The unstoppable AI revolution has been sweeping the tech world by storm and shows no signs of slowing down anytime soon.
Tools like ChatGPT have been wowing everyone with its remarkable writing abilities. AI Writing resources are ramping up, and it's doing so pretty quickly.
While AI content is not necessarily bad, it may sometimes be useful to determine if a piece of writing was written with AI. While so many factors go into qualifying something as "AI writing" other than running it through a detector, if you're looking for a general idea about content origins, running it through these are really good options.
While you can't be 100% definitive about AI writing by running writing through a program, you could generally get a decent idea about its origins.
The more content you run by the same writer & at longer lengths can really give you a decent prediction as to where it comes from. No AI detector is 100% accurate, but good ones can act as a good prediction tool.
How Does AI Writing Detection Work?
I like to think about AI detection similar to how chess cheaters get caught. Both of these involve recognizing patterns and deviations that are indicative of unnatural patterns.
In the same way that chess cheating detection software compares players' moves to powerful chess engines, AI writing detection algorithms compare text to patterns and features found in both human-authored and AI-generated text.
Like a chess player whose moves consistently align with a top chess engine's recommendations, AI-generated writing often exhibits certain linguistic patterns, phrasing, or stylistic elements that reveal its artificial origin. In both cases, the detection systems consider factors such as context, complexity, and level of creativity to discern whether the subject in question — be it a chess game or a piece of writing — is genuine or artificial.
Human writing is inherently variable. Humans don't really follow procedural patterns when writing papers and telling stories. Of course there are syntactical and stylistic standards that should be followed, but generally human writing is unique.
The more unique, random, and varying levels of words throughout a document increase the odds of something being human-generated.
The Best AI Writing Detectors of 2023
Whether you're testing content in an academic, professional, or casual setting, there are different tools that make more sense than others. Each of these tools work, and they work well, but some help organize content better (for teachers or students trying to determine AI plagiarism) or for one-time, casual writing (like a job recruiter trying to find spammy cover letters).
Each of these are very viable options for testing any content, so final judgement is based on your use-case and detection intensity preference.
I really like Originality because it's been in the game since the start and has continuously updated its models since then.
Originality describes itself as an "AI Detector Built for Serious Content Publishers." It supports both AI detection and plagiarism. Aimed at content publishers, it really goes in depth to help you determine the origins of writing. Originality stores every check you do and shows you a detailed breakdown.
Built with AI experts, Originality was created solely to predict if something was written with AI. It was built testing a variety of models like GPT-4, GPT-3, GPT-2, GPT-NEO, GPT-J. Their accuracy rate has been shown to be higher than 94% on paragraphs with over 50 words.
In the following sample, you'll see Originality has highlighted a majority of the lines as being AI (which is true). It highlighted a single line in green (predicting human writing), but ultimately resulted in the article being 100% AI.
Originality can also scan entire websites, save scans, and comes with support to add team members. Pricing starts at $20 for 2000 credits (which can scan 200,000 words). If you want more information on them, check out our review where we go in greater depth.
My absolute favorite feature is the AI visualizer. Connect a Google Docs article to Originality and you can see a sped up visual of how an article was constructed. I've had some old writers submit articles to me that came back as 100% AI. I was doubtful, since I trusted them & I know you should only take these scores with a grain of salt, so I did a little more investigating.
I submitted the doc they sent me and visualized it with Originality and was really sad to see a bunch of copy & pasted ChatGPT writing submitted as original content. It works so seamlessly and is super telling:
Content at Scale
The second tool has to be the Content at Scale AI Detector. This is an extremely powerful tool for testing one-off, individual articles. Whether it's for educational, professional, or casual writing, Content at Scale can help a lot. Best of all, it's free!
It also tends to under-detect (in a good way). It won't really predict something as being AI unless it's quite confident.
The product behind this, Content at Scale, was built to create full-length AI articles at scale, and while doing so the team realize they need to create their own tool to make sure their own product doesn't get detected as AI. When I don't need to check a ton of writing and need something super quick, I use C@S.
To use the tool, paste writing into the input box and simply submit it for detection. Once finished, you'll see a human content score (predicting the odds that text was written by a human) and see a line by line breakdown of what parts of the content are suspicious or obvious AI. You'll get predictability, probability, and pattern scores, as well as an overall human content score.
In this sample, the article is unlikely to be written by a human. It was very predictable, had probable vocabulary choices, and followed a moderate syntactical pattern. You'll also see highly likely AI writing in red, medium in orange, and possible in yellow:
If you're looking for the best AI detector in the education sphere, check out Winston. Winston works as a great tool for checking if student and educational writing was written using AI. Super great for teachers.
It supports file uploading, optical character recognition (including handwriting!), team management, project support. Winston supports multiple languages, including english, spanish, french, and german. It includes a plagiarism scanner and can export content reports. Check out our full Winston review if you want to learn the details!
Winston seems to be the first AI detector specifically built for the education industry. You could use the tool for free with limited access before deciding if you want to use it professionally.
In the same sample of text, you'll see Winston marked the text as being almost surely written with AI.
CopyLeaks is a pretty common checker that most people have found pretty easy/straightforward to use. It's built by a company that offers a lot of other AI detection, plagiarism, and grading software – so it's quite enterprise.
It's completely free to use the detector which highlights where portions of an article are likely AI. You'll get a percentage when you hover over a selection. I have noticed it's quite strict though, so you might run into a bit of false positives. Use with caution!
Some AI detectors offer a lot of features but have no results to show for its effectiveness. On the other hand, GPTZero is the straightforward AI writing detection tool that may not be as feature-rich as the other entries, but is consistently trusted by companies and educators alike for its accuracy. No frills — just detection and, sometimes, that’s all you need.
GPTZero analyzes your text word-per-word and highlights identified AI content. It also outputs a confidence level score of AI content out of 100% to give enough leeway for potential false positives and negatives. Unfortunately, it has a 5000-character limit per detection.
Apart from their main dashboard, you can also access GPTZero using their API or Chrome Extension.
Sapling AI is a jack-of-all-trades. Their flagship product is an integrated language model-powered functionality for CRMs. Buried in their product list is an AI content detector — and a pretty good one at that.
Their AI detector is developed by researchers from UC Berkeley, Stanford, Meta, and Google, so it’s really no wonder why businesses continue to trust this company. Moreover, they also claim to have a 97% detection accuracy and, from my own testing, I have no cause to doubt that.
Unlike the rest of their products, Sapling’s AI writing detection tool is completely free. They’re also consistently updated, with new additions such as AI code detection, better accuracy, and an increase to 8000-word limit in the near future.
Crossplag also offers an AI detector primarily for educators as part of their plagiarism detection platform. This tool can analyze up to 3000 words per detection and features a free trial. After that, you must buy credits to continue using the product. 5000 credits cost €9.95, which you can use for 100 words per credit.
Crossplag has a scale that outputs how likely a text is AI-generated. However, the tool doesn’t highlight specific parts of text that are possibly AI, which isn’t really a deal-breaker but is something I actively look for in an AI detector.
AI Writing Detector Comparison Chart
These are some of the best AI detection tools on the market.
Remember not to base any major decisions off of any of these. They're not conclusive. You can't use these to prove anything, but they're a good start to helping detect AI writing.
What are your thoughts on AI writing? Are we already able to replace the level of complexity that humans can offer? Let us know your thoughts on any of these tools below if you have any suggestions or comments!