The rise of artificial intelligence has led to some pretty major advances in natural language processing. AI can now generate text that sounds just as if a real person wrote it. But along with the good stuff comes some risks too, like people mass producing fake news articles, publications unknowingly publishing content, or bots spreading misinformation online.
As AI gets better at mimicking human writing, it's getting harder to tell what's real and what's computer-generated. But there are a few tools coming out now that can supposedly spot the AI content.
In this article, we'll take a look at the top AI detectors in 2023 that can identify fake text.
As generators keep getting better and better, you can use these tools to defend against the dark side of AI content creation, and keep things human.
Besides just listing AI Content detectors, we'll also run the first paragraph of this ChatGPT (3.5) generated content to show you how each detector scores the output.
GPTZero is a venture-backed AI checker that runs its own model to check if content was written by (completely or partially) by ChatGPT or other tools.
They recently raised $3M to further their mission, which should help them innovate further on AI research, UI design, and better accuracy.
GPTZero's accuracy is hit or miss. It won't falsely identify AI content, unlike some other tools out there which is good, and will stray to the side of caution. In this example, we inputted 100% AI content, yet it only noted that there was a 55% probability that the text was written by AI.
This tool seems to get more accurate as more AI content is added. We only submitted a single paragraph (1333 characters) for our test.
The best part about the GPTZero AI detector tool is that it's 100% free.
Due the the accuracy issues, this is a great option to gut check content, but shouldn't heavily be relied on, as illustrated above. But because it's 100% free (and you don't even need an account), it's a great tool to use quickly.
They do offer pro plans for business people and educators (special plan for teachers) which has features like bulk uploading
There are many similar tools online that use the API from GPTZero for their platform as well. Most of the free, ad-supported offerings are simply using this API and adding their own branding (looking at you ZeroGPT).
Of the detectors on this list, Originality AI is the most accurate that we've found. It'll deal with paraphrased content, raw ChatGPT outputs, and everything in between. In August 2023, they released the 2.0 version of the AI model that powers their platform which is even quicker and more accurate.
Originality AI's accuracy is industry leading. Here we directly paste a AI generated paragraph, and it scores it as 100% AI. In fact, the accuracy is so good, this is our go to tool for checking author submissions for this and other blogs.
There's a very limited trial of 200 credits, making this basically a "paid only" tool. You can buy credits for $30 one time, or subscribe to a pro plan starting at $14.95/mo.
Winston AI is a newcomer to the "AI Detector" tool arena. It's incredibly well designed and easy to use.
Winston gives our 100% AI paragraph a "human score" of 0%. From our testing Winston is very accurate, but has sometimes given false positives for the "possibly AI generated" highlighting.
There's a free trial of 2000 words. Paid plans start at $12/month, billed as $144/year. Paid plans also get access to a plagiarism checker.
AI Detector Pro is another checker, aimed for more professional users. It offers tools like bulk import and checking a URL.
Compared to others on this list, it also offers a unique feature called "AI Eraser", which does as you expect - helps you remove detectable AI from a document by suggesting more 'human' outputs.
The tool is very accurate, and identifies our demo AI content as "most likely written by AI". It's also a bit more 'focused' than some of the other tools in this list, pointing to the exact sentences (not just paragraphs) where it has identified AI. This is helpful if you have a large piece of content where only portions have been written by AI.
Monthly pricing comes in at $13.99/mo for 100 AI reports.
Turnitin is a company that helps school track assignment submissions and combat academic integrity violations. With the advent of AI, AI writing can fall into the bucket of "academic integrity violations", and Turnitin has their own detector tool to combat it.
Built into the assignment submission workflow, educators can now see if content was written with AI.
We can't speak on the accuracy or pricing as the tool is only accessible to schools that have signed high-value, multi-year contracts can use this.
Not really, here's an example.
The submitted text is 100% AI generated, yet ChatGPT says that it's likely written by an AI.
If you're looking for a prompt to help turn ChatGPT into an AI Detector, you can try this: "Check the following content to see if it was created by AI (LLM from OpenAI or Google). Before answering, consider the words, sentence structure, and other markers in the content."
But, as the AI itself states, "AI-generated content has become increasingly sophisticated, and distinguishing between human and AI-authored text might become more challenging as AI language models continue to improve".
Here's a notable development.
OpenAI used to have their own AI detector tool, called the OpenAI Classifier (here's the archive link if they remove the page). However, the company took it down in July 2023, noting that it wasn't really accurate and that they're working on other methods of AI detection.
Yes and no.
AI Detectors can be beat. In a sense, it's a cold war between the LLM providers (like OpenAI) and these detectors. Every time a model is upgraded, these detectors must change to be able to detect their outputs.
Furthermore, there are several tools that claim to (and consistently do) beat any of the AI detectors on this list. We've tested a few of them and they definitely do what they say. With that said, people need to pay for them limiting the number of
However, the real purpose of checking for AI content is typically to determine if somebody directly copied / pasted from the tool and submitted it to you as their own work. These tools are great at identifying "raw" AI content and will help prevent this.
As the widespread use of AI to generate content continues to grow, so does the need to identify it. This article provides an up to date list of the best AI detectors in 2023. Each of these tools were tested by us against real AI content, and you can see the accuracy in the screenshots above.
Do you have any other suggestions, comments or questions? Leave them down below.