The free release of powerful AI tools like ChatGPT means that writing an article, email, blog post, or essay is now as easy as typing in a simple prompt.
That's amazing for people who want to easily create text for marketing, generate compelling social media posts, or otherwise use this new technology in a beneficial way.
At the same time, though, it's a problem for a wide variety of fields. When consumers read an article on a website, they have no way of knowing whether it was written by an expert or an AI system pretending to be one.
Likewise, professors and educators have no way of knowing if their students' papers or essays were actually written by the student, and publishers now struggle to tell whether writers are creating their own content or using crude AI tools to generate thousands of low-quality articles.
The good news is that just as quickly as these AI systems have advanced, powerful tools to detect AI-generated text have kept pace. Large companies like Google are already using those tools, and both free and paid versions are available to smaller-scale business people and consumers.
This blog post shares a look at the current state of tools for detecting AI-generated text, and dives into some of the reasons why detecting this text is important. If you don't know where to start with AI writing, you might want to check our this AI writing course.
To be clear, AI-generated text is not necessarily bad. Advanced AI systems can generate real insights and genuinely helpful content. The company Content At Scale, for example, uses AI and other tools to produce blog posts that can be genuinely enlightening.
That said, however, there are many situations where AI-generated text is a problem.
Phishing scams, which use email to solicit personal information from victims, have existed for decades now. With AI, though, these emails are now much more effective.
Data shows that scammers are using AI to mimic the style of writing of prominent individuals and then sending personalized phishing emails to their contacts and colleagues.
For example, a scammer often train an AI system using publicly released text by a prominent person, such as a CEO. They can then use that AI to send personalized phishing messages to the prominent person’s community, mimicking their style of writing and asking for money transfers that actually go to the scammer's account.
Because AI is extremely good at mimicking specific people's writing styles, these new nefarious phishing emails can be incredibly convincing and powerful. Detecting them is important for cyber security and personal security.
The expository essay is a cherished part of the educational process. Educators use it to assess everything from reading comprehension to language learning.
A certain number of students have always been motivated to cheat, purchasing essays from essay mills or downloading them online. Plagiarism-checking tools like those from Grammarly have largely stopped this practice.
With AI systems such as ChatGPT, however, that's changing. According to The Atlantic, these systems can easily generate convincing expository essays in a few seconds. Especially because most students are novice writers, AI-generated essays may seem better and more convincing than an essay written by actual students.
For educators to assess their students properly, they need to know whether text submitted by students was actually written by AI.
For search engine optimization professionals, the appeal of AI writing systems is obvious. Creating content is hard and expensive, and these systems appear to offer an easy way to generate realistic writing.
The issue is that, from an SEO perspective, low-quality AI-written content is a major problem. Google has made clear that it considers such AI-written content to be spam, and recent research papers indicate that the company has already developed powerful systems to detect AI-generated content on blogs and websites.
This isn’t an issue for content writers who generate their own written content. But for the millions of publishers who hire external writers, it’s crucial to know whether content submitted by these writers was actually created by simple AI systems like ChatGPT.
Getting this wrong and publishing low-quality AI-generated content by mistake can easily land a publisher in hot water with Google and other search engines.
Given the importance of recognizing AI-generated content, how can one actually go about sniffing out this kind of machine-generated text?
In some cases, there are obvious indicators that a piece of content was generated by a computer and not a person. Repetitive looping text, overly perfect spelling and grammar, and other similar issues can indicate to an experienced human reader that a piece of text may be AI-generated.
That said, relying on manual detection is risky. Today’s systems are powerful enough that it could be nearly impossible for a person, even with extensive editing experience, to recognize AI-generated text on their own.
A better solution for detecting AI-generated text is to use an automated tool. Ironically, many of these tools are built using the same AI generation systems that write AI text! It turns out that catching an AI system often requires using an AI system.
Originality is a powerful new system for detecting AI-generated text. The system is trained on the outputs of the most modern text generation systems, including ChatGPT and GPT-3.
Using Originality is inexpensive and easy. Users can sign up for the service and paste text into Originality’s checker. The system will quickly provide a score, indicating the probability that the text was created by AI. Originality also includes a plagiarism checker, which is a nice feature for educators.
In my own testing, Originality performed well at detecting text treated by an AI system. It’s not a free option (though it does have a free trial), but the cost is low enough that it could easily be applied by a professor or even a consumer.
The company just launched a Chrome extension, which makes it easier for consumers to check news articles or blog posts for AI writing.
As mentioned previously, Content at Scale is an enterprise-level system for generating blog posts using AI and other advanced tools. In my testing, Content at Scale performed the best of any AI assistant system for generating compelling blog posts.
One of the advantages of Content at Scale is the fact that its output is high enough quality to avoid triggering most AI detectors. To prove that, Content at Scale provides its own free AI detection system on its website.
Even if you’re not a Content at Scale customer, you can still use the system to detect AI-generated text from other sources. Because this is a free option, it’s compelling and useful for many consumers.
If you need to generate AI articles that are high enough quality to provide real insights, you might consider using Content at Scale as a customer.
One of the most popular early AI-generated text detectors comes from Hugging Face, a community of AI programmers.
The tool was only trained on systems up to GPT-2, which was a precursor to today’s more powerful text generation models. That said, the detector is still reasonably good at detecting text generated by newer models.
Like the detector from Content at Scale, this detector is free to use.
Although today's detection systems are overall extremely good at detecting pure AI text, they often struggle to detect hybrid text. Hybrid text include elements that are generated by AI but have been edited or enhanced by a person, or by another natural language system.
For example, many writers may use a system like Jasper to generate a first draft of an essay, and then edit it to include their own insights, information in their own voice, personal experiences, and facts.
Because this kind of hybrid text includes much human intervention, it's often much harder for detection systems to see.
Of course, once a person has involved themselves in the generation of content, it's also debatable whether that content still qualifies as AI-generated.
If a human writer uses AI to generate a few paragraphs, which they then edit into a longer article, is that a problem or a process akin to using existing literary works for inspiration in a new novel?
It's an open debate, and many people probably come down strongly on one side or the other. Either way, it's an important distinction to be aware of. Even the most powerful AI detection systems have more trouble detecting hybrid text than pure AI text.
Today’s systems also sometimes yield false positives. Poorly written human-generated text often looks similar to AI-generated text, and thus detection systems might flag it as AI-generated when it was really written by a person.
The ease of use and ubiquity of AI text generation systems is challenging for many industries, including cyber security, education, and publishing.
Although many editors or educators may like to think that they can recognize text written by a machine, today's machine-generated text is convincing enough that it's often hard to recognize manually.
Thankfully, AI detection tools are advancing at the same rate as AI text generation tools, or perhaps even faster. Tools like Originality, Content at Scale, and Hugging Face provide an easy way to check whether text may be AI-generated.
To be clear, the systems aren't perfect and can still generate both false positives and false negatives. If you need to get a sense of whether text may be AI-generated, however, these tools are an excellent resource to use in your overall content evaluation process.
This article contains affiliate links. If you make a purchase, I may receive a commission at no cost to you. Thanks!
🤓 I need more Information about Ai Content Detectors
🤨 I have a question about a particular Ai Content Detector
🤔 I don’t know which Ai Content Detectors is right for me
🙄 Something else…
The New York Times referred to Thomas Smith as a “veteran programmer.” For over a decade, Smith has been leading Gado Images, an AI-driven company specializing in visual content. As a professional news, travel, and food photographer, and CEO of Gado Images, Smith’s work routinely appears in publications including Time Magazine, People Magazine, and Food + Wine. Gado Images, with its emphasis on AI technology, has contributed visual content to thousands of publications worldwide.
AI Secrets is a platform for tech decision-makers to learn about AI technology. Our team includes experts such as Amos Struck (20+ yrs ICT, Stock Photo, AI), Ivanna Attie (expert in digital comms, design, stock media), and more who share their views on AI.
Get AI news in your inbox & join thousands of engineers and managers using AI to boost sales and grow market share