The Most Accurate AI Content Detector
Try Our AI Detector
AI Studies

Can Humans Detect AI-Generated Text? 6 Studies Would Suggest They Can’t

As AI continues advancing, it’s essential to establish whether humans can identify AI-generated text. To evaluate this further, we reviewed six independent studies on the subject.

Trusted By Industry Leaders
Trusted By Industry Leaders

Introduction

Our text compare tool is a fantastic, lightweight tool that provides plagiarism checks between two documents. Whether you are a student, blogger or publisher, this tool offers a great solution to detect and compare similarities between any two pieces of text. In this article, I will discuss the different ways to use the tool, the primary features of the tool and who this tool is for. There is an FAQ at the bottom if you run into any issues when trying to use the tool.

What makes Originality.ai’s text comparison tool stand out?

Keyword density helper – This tool comes with a built-in keyword density helper in some ways similar to the likes of SurferSEO or MarketMuse the difference being, ours is free! This feature shows the user the frequency of single or two word keywords in a document, meaning you can easily compare an article you have written against a competitor to see the major differences in keyword densities. This is especially useful for SEO’s who are looking to optimize their blog content for search engines and improve the blog’s visibility.

Ways to compare

File compare – Text comparison between files is a breeze with our tool. Simply select the files you would like to compare, hit “Upload” and our tool will automatically insert the content into the text area, then simply hit “Compare” and let our tool show you where the differences in the text are. By uploading a file, you can still check the keyword density in your content.

URL compare

Comparing text between URLs is effortless with our tool. Simply paste the URL you would like to get the content from (in our example we use a fantastic blog post by Sherice Jacob found here) hit “Submit URL” and our tool will automatically retrieve the contents of the page and paste it into the text area, then simply click “Compare” and let our tool highlight the difference between the URLs. This feature is especially useful for checking keyword density between pages!

Simple text compare

You can also easily compare text by copying and pasting it into each field, as demonstrated below.

Features of Originality.ai’s Text Compare Tool

Ease of use

Our text compare tool is created with the user in mind, it is designed to be accessible to everyone. Our tool allows users to upload files or enter a URL to extract text, this along with the lightweight design ensures a seamless experience. The interface is simple and straightforward, making it easy for users to compare text and detect the diff.

Multiple text file format support

Our tool provides support for a variety of different text files and microsoft word formats including pdf file, .docx, .odt, .doc, and .txt, giving users the ability to compare text from different sources with ease. This makes it a great solution for students, bloggers, and publishers who are looking for file comparison in different formats.

Protects intellectual property

Our text comparison tool helps you protect your intellectual property and helps prevent plagiarism. This tool provides an accurate comparison of texts, making it easy to ensure that your work is original and not copied from other sources. Our tool is a valuable resource for anyone looking to maintain the originality of their content.

User Data Privacy

Our text compare tool is secure and protects user data privacy. No data is ever saved to the tool, the users’ text is only scanned and pasted into the tool’s text area. This makes certain that users can use our tool with confidence, knowing their data is safe and secure.

Compatibility

Our text comparison tool is designed to work seamlessly across all size devices, ensuring maximum compatibility no matter your screen size. Whether you are using a large desktop monitor, a small laptop, a tablet or a smartphone, this tool adjusts to your screen size. This means that users can compare texts and detect the diff anywhere without the need for specialized hardware or software. This level of accessibility makes it an ideal solution for students or bloggers who value the originality of their work and need to compare text online anywhere at any time.

Can humans detect AI-generated text?

Some argue they can, while others rely on high-performing AI detection tools such as Originality.ai.

With the high impact of identifying (or misidentifying) AI-generated content, it’s essential to establish if humans can detect AI-generated text. To evaluate this further, we reviewed six independent studies on the subject.

Key Takeaways (TL;DR)

  • Despite confidence in their own ability, humans tend to struggle to spot AI-generated content.
  • AI content detectors are significantly more effective than humans at identifying AI content.
  • There are some significant implications of not being able to detect AI content, such as the spread of misinformation, academic dishonesty, and a lack of authenticity online.

Can Humans Detect AI-Generated Text? 6 Studies Explained 

Despite high confidence, people typically struggle to detect AI-generated text. Our study on the most commonly used ChatGPT words found that ChatGPT is harder to identify than most think. 

To evaluate this further, we’re summarizing and reviewing six third-party studies on the subject of humans detecting AI-generated text.

1. Do teachers spot AI? Evaluating the detectability of AI-generated texts among student essays

Study overview

The first study is available through ScienceDirect and investigates the ability of teachers to detect AI-generated content.

As AI continues to advance, detecting AI-generated content in education is becoming increasingly important. So, the researchers in this study conducted a series of experiments to see if teachers could differentiate between AI text and human-written text.

The study incorporated two groups of participants:

Study 1 (participants were pre-service teachers)

 “Pre-service teachers identified only 45.1 percent of AI-generated texts correctly and 53.7 percent of student-written texts.”

Study 2 (participants were experienced teachers)

 “They identified only 37.8 percent of AI-generated texts correctly, but 73.0 percent of student-written texts.”

Key findings

  • The participants were unable to consistently identify AI and human-written content.
    • Pre-service teachers only accurately identified 45.1% of AI texts.
    • Experienced teachers only accurately identified 37.8% of AI texts.
  • Even when participants felt confident, their judgements were often incorrect, highlighting an overestimation of the human ability to detect AI.

Conclusion

The findings show that teachers struggle to distinguish AI-generated content from human-written work without support from leading AI detection tools.

Study: Do teachers spot AI? Evaluating the detectability of AI-generated texts among student essays

Source: ScienceDirect

2. All That’s ‘Human’ Is Not Gold: Evaluating Human Evaluation of Generated Text

Study overview

In this study, accessible through Cornell University, researchers again assessed participants' ability to differentiate between content produced by AI or written by humans.

The content researched in this study included “stories, news articles, and recipes.” 

  • 255 stories were reviewed for the study (sourced from social media posts).
    • 50 of the 255 human-written stories were chosen.
    • 50 additional stories were generated from prompts.
  • 2,111 news articles were reviewed for the study (sourced from 15 news publications).
    • 50 of the 2,111 articles were chosen.
    • 50 additional articles were generated from prompts.
  • 50 human-written recipes from the RecipeNLG dataset were included.
    • 50 additional recipes were generated from prompts.

Key findings

  • Again, the results showed that participants had a poor success rate in identifying AI-generated content.
  • Participants had 57.9% accuracy when identifying GPT2 vs. human content (rounded to 58% in the published table).
  • Then, study participants had a 49.9% accuracy at identifying GPT3 vs. human content (rounded to 50% in the published table).
  • When participants were trained in AI recognition before the test, their scores slightly improved, but the results were not substantial.

Conclusion

The findings further show that humans struggle to identify AI-generated text, even if they have received some form of training prior to the task.

Find out about the accuracy of AI detectors according to third-party researchers in a meta-analysis of AI-detection accuracy studies.

Study: All That's 'Human' Is Not Gold: Evaluating Human Evaluation of Generated Text

Source: Cornell University

3. People cannot distinguish GPT-4 from a human in a Turing test

Study overview

This study, also accessible through Cornell University, offers additional insight into the ability of humans to detect AI-generated content produced by three AI models: GPT-4, GPT-3.5, and ELIZA.

The study focused on how successful humans are at identifying AI-generated text. It included 500 participants and evaluated their AI-detection efficacy using a ‘game’ that resembled a messaging app.  

Further, the study evaluated whether or not prior knowledge or familiarity with AI and large language models (LLMs) could impact the results.

Key findings

  • Similar to previous studies, research shows that participants were overall unable to detect AI-generated text and differentiate it from human-written text.
    • Human text was correctly identified as human-written 67% of the time.
    • GPT-4 was incorrectly identified as human 54% of the time.
    • ELIZA was incorrectly identified as human 22% of the time
  • With regard to the unique aspect of familiarity with LLMs in this study, LLM knowledge had a marginal impact on accuracy.
  • Interestingly, age was the largest influential factor in terms of accuracy. Younger age groups generally identified AI models more accurately than older age groups.

Conclusion

This study further highlights that humans struggle to identify AI-generated text consistently, even if they are familiar with LLM subject matter.

Interested in learning more about AI detection and using AI detectors? Read the Originality AI Content Detector Accuracy Review for more information.

Study: People cannot distinguish GPT-4 from a human in a Turing test

Source: Cornell University

4. Can you spot the bot? Identifying AI-generated writing in college essays

Study overview

The International Journal for Educational Integrity also conducted a detailed study into AI detection, focusing specifically on academic writing. The aim was to see whether participants could identify AI-generated essays from human-written ones.

The study included 140 college instructors and 145 college students as participants. Researchers provided each participant with one human-written student essay and one essay generated by AI. Then, they were asked to evaluate which was AI-generated and which was student-written.

Key findings

  • The study showed that participants struggled to accurately identify AI-generated essays.
  • College instructors correctly identified ChatGPT 70% of the time.
  • College students correctly identified ChatGPT 60% of the time.

Conclusion

The results demonstrate that both educators and students have difficulty identifying AI usage in academic work.

Study: Can you spot the bot? Identifying AI-generated writing in college essays

Source: International Journal for Educational Integrity via BioMed Central Ltd  

5. Identification of ChatGPT-Generated Abstracts Within Shoulder and Elbow Surgery Poses a Challenge for Reviewers

Study overview

The fifth study on this list is available in Arthroscopy: The Journal of Arthroscopic & Related Surgery, and focuses on a reviewer’s ability to identify ChatGPT-generated abstracts.

As with other studies, participants were shown a mixture of AI-generated and human-written content, this time in the format of scientific abstracts, and were asked to select which ones they believed were AI-generated.

Key findings

  • Reviewers correctly spotted AI content 62% of the time
  • 38% of the time, original human-written abstracts were incorrectly identified as AI.
  • As a result of the difficulties reviewers faced, the study found that AI detection methods beyond human evaluation are needed.

Conclusion

As with the other studies, this further proved that humans struggle to detect AI-generated content.

Study: Identification of ChatGPT-Generated Abstracts Within Shoulder and Elbow Surgery Poses a Challenge for Reviewers

Source: Arthroscopy: The Journal of Arthroscopic & Related Surgery

6. Can ChatGPT Fool the Match? Artificial Intelligence Personal Statements for Plastic Surgery Residency Applications: A Comparative Study

Study overview

Lastly, we have the study from the Canadian Society of Plastic Surgeons, focusing on content in the highly specialized context of medical residency applications.

The participants were two recently retired surgeons with 20 years of experience in the CaRMS (Canadian Residency Matching Service). They were asked to evaluate 11 AI and 11 human applications and identify whether they were human-written or generated by ChatGPT-4.

Key findings

  • The study showed that evaluators successfully differentiated between AI and human applications 65.9% of the time.
  • The paper acknowledged that the small sample size was a limitation of the study, while the evaluator’s expertise was a notable strength. 

Conclusion

The findings once again show that identifying AI content is challenging for humans. It even proves challenging for experienced professionals with over two decades of experience in medical residency applications to identify human vs. AI residency applications.

Study: Can ChatGPT Fool the Match? Artificial Intelligence Personal Statements for Plastic Surgery Residency Applications: A Comparative Study

Source: Canadian Society of Plastic Surgeons

Final Thoughts

So, can humans detect AI-generated text? As demonstrated by these studies, the answer is no. 

While humans can correctly identify AI text in some instances, they still struggle overall to differentiate AI content from human-written content. As a result, for additional clarity, an accurate AI content detector is highly beneficial.

FAQs About AI Detection

Can humans detect AI-generated content?

In general, humans struggle to detect AI-generated content. Studies show that human accuracy can vary significantly.

How accurate are AI detection tools?

The efficacy of AI detection tools varies depending on the tool. The Originality.ai Lite AI Checker exhibits 98% accuracy in AI detection. Additionally, the Turbo model demonstrates 99%+ AI detection accuracy.

To compare the efficacy of popular AI tools, check out our AI detection review series

What are the implications of not being able to detect AI content?

There can be significant implications, including the spread of misinformation, academic dishonesty, and a lack of authenticity in online information.

Jonathan Gillham

Founder / CEO of Originality.ai I have been involved in the SEO and Content Marketing world for over a decade. My career started with a portfolio of content sites, recently I sold 2 content marketing agencies and I am the Co-Founder of MotionInvest.com, the leading place to buy and sell content websites. Through these experiences I understand what web publishers need when it comes to verifying content is original. I am not For or Against AI content, I think it has a place in everyones content strategy. However, I believe you as the publisher should be the one making the decision on when to use AI content. Our Originality checking tool has been built with serious web publishers in mind!

Frequently Asked Questions

Do I have to fill out the entire form?

No, that’s one of the benefits, only fill out the areas which you think will be relevant to the prompts you require.

Why is the English so poor for some prompts?

When making the tool we had to make each prompt as general as possible to be able to include every kind of input. Not to worry though ChatGPT is smart and will still understand the prompt.

In The Press

Originality.ai has been featured for its accurate ability to detect GPT-3, Chat GPT and GPT-4 generated content. See some of the coverage below…

View All Press
Featured by Leading Publications

Originality.ai did a fantastic job on all three prompts, precisely detecting them as AI-written. Additionally, after I checked with actual human-written textual content, it did determine it as 100% human-generated, which is important.

Vahan Petrosyan

searchenginejournal.com

I use this tool most frequently to check for AI content personally. My most frequent use-case is checking content submitted by freelance writers we work with for AI and plagiarism.

Tom Demers

searchengineland.com

After extensive research and testing, we determined Originality.ai to be the most accurate technology.

Rock Content Team

rockcontent.com

Jon Gillham, Founder of Originality.ai came up with a tool to detect whether the content is written by humans or AI tools. It’s built on such technology that can specifically detect content by ChatGPT-3 — by giving you a spam score of 0-100, with an accuracy of 94%.

Felix Rose-Collins

ranktracker.com

ChatGPT lacks empathy and originality. It’s also recognized as AI-generated content most of the time by plagiarism and AI detectors like Originality.ai

Ashley Stahl

forbes.com

Originality.ai Do give them a shot! 

Sri Krishna

venturebeat.com

For web publishers, Originality.ai will enable you to scan your content seamlessly, see who has checked it previously, and detect if an AI-powered tool was implored.

Industry Trends

analyticsinsight.net

Frequently Asked Questions

Why is it important to check for plagiarism?

Tools for conducting a plagiarism check between two documents online are important as it helps to ensure the originality and authenticity of written work. Plagiarism undermines the value of professional and educational institutions, as well as the integrity of the authors who write articles. By checking for plagiarism, you can ensure the work that you produce is original or properly attributed to the original author. This helps prevent the distribution of copied and misrepresented information.

What is Text Comparison?

Text comparison is the process of taking two or more pieces of text and comparing them to see if there are any similarities, differences and/or plagiarism. The objective of a text comparison is to see if one of the texts has been copied or paraphrased from another text. This text compare tool for plagiarism check between two documents has been built to help you streamline that process by finding the discrepancies with ease.

How do Text Comparison Tools Work?

Text comparison tools work by analyzing and comparing the contents of two or more text documents to find similarities and differences between them. This is typically done by breaking the texts down into smaller units such as sentences or phrases, and then calculating a similarity score based on the number of identical or nearly identical units. The comparison may be based on the exact wording of the text, or it may take into account synonyms and other variations in language. The results of the comparison are usually presented in the form of a report or visual representation, highlighting the similarities and differences between the texts.

String comparison is a fundamental operation in text comparison tools that involves comparing two sequences of characters to determine if they are identical or not. This comparison can be done at the character level or at a higher level, such as the word or sentence level.

The most basic form of string comparison is the equality test, where the two strings are compared character by character and a Boolean result indicating whether they are equal or not is returned. More sophisticated string comparison algorithms use heuristics and statistical models to determine the similarity between two strings, even if they are not exactly the same. These algorithms often use techniques such as edit distance, which measures the minimum number of operations (such as insertions, deletions, and substitutions) required to transform one string into another.

Another common technique for string comparison is n-gram analysis, where the strings are divided into overlapping sequences of characters (n-grams) and the frequency of each n-gram is compared between the two strings. This allows for a more nuanced comparison that takes into account partial similarities, rather than just exact matches.

String comparison is a crucial component of text comparison tools, as it forms the basis for determining the similarities and differences between texts. The results of the string comparison can then be used to generate a report or visual representation of the similarities and differences between the texts.

What is Syntax Highlighting?

Syntax highlighting is a feature of text editors and integrated development environments (IDEs) that helps to visually distinguish different elements of a code or markup language. It does this by coloring different elements of the code, such as keywords, variables, functions, and operators, based on a predefined set of rules.

The purpose of syntax highlighting is to make the code easier to read and understand, by drawing attention to the different elements and their structure. For example, keywords may be colored in a different hue to emphasize their importance, while comments or strings may be colored differently to distinguish them from the code itself. This helps to make the code more readable, reducing the cognitive load of the reader and making it easier to identify potential syntax errors.

How Can I Conduct a Plagiarism Check between Two Documents Online?

With our tool it’s easy, just enter or upload some text, click on the button “Compare text” and the tool will automatically display the diff between the two texts.

What Are the Benefits of Using a Text Compare Tool?

Using text comparison tools is much easier, more efficient, and more reliable than proofreading a piece of text by hand. Eliminate the risk of human error by using a tool to detect and display the text difference within seconds.

What Files Can You Inspect with This Text Compare Tool?

We have support for the file extensions .pdf, .docx, .odt, .doc and .txt. You can also enter your text or copy and paste text to compare.

Will My Data Be Shared?

There is never any data saved by the tool, when you hit “Upload” we are just scanning the text and pasting it into our text area so with our text compare tool, no data ever enters our servers.

Software License Agreement

Copyright © 2023, Originality.ai

All rights reserved.

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

  1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.

  1. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

Will My Data Be Shared?

This table below shows a heat map of features on other sites compared to ours as you can see we almost have greens across the board!

More From The Blog

Al Content Detector & Plagiarism Checker for Marketers and Writers

Use our leading tools to ensure you can hit publish with integrity!