In spring 2023, a new entry appeared on Google’s list of most searched terms.
‘ChatGPT’ amassed almost 120 million searches worldwide from people eager to test out the powers of the new generative AI language model sweeping the planet.
Fast forward to spring 2024 and we may see another new entry in response to this – ‘Best AI content detector’.
We’re now surrounded by AI-generated content.
AI writing tools like ChatGPT or Google Bard mean there’s a good chance that the product review, social media post, or email you’re reading has come directly from a robot.
Even a lot of business content, such as landing and product pages, may have come from a machine instead of a human’s keyboard.
This poses a problem. As human beings, we tend to seek a personal touch from the content we consume. At the very least, this helps build trust. At most, it gives a sense that the writer is being genuine.
We want to know what a human thinks about the global economy or whether Ryan Gosling really was funny in the Barbie movie.
We don’t crave just information, but the unique insights that only other people can provide.
Forbes research from earlier this year reflects this concern. Most of us are worried about the use of artificial intelligence behind a piece of content, be it advertising or a product review.
Worse, almost half of us (46%) say that we don’t know or are unsure of the difference between human or AI-written text.
Source: Forbes
Chatbots are training themselves to appear more human as each day passes, so this level of doubt is only going to grow.
The use of misinformation will also rise as bad actors use AI-generated text to achieve their aims. Society is soon going to need a widespread ability to detect false text so that “fake news” isn’t regularly spread unchecked.
But we’re going to need more than the naked eye to do this.
This is where AI content detectors step forward.
Built to identify generative AI content using machine learning and natural language processing (NLP), a detector tool uses algorithms to spot AI algorithms.
It takes one to know one, after all.
As a human being (we presume), you’ll probably want to know if what you’re reading was written by another human.
Knowing which is the best generative AI content detector tool will help you do that.
Looking for SEO-focused content marketing that gives your company a human voice while generating organic leads? Book a consultation call with us to find out how Mint Position uses the perfect blend of humans and AI-based SEO tools to get you ranking higher.
First, do AI content detectors even work?
Make no mistake – inaccurate AI detection can be dangerous.
Imagine putting long hours of hard labor into an original essay only for a smart-alec AI detector to tell your editor/teacher/client that it’s probably AI-generated.
This is called a false positive and it threatens the integrity of honest content authors everywhere.
Unfortunately, the perfect AI content detector doesn’t exist (yet), but the best examples can give you an excellent idea of whether you’re dealing with a person or a robot.
A recent study from the University of Kansas found that an in-house detection tool identified AI content with a 99% accuracy rate, but this was geared toward academic material and white papers.
Detector tools designed for more general content have lower success rates, although some claim up to 99% predictability with content generated by GPT-4, OpenAI’s latest AI model.
Accuracy is the most important factor, but we should also bear usability and pricing in mind when choosing the best generative AI content detector.
So, how do we know which is the best English-language AI detection tool out there?
We have compiled the following methodology to help decide.
What is the best AI content detector? We test 9 tools with our in-house methodology
If you wish to find the best generative AI content detector then you’re going to need a lot of time on your hands.
To help you out with this, we decided to put nine AI content detection tools to the test by analyzing the accuracy of each with nine pieces of content, as follows:
- 3 pieces of 100% human-generated content
- 3 pieces of 100% AI content from OpenAI’s GPT 3.5 model (the most commonly used)
- 3 pieces of hybrid content (50% human and 50% AI).
We deemed the ‘three points for a win’ scoring system to be the fairest way of analyzing these as it has a tiered grading system that rewards pinpoint accuracy.
If the tool is completely accurate, or its score is within 10% of the right answer, then it gets three points.
If its accuracy is within 11% and 40% of the correct answer, it gets one point.
If it’s more than 40% outside of the right answer, it gets zero points.
Here are three examples to make this system crystal clear.
- I enter a text that’s 50% AI generated and the tool grades it as 60% AI – 3 points
- I enter a text that’s 100% AI and the tool grades it as 70% AI – 1 point
- I enter a text that’s 100% human but the tool flags it as 80% AI – 0 points (false positive)
These results will yield an accuracy percentage from a maximum score of 27, which I’ll combine with other factors, such as ease-of-use and cost to come up with a star rating out of five.
Here’s how they performed.
NOTE: Made using Mint Position’s in-house scoring system.
Now let’s look at each detector, one by one, to see how they stack up.
Originality.ai – The Best AI content detector tool
✅ One of the most accurate tools around
✅ Excellent design
✅ Clear and colorful grading system
❌ Have to request a free trial which only includes a few scans
Originality is one of the most talked-about detectors right now. It appears on many ‘Best AI content detector’ lists thanks to its high accuracy, which has been backed up by well-publicized in-house tests
Founder Jonathan Gillham has given a series of interviews talking up his detector’s accuracy. Central to these is a test of over 10,000 pieces of text that he claims yielded an accuracy rate of over 94% and reached close to 100% much of the time.
The workings of Originality’s methods aren’t clear. Not even Gillham himself knows how the tool works but instead compares it to existing technology to reassure the public.
“Half the trading decisions on stock markets are made by machines without any direct human oversight,” he says, “I know how the tests work and I’m very confident in (Originality’s) accuracy.”
But does Originality back up its founder’s positive talk according to our tests?
Well, our results show that Originality is the most accurate tool we came across. It achieved its score of 85% by correctly detecting all human and AI texts, only tripping up slightly on the hybrid texts.
Originality uses a bright, color-based confidence system with green for human-sounding copy through to red for mechanical content. It even lets you see a history of previous scans.
The one downside is that Originality is mostly a paid-for tool, although we found we could get a small trial by request. Prices start at $14.95 a month for 2,000 credits (or around 200,000 words) but you can make a one-off payment of $30 for 3,000 credits.
If you have a budget for an AI detector that’s right most of the time, then Originality.ai is the one to go for.
Accuracy percentage: 93%
Overall rating: 4.5 out of 5
SEO.ai – The best free AI content detector
✅ A superbly accurate and free tool
✅ Informative content feedback
✅ Clear and colorful grading system
❌ Struggles with hybrid text
With two decades’ experience in the SEO industry, SEO.ai helps businesses produce high-ranking human content.
The company recently released its free AI content detector which “uses four unique AI content detection models at once” to sniff out robot-written copy. We can summarize these as follows:
- Classifiers: Recognizes the language patterns of artificial text.
- Embeddings: Analyzes how much similar-sounding text an extract has: AI tends to use more than humans.
- Perplexity: Measures the “surprise” element of the text, or how creative it is. AI texts strive for low perplexity so that the text is more smooth and understandable.
- Burstiness: Look at the variation of words in a sentence. AI uses less of this.
SEO.ai explains the science behind its detector well, but can it sniff out artificial text?
Our tests showed that it mostly can, with its 88% accuracy the best we found from a free tool.
It flagged the artificial texts with ease and achieved high scores for the human content, too. The hybrid copy is where it struggled the most, choosing to label it as mostly human or artificial instead of the 50% split.
The tool also provides a clear and informative analysis of the text, with a breakdown of the percentage scores of each of its four models. This means we can see not only if our content sounds robotic, but why it does.
SEO.ai’s informative analysis
If you’re looking for a free detector, then you won’t do much better than SEO.ai’s offering.
Clear, informative, and simple to use, it beats most paid-for tools in terms of accuracy and analysis.
Accuracy percentage: 88%
Overall rating: 4.5 out of 5
Sapling AI Detector – a solid free AI detector
✅ It’s free – with no apparent limit or sign-up
✅ Has a clear and precise scoring system
✅ No false positives
✅ Has an informative FAQ section
❌ Could do better with detecting AI text
❌ Tries to push an unnecessary browser extension
Sapling.ai is targeted at businesses that wish to root out AI-sounding text from their communications, but the general public are free to use it, too.
Human-generated text fared the best with this model. It returned a 0% chance of AI for all text we tested, which is as far from a false positive as you can get.
Sapling’s scoring system is also easy to understand, using a simple percentage and striking colors to get its point across.
This tool lets itself down slightly when it comes to AI text, though. Like with ZeroGPT, it tends to be more accurate the more text you paste in, but it still only graded the longest entry as 72.5% chance of being artificial. The shortest entry received just 60%, which leaves a lot of room for doubt.
Sapling’s accurate human text reading mitigates concern, however: if you get any kind of AI percentage, then you can assume that it’s artificial by default.
Sapling is one of the best free tools we came across: it can tell the difference between human and AI content, it’s free, and it’s easy to use.
Accuracy percentage: 74%
Overall rating: 4 out of 5
Content at Scale – Lots of free scans and a fairly high accuracy rate
✅ A free tool allowing 25,000 characters (around 3,000 words) each scan
✅ A high accuracy rate with no false positives
✅ Grades AI-sounding text using a traffic light system
❌ Pricing for a full version isn’t immediately available
Content at Scale is one of those companies that the media told content creators they should be worried about: a content automation platform that “was built to scale content marketing by replacing the need to hire a team of writers”.
While this claim may be wide of the mark – top-quality content will come from striking a balance between AI and human creators – its detection tool claims to be one of the best generative AI content detectors around.
Our tests showed that it does achieve a very good level of accuracy. It sailed through the human-text checks by correctly identifying all entries as 100% authentic content.
It also correctly flagged AI-written content, albeit with a less-than-perfect 30% probability of human creation, and its hybrid detectability was also fine.
The tool is user-friendly and employs a clear traffic light system when grading AI, with red flagging the most obvious AI-produced content and green showing up for human-written text.
Content at Scale’s free version will be enough for most users, but if you’d like to upgrade to Pro standard then the pricing isn’t immediately available – instead, you’ll need to “request an invite”, which is a hassle and means we can’t award this tool a full 10 out of 10 rating.
Accuracy percentage: 74%
Overall rating: 3.5 out of 5
Winston.ai – Great for detecting AI content, but not human text
✅ No credit card required for the free version
✅ Great design
✅ Excellent at detecting AI text
❌ Struggled with detecting human text
❌ You only get a few scans before having to upgrade to a paid version
Winston has a background in plagiarism detection and you get both its plagiarism tool and AI detector when you sign up to a paid version.
You can access a free trial without signing up or sharing card details, but this extends to only a few scans, which can be in English or French.
From our tests, Winston fared perfectly when it came to detecting AI text, making it a useful tool for teachers everywhere, in particular. However, its hybrid and human text detection left a lot to be desired, only achieving a 100% accuracy score in one of the six entries in these categories.
Like several AI detectors, Winston uses a striking coding system to clearly illustrate what it thinks is AI or not. You can also get started in just a few clicks instead of searching around a website.
However, its poor non-human detection means there’s a lot for it to improve on.
Accuracy percentage: 70%
Overall rating: 3 out of 5
Copyleaks – Easy to navigate but inaccurate detection
✅ Free daily quota with no sign-up required
✅ Highlights AI-sounding text
✅ No false positives
❌ Identified AI content as human
❌ Limited daily quota even if you sign up
Copyleaks has been available as a plagiarism checker since 2015 which, it claims, makes it an expert in detecting AI content.
At first glance, it appears impressive. It works with 15 languages (including French, German, and Spanish), compared to the rest on this list which mostly work with only English, and claims to offer five times more daily scans than your average detector, and has a 99.1% accuracy.
For our tests, the human-text checks went well. Copyleaks correctly identified it as AI-free with a clear thumbs up. It also scored well with the short AI word count.
The long AI text and hybrid texts, though, are where it fell down. Here, its results veered wildly with all turning out to be inaccurate.
Copyleads is easy to navigate, uses full API integration, and offers affordable pricing plans on top of its free daily quota. But if it’s infallible detection that you’re looking for, then there are more accurate tools out there.
Accuracy rating: 59%
Overall rating: 2.5 out of 5
Kazan SEO – A messy design with low accuracy
✅ Unlimited free checks (we think)
✅ Very quick response
✅ No false positives
❌ Identified AI content as human
❌ A messy design with adverts
Should you Google ‘Best AI content detector’, then you’ll see the name ‘Kazan’ pop up regularly.
There isn’t much info about Kazan online. It just has a webpage where you can try out a host of its tools from content optimization to text extraction.
Its AI detection tool appears on a clumsily designed page littered with adverts, but it quickly identified all our human entries as authentic.
However, it appears incapable of detecting AI content, ranking all of our AI and hybrid entries as close to 100% human.
An unlimited number of free checks doesn’t mean much if it doesn’t flag generative text.
Kazan’s strange design thrown in for good measure means you should probably avoid this tool if you’re looking for accurate detection.
Accuracy percentage: 33%
Overall rating: 2 out of 5
Writer.com AI Content Detector – A pretty but ineffective tool
✅ It’s free, with unlimited 1,500-character (300 words) checks
✅ Boasts a beautiful design with a clear scoring system
❌ Didn’t pick out any AI across several clearly artificial examples – what’s the point?
❌ 1,500 characters is frustratingly short
Now, we don’t wish to be harsh to the makers of any of these tools. Somebody somewhere has put a lot of time and effort into creating them. But in the case of Writer.com’s AI Content Detector, you have to wonder what purpose an AI detector that doesn’t detect AI content serves.
Writer is one of the best-marketed tools out there, appearing proudly at the top of Google’s search engine results. After giving it a few tests, we were encouraged by the “100% human-generated content” scores each time. “Great!”, we thought, “our writing sounds authentic!”.
Well, it turns out that it doesn’t mean that much. “100% human” appears to be Writer’s default response, even after we desperately tried to make it say something else.
We entered our hybrid text — “100% human” was the response.
We put the most obvious robot-made text in there, full of clunky phrases like ‘navigating the evolving tech landscape’ — it came back human.
We even dropped ChatGPT’s tell-tale ‘regenerate’ in there. This button gives the user an option to produce an alternative answer, but it’s very easy to accidentally copy and paste it from the editor along with the text (see our text in the box below). The result when we did this? Still 100% human!”
Writer’s 100% ‘human content’ score — with ChatGPT’s ‘regenerate’ included
It also recognized our genuine text as human, of course, but to the Writer detector, everything’s real.
To summarize, if you’re looking to pick out AI content generation, then this tool won’t help you much.
It’s very nice to look at, though, if you like that kind of thing.
Accuracy percentage: 33%
Overall rating: 2 out of 5
GPTZero – A free tool that fails to hit the mark
✅ It’s free – limited to 5,000 characters (or around 1,000 words per check)
✅ It highlights sentences it believes to be AI-written text
❌ Poor accuracy
❌ Offers vague explanations
❌ Gives false positives
GPTZero appears to be one of the most generous detector tools out there.
It’s completely free and allows you to input around 1,000 words each time, although you’ll need to sign up after your first few tests.
Its grading criteria also sounds impressive, like with SEO.ai above, it uses words like ‘burstiness’ and ‘perplexity’ to describe how the detector works.
Yet, despite all that, GPTZero lets itself down for poor accuracy. For each of our tests, it failed to accurately guess the source of the text and generated a false positive more than once.
It appears that the shorter the text, the less accurate the tool is: for our 300-word human-written content, it generated a 50% chance of human input; but for 1,000 human words this reduced to 30%. Seeing as many users will paste snippets of text into the editor, this is a poor level of precision.
In terms of AI text, GPTZero again only picked up a 50% probability: way below what we should expect, and it fared only slightly better on hybrid texts.
While some of the best things in life are free, GPTZero isn’t one of them.
Accuracy percentage: 11%
Overall rating: 1 out of 5
So, which is the best generative AI content detector?
Finding the best generative AI content detector comes down to whether you’re prepared to pay money for it or not.
If you are, then we found Originality.ai’s algorithms to achieve incredibly high accuracy scores – it’s the winning option for the general internet user. Quick and a breeze to use, $15 a month isn’t going to break the bank, and its 200,000-word limit will take you a long way.
If you’re a more casual user, though, Seo.ai’s free tool is the best choice. You get highly accurate unlimited checks for free, all backed up by well-explained methods.
Looking for high-quality content marketing that uses journalistic-quality research like that found in this article? Get in touch with us and we’ll show you how quality human content backed up by AI-powered SEO ranking tools can help you increase organic traffic and boost conversions.
Disclaimer: The author’s comparison of AI content detectors is based on personal analysis independent of Mint Position LLC. The article does not imply endorsement by the mentioned organizations. Readers should verify information independently due to evolving technology.
