Academic integrity is a critical factor in education, and the introduction of AI writing tools haven't helped. AI tools like ChatGPT pose major threats to the modern education system. It's becoming easier and easier for students to use AI writing tools to pass off fully AI-generated work as their own. We've entered a grey zone where educators seem to be facing a constant battle to detect and prevent AI plagiarism.
With this recent AI boom, tools have popped up that help detect the presence of AI writing by finding patterns in submitted text. With the use of advanced algorithms and natural language processing models, AI plagiarism detection tools are meant to quickly identify instances of AI plagiarism regardless of assignment size. These tools are seemingly becoming an essential tool for teachers and educators across the world, in ensuring the authenticity of student work. Additional issues arise as teachers have to strive harder to ensure a fair academic playing field is set for all students.
In this article, we will explore the best AI plagiarism detection tools for teachers and educators in 2023. These tools will help educators ensure that the work their students are submitting is original and authentic. We've tested each one of these detectors thoroughly on an abundance of both short and long-form content. As new detection tools and technologies are released, we'll test them and replace them on this list.
How Does AI Writing Detection Work?
Although not definitively accurate, AI writing detection tools try to re-predict the same text you're trying to test. Because AI works on patterns, it generally produces fairly consistent text structures. Currently, you can't have AI accurately write things like recent product reviews or creative analytical opinions, but AI is great for analyzing static content like historical events, books, and textbook content.
And because humans are so versatile and complex, human writing doesn't include as many predictable patterns & syntaxes. The best writers often have an unpredictable vocabulary that engages and questions their audiences in ways Artificial Intelligence cannot currently do.
AI writing follows formulaic patterns of what it was trained on, since that's what it knows best. The more accurate a detection tool can recreate the same sentences you are trying to detect, the higher chance what you're looking at has been written by or at least assisted by AI. If you want more of an in-depth explanation, check out our more technical explanation.
The Best AI Writing Detection Tools for Educators in 2023
Over the last few months a few unique AI detection tools have stepped up out of nowhere as a result of monster tools like Jasper and ChatGPT. Each detection tool works slightly different & carry certain purposes over others. Convenience is also a big issue, especially if you're testing 30 pages of essays each day. With all of this in mind, here are the current best AI text detection tools currently on the market.
1) Originality (best for lots of writing)
The first tool on the list is going to have to be Originality. I've been using this AI detection tool for about 2 months now to check various types of content (academic papers, business industry reports, and online blog posts). It's fast, easy, and very accurate. Originality claims to be made for serious content publishers over those in academia, but its UI and flexibility are vasty ahead of other tools, especially when checking a ton of submissions at once.
Originality costs 1 cent per hundred words, making it an affordable option if checking many papers in a short time. To put this into perspective, to check 30 essays each around 1500 words, will run you about $4.50.
Compared to other tools, Originality also includes a percentage in its prediction score. You'll see something was 2% likely to be AI and only 98% likely to be human written (these will add up to 100). Here's what it looks like in action:
A great feature about Originality is line-by-line highlighting. You'll see sections marked in orange which represents it might be written with AI. Green sections show human-writing. The more text you enter, the larger sample size Originality has to work with (increasing the reliability of your result).
As you can see below, a decent amount of the paragraph was highlighted in orange, but as a whole the article is in the clear. (This is true, I tested it with a college academic essay!)
2) Content at Scale (best detection quality)
The next tool we'll talk about is called Content at Scale. If you're looking for the most accurate detection tool, this is probably it. I've tested old academic essays, ChatGPT responses, and random articles I found online and have found the tool holds up very well. This detection software is also free, it's just a bit harder to test dozens of articles quickly with it. I would give this tool the highest AI detection score as it's based on a ton of advanced language learning models. The company behind this actually creates fully written, complex website blogs; so they built a detection tool to see if their writing is able to be detected by AI (it's generally not!). Here's a few examples of an academic research paper, poorly written paper, and ChatGPT answer:
3) GPTZero (individual academic content)
The third resource that could be used in detecting AI writing is GPTZero. This tool was actually created by a student at Princeton in early January. It works by assigning text both a perplexity and burstiness score. Perplexity measures how random the text is while burstiness measures its variation. The higher both of these values are, the more likely the document was human-produced.
The tool also includes feature that will highlight text that might be written with AI (if it's only a few lines in an essay, compared to the entire document). Again, these tools should be used with your own judgement, but it's a good initial starting point if you detect something was written by some AI-writing software.
You can use GPTZero by uploading a full document (as a pdf, docx, or txt) or pasting text directly into the tool. Once uploaded, you'll see the predicted result in large bolded text.
You can see what text had been predicted as likely AI if you scroll down. This text will be highlighted in yellow and can appear at any point or sentence throughout a submission.
You'll also be provided with the perplexity and burstiness scores of the submission, but these scores don't really tell you anything beyond the scope of their definitions. It's best to use GPTZero based on its final prediction & color flagging.
4) Official OpenAI Classifier (made by ChatGPT creators)
The last tool we'll recommend is the official Text Classifier made by OpenAI. This tool was released very recently after the company was met with criticism as they released one of the most groundbreaking recent technological innovations in recent years, with no way of detecting what was made with it.
The classifier is very easy to use, and just requires a free OpenAI account. To use, paste a section of text you want to test in the input box & press submit. You need to submit a minimum of 1000 characters (about 200 words) for the tool to run. Like any AI detection tool, the more words you provide it to test, the higher reliability the tool becomes. You'll get one of five predictions from the detector that include:
- The classifier considers the text to be likely AI-generated.
- The classifier considers the text to be possibly AI-generated.
- The classifier considers the text to be unclear if it is AI-generated.
- The classifier considers the text to be unlikely AI-generated.
- The classifier considers the text to be very unlikely AI-generated.
I haven't had the greatest success using the classifier, but it's also new. I've found that it detects AI as human quite frequently, and human-written text as being AI. It is strange that OpenAI had such a long head start of the rest of the tools on this list yet still released a tool that isn't highly accurate (especially because they have the data ChatGPT was trained on!)
5) Passed AI
Passed.AI is a newer AI detection tool that is specifically for teachers. The model is trained on GPT-2, GPT-J, GPT-NEO or GPT-3 to detect AI writing based on patterns.
They also have a Chrome Extension for quick-checking documents, which you can even do inside of Google docs. A unique feature about passed is how they let you "replay" a Google doc to see edit history. This is great for auditing an entire document based on how it was written
The company claims that AI Detection alone is NOT enforceable. But when combining AI detection with a document history audit will provide an unparalleled confidence in AI detection. All you need is edit access to the Google document the student submitted and Passed.AI can provide you a detailed audit.
- "Replay" at 10x speed
- See the number of contributors and the size of their contributions to the text
- See if the words/minute were natural or not
- No software for students to install
The actual AI detection page is similar to any other software. To use, paste suspected text into the scanning box and start scanning. You'll get a percentage score indicating the likelihood a sample was written with AI. A score of 100% Not AI and 0% AI should be thought of as "We are 100% confident that this content was created by a human."
Final Thoughts
All of the tools on this list serve their specific purpose. Before going all in on one specific AI detection tool, you should carefully consider the features and use cases for each tool, as well as features that would best align with your classroom and school standards.
Remember to take the concept of AI detection with a grain of salt as there are many ways to help detect content that include your very own judgement. Keeping all these things in mind, you can always test out various detection tools to see what fits right for your assignments and workload.
Have I missed any of your favorites on this list? Which of these detection tools have you had the best luck with? Let me know in the comments!