Since its launch in November 2022, no app has had a larger impact on the world than ChatGPT.
It has almost single-handedly brought about the AI revolution that is working its way through nearly every industry on Earth.
At the top of that list is the world of content creation and publishing.
For thousands of years, humans have been putting pen to paper or fingers to a keyboard and manually pushing words to print.
But with the introduction of ChatGPT, all of that changed forever.
With just a few simple prompts, people can use ChatGPT to produce articles within a matter of seconds.
But with over 100 million users worldwide, people started to ask the question of how to detect ChatGPT when reading content online.
So can ChatGPT actually be detected? Absolutely.
This article will discuss the best tools and tips for detecting ChatGPT and how you can use them to your advantage.
What is ChatGPT?
It was originally introduced in 2022 by the company OpenAI as a large-language model or LLM to assist us in completing a wide variety of tasks.
LLMs are often described as chatbots since users can casually converse with ChatGPT to obtain their desired output.
The GPT stands for Generative Pre-trained Transformer which has been fine-tuned to provide information and responses in a casual conversational tone.
The current model of ChatGPT is GPT4 which was released in March 2023.
ChatGPT has continued to evolve into more than just an LLM.
Today, the system can identify images and use text-to-speech to provide human-like audio.
It is also integrated into thousands of different applications including Microsoft’s suite of apps like Teams and the Bing web browser.
Microsoft owns a 10% stake in OpenAI and has hired the former founder and CEO Sam Altman to head its own AI taskforce.
Can ChatGPT Actually Be Detected?
The answer is yes but also no, and we’ll explain.
If you want to be literal, ChatGPT cannot be detected with 100% certainty by any human or AI detector tool.
Even if you use an AI detector to analyze a particular article or passage, it will only tell you the likelihood that it was created by AI.
It cannot specifically tell you that it was ChatGPT that produced it or another LLM like Claude AI or Google Bard.
What we can identify in AI-generated writing are certain predictable patterns.
These patterns will appear concerning things like the syntax and sentence structure of the text.
Other measurements are used like the burstiness of the text which essentially means the variation of sentence length and the perplexity which is how complicated the text is.
Who Would Need a ChatGPT Detector?
Anyone who deals with published material or content in any way should use a ChatGPT detector tool.
These tools have been particularly useful for teachers to determine academic dishonesty and for editors who receive plagiarized, AI-generated work.
But AI detectors can assist nearly anyone who reads articles or blogs online.
Part of the issue with LLMs like ChatGPT is that they can provide plagiarized or factually inaccurate outputs.
This is known as AI hallucinations and unless you research and check the facts, you could unknowingly publish undeniably false information.
How Do I Know if Something is Written by ChatGPT?
While it is difficult to determine if something is written by ChatGPT, you don’t need an AI detector tool to recognize some of the patterns.
Here are 8 things that might tip you off that what you are reading was produced by ChatGPT.
1. Look for Patterns and Inconsistencies
As we mentioned, there are plenty of recognizable patterns that appear in AI-generated text.
This doesn’t just need to be patterns in the syntax or structure of the text, but also in how predictive the text itself can be.
AI is not capable of thinking outside of the box and would rather provide a safe and consistent output that is easy to read and understand.
Oversimplicity and predictable language are hallmarks of text that was created by AI.
2. Check for Signs of Human Error
You might not realize it, but most human-written text has some sort of errors in it.
This is completely normal and has been the norm for centuries now.
Human writers with human editors can certainly miss something like incorrectly placed punctuation or grammatical or tense issues.
But compared to AI text, human writing is more polished and natural.
AI tools like ChatGPT aren’t writing the text but rather producing it from all of the data that they have been trained on.
This can include errors in grammar or word choice that will sound awkward to us but indiscernible to the AI tool itself.
3. Look for a Lack of Descriptive Language
This is exactly what the perplexity score is for AI-generated content.
Most LLMs are designed and programmed to not confuse the user. For this reason, these tools use language that is simpler and easier to understand.
There is also a lack of diversity when it comes to word choice and you will realize this as the same words or phrases will often appear multiple times within the same text.
4. Pay Attention to the Context
Why would you need to pay attention to the context? Because AI tools are often quite bad at providing it.
Remember that AI is essentially producing the output and responses from previous data sets.
It isn’t necessarily easy for ChatGPT to continuously understand new contexts for each conversation.
ChatGPT is getting smarter and can both learn and be trained in understanding the context of previous conversations.
But when it comes to the text output of the conversation, ChatGPT and other AI tools are still quite poor at providing context that is otherwise very easy for human writers to include.
5. Excessive Use of Transitional Words
Transitional words are excellent for tying together sentences and ideas.
This structure comes naturally to human writers but for AI tools, they seem to overcompensate with transitional words.
These LLMs are trained to provide similar sentence structures and lengths.
To combat this, it seems like they also have an overactive algorithm that tries to tie together sentences or concepts in an unnatural way.
If it sounds strange when you read it, AI was likely involved in the production of that text.
6. Sentences That Look Right but Don’t Make Sense
If you ever see these sentences that look right but are slightly off when it comes to trying to make sense of them, then it was likely written by AI.
These tools are well-trained when it comes to things like sentence structure but can often miss the mark when it comes to the subject of that sentence.
Always remember that ChatGPT and other AI tools are just regurgitating data they’ve learned somewhere else and this can result in some awkward phrases and sentences.
7. Lack of Originality
When you read most AI text, there just isn’t any originality to it.
It is missing the nuance and soul of a human writer. There is a reason why Google’s EEAT guidelines are making it difficult for AI-generated content to rank highly.
Google is looking for examples of Experience, Expertise, Authoritativeness, and Trustworthiness from the content.
These are all things that AI has difficulty in providing, particularly in an original and thought-provoking way.
There have even been examples of AI text that is completely plagiarised from another source.
AI tools are incredibly helpful but lack originality compared to humans.
8. Factual Errors
We already briefly touched on AI hallucinations and unfortunately, these are still the cost of doing business with an AI writing tool.
If ChatGPT provides a factually inaccurate output it is up to the user to check for authenticity and accuracy.
ChatGPT itself can provide factual errors without blinking an eye.
If you notice that certain things are blatantly incorrect in a text, then there is a good chance that it was either an AI tool or a horrendously misinformed human author.
How to Spot Misinformation Generated by ChatGPT?
Unfortunately, misinformation comes with the territory when using AI tools to produce content.
But how can you spot misinformation in the text that was generated by ChatGPT?
It’s difficult and you do have to do a bit of sleuthing but eventually, you’ll be able to recognize the signs of misinformation.
Start with any awkward sentences or repeated facts or phrases that stick out in the text.
Here’s your first clue: ChatGPT and other LLMs will often have punctuation or grammatical errors when they are hallucinating.
Another way to spot misinformation is the human common sense test: if something you read sounds fishy, fact-check it.
When an LLM is hallucinating, it will provide its response in the same casual and conversational output.
There is no way for it to know that its output is factually incorrect!
What Tools Are Used to Detect ChatGPT?
If you need a tool that can detect ChatGPT writing, you’re in luck.
The AI detector industry has exploded with dozens of different apps that all claim to be able to identify AI-generated text.
Do they work? To varying degrees. Here are our choices for the best tools to detect content that was created by ChatGPT.
SEO.ai
Apart from having one of the coolest URLs in the AI industry, SEO.AI also provides a long list of free and paid AI tools for content creators.
The site’s main offering is a platform that helps to generate SEO-friendly content that has a high SERP rank and is Google EEAT optimized.
But SEO.ai also offers a free AI detection tool. This tool can detect the latest GPT-3.5 and GPT-4 versions of ChatGPT and other LLMs like Google Bard.
SEO.ai claims that the detector can flag AI-generated content with a 98.4% accuracy.
Copyleaks
Copyleaks is a well-known AI tool that can help identify plagiarized or AI-generated content.
A July 2023 report from a team at Cornell University shows that Copyleaks was the most accurate AI detector tool on the market with a 99.1% accuracy.
This tool comes with a free version as well as a premium paid version that offers plenty of additional features like detection of up to 30 different languages and unlimited scans.
It even comes with a Google Chrome extension so that you can use Copyleaks with any site you have open in a Chrome window.
GPTZero
GPTZero, as its name suggests, is an AI tool that can be used to detect if the content was written by ChatGPT or any other LLM.
It claims that it is the gold standard in AI detection and the most trusted AI detector tool on the market.
GPTZero has partnered with the American Federation of Teachers to ensure academic honesty in millions of classrooms across the country.
This app also provides both a free and premium version.
The free version has a 5,000-character limit for scans and a 10,000-word per-month limit in total.
A price of $10/month will get you up to 150,000 words and $16/month will get you up to 300,000 words per month.
GPTZero also offers batch file and document scanning, plagiarism scanning, and an API and Google Chrome extension.
This app claims it has a 99% accuracy when it comes to detecting human-written content and an 85% accuracy in identifying AI-generated content.
GPT-2 Output Detector
The GPT-2 output detector is an open-source tool that can detect AI-written text vs. human-written text.
This tool was created by using a fine-tuned version of the RoBERTa model and utilized the 1.5B-parameter GPT-2 model.
This AI detector has flaunted an accuracy rate of 99.8%, although it is potentially limited by not including the newer GPT-3.5 and GPT-4 models.
PoemOfQuotes
PoemOfQuotes might not be the most obvious AI detector tool but it certainly works.
This site is used as an AI poem-generation page that has been trained on over 178 billion parameters.
The AI detector tool is provided by the site AI Detector Pro and can integrate directly with Microsoft Word and Google Docs.
This app will even provide you with reports outlining what parts of your text were likely written by AI and which were written by a human.
There is a premium paid version which will run you $13.99/month for 100 reports or $24.99/month for unlimited reports.
Corrector
The Corrector AI detection tool is completely free to use although it does have a word limit of 800 words per scan.
The site claims that its AI detector is 100% accurate and unlike other detectors, it can detect content created by every iteration of ChatGPT including GPT-4.
Corrector’s AI detection tool can identify content created by AI tools like ChatGPT, Google Bard, Jasper AI, SurferSEO, and Quillbot.
Content at Scale
Content at Scale is a full-solution AI content tool that offers an AI detector as one of its options.
This AI detector has an alleged 98.3% accuracy rate and has recently upgraded its modelling to match that of GPT-4, Google Bard, and Claude AI.
Content at Scale boasts that it has the simplest scoring system that alerts users if the content was created by a human or an AI writing tool.
Following each scan, Content at Scale will also provide sentence-by-sentence analysis, flagging the sections that appear to be AI-generated.
Included in the cost of the AI detector is Content at Scale’s generative AI writing tool that will not only help you produce high-quality content but also help to re-write old content and AI-generated content in a more human-sounding manner.
For unlimited AI detector scans and up to 25,000 undetectable AI words, Content at Scale charges just $49/month.
Roberta OpenAI Detector – Huggingface
Huggingface is a French-American online AI community that allows developers to collaborate and share machine learning applications.
Not surprisingly, Hugginface has produced quite a few AI tools that have become popular not just in the developer community but also in the mainstream content industry.
The Roberta OpenAI detector was already featured earlier in this article as the platform for the GPT-2 Output Detector.
Roberta, or roBERTa as it is known, is an AI model that was based on the Google BERT model from 2018.
Roberta was trained on a much larger dataset than BERT and uses a slightly different program architecture.
ChatGPT Detector – Huggingface
(This seems like the same as the other Huggingface AI detectors?) The webpage that I found was https://huggingface.co/spaces/Hello-SimpleAI/chatgpt-detector-single which seems odd since there are spelling mistakes and half of it is in Chinese.
GLTR
GLTR or the Giant Language Model Test Room is an AI detector that provides a visual overlay of the text that colour codes which words or phrases were created by AI.
GLTR.io uses the GPT-2 117M platform as its foundation for analyzing and identifying AI-generated text.
Rather than a paid subscription, GLTR is an open-source code that is available for access by developers on GitHub.
The free demo is also available to use on its website for anyone who wants to test it out.
The tool’s colour-coded scheme is great for identifying different language patterns, some of which are obviously used more frequently by AI writing tools.
The one drawback is that it is trained on the now outdated GPT-2 model which means it might not be as effective in flagging AI-generated content from the more recent GPT versions.
Writer
The Writer AI detector tool is free for use at its website and is fully trained in GPT-4 and ChatGPT.
Currently, the limitation for scanning content on Writer is 1500 characters although there are no limits for the number of times you scan content.
It is an extremely simple tool to use and provides an immediate detection score once you scan your text.
Writer comes with other tools including a generative AI content tool and an AI-powered plagiarism checker.
These other products allow for the integration of extensions directly into other applications including Figma and Google Chrome.
Here is an example of this paragraph scanned by Writer for AI detection.
Does ChatGPT Have a Watermark?
As of now, ChatGPT does not currently watermark its text or images. This has been a controversial topic ever since ChatGPT was introduced back in 2022.
With recent pressure from US regulators, the White House has reported that several companies including OpenAI and Meta have agreed to work on adding watermarks to their AI-generated content.
What exactly is an AI watermark? It is a form of embedding digital indicators into the text itself.
For the most part, these signatures will not be detectable by the human eye, but AI detector tools and scanners will be able to recognize the sequence as unique to ChatGPT.
It is believed that these watermarks will allow people to easily identify when text, videos, or images have been created by AI rather than a humans.
Is Google Interested in Detecting Whether ChatGPT Writes a Text?
This has been another interesting debate in the content industry.
As you know, the key to monetizing a website or blog is to rank highly on SERPs through keyword and SEO optimization.
Google sees more than 90% of the world’s internet traffic so if there is one set of guidelines you’ll want to follow, it’s the Google SEO guidelines.
After ChatGPT was released Google revised its guidelines to include the principles of EEAT.
This stands for Experience Expertise Authoritativeness and Trustworthiness.
Google does not explicitly care if you use AI to produce content. It specifically mentions AI-generated content as not being against its guidelines as long as it is not deemed to be spam.
The problem with AI-generated content is that it will never provide a human’s perspective or provide first-hand experience.
According to Google’s EEAT guidelines, this means that strictly AI-produced content will usually not rank very highly.
This requires you to go in and manually edit the content yourself or you could use a tool like the one we’ll discuss in the next section.
How to make ChatGPT Text Undetectable?
So if AI-produced content does not rank highly, how can we make it so that AI detectors and even Google itself cannot detect that you used ChatGPT to produce your content? By making it Undetectable.
Undetectable.AI is the leading AI scrambling tool that takes your human or AI-generated content and humanizes it with the click of a button.
This tool guarantees that your humanized content will not be flagged as AI-generated by some of the best AI detectors on the market. These include ZeroGPT, Writer, Copyleaks, and Sapling.
Whether you are using ChatGPT to produce your content or writing it yourself, Undetectable.AI is a must-have for any content creator who is using AI to scale their content output.
For as low as $9.99/month for up to 10,000 words or just $5.00/month when you pay annually, Undetectable.AI is a minimal cost for every writer’s peace of mind.
Conclusion
As AI tools like ChatGPT continue to evolve and improve, there will always be a demand for ways to detect its presence.
There is a long list of AI detector tools that are on the market these days. Some are free, some are open source and some require a paid subscription.
But all of them are working to cut down on writers who are using AI as an easy way to mass produce low-quality content.
While they can be a nuisance to writers, especially if your content is flagged as AI by Google, there are ways around it.
First, you can take the old-fashioned route and either write the content yourself or manually edit your AI-generated text.
The other method is to use a tool like Undetectable.AI to complete that task for you.
With just the click of a button, this tool can humanize your content and protect your content from being identified as produced by AI.
Ready to see the difference? Test the Undetectable AI widget below and experience firsthand how it can enhance your writing (English only).
Just input your text and watch as it transforms into a more humanized and polished version. Give it a try now!