How to Tell if Something Was Written by Chat GPT

+8 Free Chat Detector Tools 

Have you recently looked at a phishing email and questioned its legitimacy because it was decently written? Or read a news article that didn’t seem quite accurate, but you weren’t sure what was wrong with it?

It was likely written with the help of AI. Artificial intelligence is a hot-button issue, posing ethical and political questions to society, as well as concerns around personal privacy and false information for individuals.

Although AI has been with us for years (think smartphones and search engines), ChatGPT is getting a great deal of attention, as it pulled back the curtain on AI-generated writing for the general public, allowing us to ask questions and see what it can generate in real-time. So, how does this all work, and how can someone tell if something is written by AI?

Summary; TLDR 

AI-generated content is under the spotlight in the form of ChatGPT, an open experiment in language processing led by OpenAI.

AI-generated writing is becoming more common across the internet and improving in its quality as the technology learns through more utilization. Online tools are available to detect text written by artificial intelligence, but their accuracy is not always spot-on. And questions are raised for consumers, students, educators, and writers about its practical application, ethics, and potential threats it poses to our current modes of operation.

What is Chat GPT?

Who better to answer this than ChatGPT itself?

ChatGPT is a language model developed by OpenAI that uses artificial intelligence to generate human-like text responses.

It is designed for natural language processing tasks, such as answering questions, generating text, and engaging in conversation, and can be used in various applications like chatbots and virtual assistants.

ChatGPT is trained on a vast dataset and can understand and produce text in multiple languages, but its responses are based on patterns in the data and may not always be entirely accurate or contextually appropriate.”

How Does AI Work?

Artificial intelligence (AI) works by showing large amounts of data into a computing system repeatedly until it can distinguish differences in the patterns it has seen, therefore programming itself through that data. The algorithms that are programmed into the system help it to understand the question it is being asked to answer.

As a simple example, a data scientist can feed thousands of pictures of dogs into a system, labeling them as “dog” or “not dog” until it learns the general characteristics of dogs well enough to distinguish the difference itself.

ChatGPT and other AI content generators use large amounts of data by combing through human-written text, gathering knowledge, and producing human-like text predictions to answer questions and write paragraphs. The more of this it does, the better the algorithm gets at producing text content that humans like to read — or images that humans like to see.

How to Detect AI-Generated Text Yourself

Are you afraid of being fooled by ChatGPT or other AI-written text? Here are some ways to detect AI-generated text with the naked eye. Keep in mind that when AI makes mistakes, they seem strange–not the kind of mistake that a human would make.

ChatGPT gave these titles when asked for books on Chester Alan Arthur.
All titles exist and were findable on Google except the last one.
  • Patterns: Most AI uses predicted word writing, making its next word the most common one it has found, which may also result in short sentences with lots of repetition. Text written by a human will contain more variety, more synonyms, and longer sentences with a more appealing structure.
  • Factual Inconsistencies: ChatGPT can make some interesting claims with statistics, like stating that ADHD and Asperger’s co-occur as diagnoses in 20-70% of children, a range so large that it appears dubious. It can also get basic facts wrong– like when asked which institution granted a professor his Ph.D., it produced a confident answer of Oxford University, which was incorrect and easily verifiable. It is always a good idea to double-check ChatGPT’s facts, especially around statistics, dates, and other readily verifiable information.
  • Obvious Misinformation: When asked to provide a list of books about President Chester A. Arthur, ChatGPT generated a list of six books: four were actual published books, one was listed twice (once with its subtitle, once without), and one was completely made up, even though its title made it sound plausible enough. A quick run through a search engine revealed the real from the fake.
  • Irrelevant Text: Occasionally, ChatGPT text will occasionally include information that is unrelated to the topic at hand or what has already been written, or simply does not make sense in context. This is a dead giveaway that an AI tool was used in writing.
  • Signs of Human Error: AI is only as good as its data set, and the humans providing the data are full of bias, which means the product produced by ChatGPT can also be biased. Common sense is your friend in detecting evidence of human error in ChatGPT.

|| Related: 15 Instant Blog Ideas for the Uninspired, Unmotivated, or Reluctant B2B Blog Writer.

8 Free Tools to Detect AI Written Content

Chat GPT-2 Output Detector predicted the AI-written text I asked it to examine was “fake” or robot-written.

Have you recently looked at a phishing email and questioned its legitimacy, or read a news article that didn’t seem quite right? AI is becoming more common, and detecting AI-generated text may be more difficult over time. Let’s look at a few of the AI detection tools available

AI is a master imitator, using large language models to figure out relationships between words and phrases, predicting what it thinks people would like to read.

With a 77% degree of probability, GPTZero also predicted the AI-written text to be non-human.

As the AI tools are accessed by more people, it learns from its users and the AI writing improves, making it harder to detect AI-generated text. AI detection tools work to find the patterns in AI writing, therefore predicting which texts are AI-written instead of human-written. Here are a few to try out.

  1. GPT-2 Output Detector: With a simple copy-and-paste of text into the box, this basic ChatGPT detection tool will give a percentage estimate as to whether a passage is AI-generated text. Its prediction gets stronger with more text entered.
  2. GPTZero: By uploading a file or copy-and-paste, GPTZero can detect text ChatGPT written text, Google’s Bard AI generator, human-written text, or a combination of human and AI content. This free tool does want you to register an account with your email.
  3. DetectGPT: This is another basic AI detection tool, similar to the GPT-2 Output Detector. It is based on an older language model.
  4. Writer AI Content Detector: This is a great tool for catching writers who are trying to mask GPT-written text. Bonus: In addition to copy-and-pasting text, you can throw a URL into this content detector and it will score your work (or someone else’s) as a percentage likely to be written by a human.
  5. Content at Scale AI Content Detection: This is a user-friendly AI detector with bells and whistles, having tools to work with both text and images. This company claims it can also help advertising agencies and content writers become more efficient by harnessing the power of AI–for a fee, of course.
  6. Undetectable.AI: Undetectable.AI takes AI-generated text, analyzes it for the likelihood of AI content, and then can humanize the writing to different levels and audiences so that it beats AI detectors. It costs around$9.99/month for 10,000 words, makes the user agree to terms of service (including no academic misconduct), and watermarks drafts.
  7. GLTR: Through its color-coding of text based on probabilities of text, the GLTR AI text classifier “enables forensic inspection of the visual footprint of a language model on input text to detect whether a text could be real or fake.” This visual representation of human writing vs. AI content is striking.
  8. OpenAI Classifier: OpenAI, maker of ChatGPT, has shut down its own data classifying tool in July of 2023 “due to its low rate of accuracy.”

How Do These Chat Detection Tools Work?

An AI content detector works by reverse engineering the process that AI uses to generate its text in the first place.

ChatGPT predicts one word at a time. Based on the question or prompt entered, it produces text based on the word most likely to be used based on human speech patterns. These tools look for those statistically likely words or phrases and use those occurrences to predict whether or not content was produced by an AI tool.

Fed by the AI-Language Models They’re Spying On

Chat GPT detectors can detect if text was written by AI and which AI tool was used.

As with the AI content generators, the content detectors are only as good as the AI language models it has learned. Some of the available tools can detect ChatGPT-2, but ChatGPT-3.5 (OpenAI’s currently available free tool) is based on a newer, larger data set.

Other AI generators, like Bard by Google which uses its own AI language model with an emphasis on conversation, mean that the AI detectors need to be trained in those models to accurately predict AI-generated content.

Designed to Go Undercover

QuillBot will paraphrase AI-written text you provide.

While some tools specialize only in detecting AI-generated text, others are capitalizing on the movement toward AI. Companies are targeting students, academics, journalists, and content and copywriters, promising to teach them how to harness the power of AI chatbots and use it as a tool to improve their own writing processes.

Some, like Undetectible.AI and QuillBot, help rephrase AI-generated content into unique words that are unlikely to be red-flagged for plagiarism or AI authorship. Others like Content At Scale offer long-form, AI-generated text as the basis for content writers in addition to scanning articles for AI detection. Most of these are paid services, with costs based on scale, be it word count, number of users, or number of articles written.

You Can’t Use Bots to Beat Bot Detection

If you’re using an AI bot to outwit chatbot detectors, it’s a losing game. Eventually, there will be a chat detection program designed to understand the type of content the “undetectable chatbot” produces. And this detective AI will ferret out articles or essays that the “invincible” bot has written.

If a bot is writing something for you, another bot already can — or will eventually be able to — suss it out.

How Accurate Are Chat Detection Tools?

No tool will tell you something was 100% written by a human or bot. But if tells you there was a strong likelihood of AI, or that it’s unclear, a teacher or boss will probe further for answers.

ChatGPT detection tools can give you a good general read but with some caveats. None of the online tools is 100% accurate, nor do they claim to be. Rather, they give a score for the likelihood that a bot was used to write a particular article.

The best way is to use several AI detection tools. If 5 tools all rate the piece extremely likely to have been written by a bot (i.e., 85 to 95 percent likelihood), that’s a solid indicator. But if the tools give mixed results, then you may need more information. This might include comparing earlier writing samples (from essays written in the classroom in text booklets) against the tax you are reviewing for possible chatbot authorship.

Even if you use a dozen tools that all detect a high probability of bot usage, tread carefully with accusations. False positives exist. Human-written language that is hyper-generic, sounds “salesy,” or does not cite sources may be flagged in error. And another reason to stress the importance of referencing sources, even when paraphrasing.

Chat tools may be most accurate when you use three or more at once, and use these checkers in tandem with plagiarism checker tools, like Copyscape.

Common FAQs

Can a Teacher or Professor Tell if I Used Chat GPT to Write a Paper?

In short, yes — mostly. Teachers and professors are generally able to recognize certain patterns of language and structure in papers written by ChatGPT. Large institutions are also providing programs to check for AI and plagiarism.

ChatGPT is known to produce factual errors and show a lack of original analysis in its writing. It pulls data and summarizes what already exists, and it pulls from the analyses of other writers. When offering an analysis or reflection on a subject that is not yours, credit should be given to the original author even when you paraphrase their words.

This means academic honesty and hard work are your best bet, but this doesn’t mean Chat GPT is the enemy. Academic institutions are exploring how to teach students to use AI in responsible ways while innovating their own means of assessing students’ work (like this Princeton University memo to faculty and students, “AI and the classroom.”)

What’s the Difference Between Tools to Check for Plagiarism vs. Chatbots?

Programs that detect plagiarism will look for changes from active to passive voice, rearranging of a sentence or paragraph, usage of synonyms, and paraphrasing (depending on its proprietary algorithms), as well as checking text against a database of journal articles.

Some of these tools, like Grammarly, will flag texts for plagiarism and AI writing. While none of these programs are totally accurate, students and writers must be aware of both their capabilities and shortcomings.

Chatbot detector tools understand how the chatbots operate and the tell-tale signs to look for; they’re bots programmed to understand how other bots are programmed.

Is Chat GPT Content Bad for SEO?

ChatGPT can be a useful tool for SEO writers, helping to generate ideas and overcome writer’s block, potentially leading to better productivity.

Image courtesy of

It cannot replace a human’s voice, accuracy, fact-checking, or creativity and insight. More content does not equal quality content.

Finally, ChatGPT cannot produce original content, draw from personal experience and expertise, or conduct interviews or primary research. Google’s SEO algorithm values Expertise, Authoritativeness, and Trustworthiness, referred to as E-A-T.

When factual accuracy improves, ChatGPT may become a more reliable workhorse for churning out generic articles based on second-hand sources.

Can Chat GPT be Detected if You Paraphrase?

Turnitin detects AI-written content, including AI content that has been paraphrased. Their product is geared toward schools and universities.

Maybe. Some companies like Turnitin claim that they can detect paraphrasing, while others like Wordtune offer paraphrasing services to help improve your rewrites.

As this technology is rapidly changing, capabilities will vary by company–they are only as good as their algorithms and AI language models.

Bottom Line

With AI, we find ourselves amid a revolution — if not revolt — which brings new opportunities and challenges, both to individuals and the larger society.

Regulation is lagging the technological advances, but no one is heeding the temporary cease-fire called for by tech leaders. They’ve expressed their concerns over the rapidly advancing technology and requested a six-month break in new AI developments “to give society a chance to adapt.”

Yes, it’s possible to use Chat GPT to write a decent article or essay for you. Beyond concerns of personal ethics, there are two main ways of getting caught.

Chat GPT and similar AI written contact are factually inaccurate. You’d have to put in a significant amount of work to fact-check and fact-correct the output– so much so that you might as well just write it yourself and ask Chat GPT to proof it for grammar, structure, and style afterward.

The second issue is the AI detection tools work well enough that if you use a handful of them, in concert, they can red-flag content that was probably written by bots.

Related Trending Reads

Leave a Reply

%d bloggers like this: