level of hedging is expected in student writing. You can’t sound too confident in anything, especially if the topic is controversial, with multiple conflicting perspectives. Still, review each hedging word and phrase to make sure it’s actually necessary.
Get rid of the meaningless words that are universally considered AI markers
Words like robust, cutting-edge, leverage, and the like don’t really mean much, yet AI models love them. Find a good list of AI-typical words and do your best to avoid them.
Make sure your writing reads as one coherent work with one central idea
AI-generated texts often read like a string of disjointed ideas united by nothing more than a few transitional words between paragraphs. Make sure this isn’t the case for your work.
The “high burstiness and perplexity” prompt
There is one more strategy that can help bypass AI detectors – the “high burstiness and perplexity” prompt. If you ask the AI model you use to write using high burstiness and perplexity, your text won’t be monotonous or dull in the typical AI fashion.
To give you an idea, here’s how a passage from a text written with the “high burstiness and perplexity” prompt can look:
“Exercise does more for the mind than most people expect. One day it’s a brisk walk that clears the mental fog; another day it’s a heart-pounding workout that shakes loose a week’s worth of stress.”
As you can see, the problem with this prompt is that it makes writing completely unfit for academic assignments. The text you get is fine for creative writing, but not for a research paper or academic analysis. So, while the “high burstiness and perplexity” prompt can trick some AI detectors, it’s not a particularly helpful strategy for students.
Can the most reliable AI detector get past the “high burstiness and perplexity” prompt?
Also, the “high burstiness and perplexity” prompt isn’t as bulletproof now as it once was. It used to be a sure way to bypass AI detection, and it worked on most, if not all, detectors. This isn’t the case anymore.
You’ll likely be able to fool a random AI detector online, but not the most advanced tools. This prompt has been known for a few years now, so the best detectors are specifically trained to be able to identify AI-generated writing despite “high burstiness and perplexity.”
Conclusion: Do you need an AI detector for essays?
As you’ve probably realized by now, AI detection is complex, with tons of variables involved. We don’t even know a lot of them, because the developers of AI detection software aren’t exactly eager to reveal their secrets.
Nonetheless, in most cases, you can bypass most detectors by heavily editing AI-generated work. Watch out for burstiness, perplexity, repetitions, and hedging. But make sure you’re using not just one free AI detector for essays but a bunch of them. Different detectors often show widely different results, so it’s better to run your work through at least 3–4 tools.
Also, as AI detection software is continuously getting better, you’ll need more and more editing to be able to trick detectors. At some point, it might actually be easier to write your assignments without AI tools than to edit AI-generated content enough for it to bypass detection.
Vincent van Vliet
Vincent van Vliet is co-founder and responsible for the content and release management. Together with the team Vincent sets the strategy and manages the content planning, go-to-market, customer experience and corporate development aspects of the company.
