OpenAI released GPT-5 on Aug 7, 2025 and this study looks to understand if AI detectors remain accurate on GPT-5 generated content.
Additionally, we examine what makes GPT-5 unique in terms of writing. Not just is it still detectable but what are its unique features, does it overuse delve, em-dash etc?
Yes, GPT-5 content is able to be detected by Originality.ai at a 96.5% accuracy rate. Currently, GPT-5 content is less detectable than GPT-4o; however, that gap will close quickly as Originality.ai trains on more content from GPT-5.
99% Accurate for Rewritten Content based on a quick test of 100 samples
94% Accurate for Newly Generated GPT-5 Content based on a quick test of 100 samples
Method:
Results:
Can other detectors such as GPTZero, TurnitIn, CopyLeaks, Quillbot and Grammarly detect GPT-5 generated content?
Method:
The first 10 samples from the GPT-5 rewritten content are tested against multiple detectors.
Results:
Note - I expect the performance of all AI detectors to increase significantly over the coming weeks/months as models are updated.
Open AI says that GPT-5 is the “most capable writing collaborator yet”
Basically, what this means is that OpenAI says GPT-5 can…
Out of the 100 human-AI rewrites, none of the 100 human samples used any em-dash’s. However, the GPT-5 rewritten content included the em-dash in 67 of the articles.
Previous models GPT-4o etc often rewrote content and the output of the content often clustered around similar Flesch Kincaid Readability scores.
GPT-5 is being positioned as a superior writing tool that is able to follow a specific style more consistently.
If that is the case it should show up in readability scores staying consistent with the human text it is being asked to re-write.
To test this we looked at dataset of Human content and AI content to see how consistent the readability scores were between the input and output using our Readability Checker.
Results:
GPT-5 mirrored the readability score of the human content
Average Flesch Kincaid Readability Score:
With more spread for AI written content (19.1 vs 14.9 for human written)
There is a clear, moderately strong positive relationship. Higher-readability human originals tend to yield higher-readability AI rewrites.
GPT-5 rewrites generally mirror the clarity of their human source material. In our 100-sample test, human passages averaged a Flesch–Kincaid score of 49.6, while GPT-5’s versions landed at 46.5—only about three points lower and still in the same readability band. The two sets of scores are moderately correlated (r ≈ 0.58), so a well-written original usually stays well-written after the AI pass. The main difference is consistency: AI outputs show a wider spread in scores (standard deviation 19 vs 15), meaning they sometimes swing from ultra-concise to slightly dense. Bottom line: GPT-5 tends to keep the readability level you give it.
GPT-5 content is detectable by Originality.ai but for now it makes AI detectors less accurate. We would expect all leading AI detectors to get more accurate in the coming weeks/months on the latest model release from GPT-5.
Yes - GPT-5 still loves to use the Em-Dash
GPT-5 mirrors the style/readability score of human writing more than previous models.
No, that’s one of the benefits, only fill out the areas which you think will be relevant to the prompts you require.
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.
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
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.
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.
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.
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.
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.
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.
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.
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.
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:
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.
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!