What is GPTZero
Large language models (LLMs) such as ChatGPT have made it easier than ever to create content. Of course, AI content is not suitable for every situation. That’s where AI detectors like GPTZero come in.
GPTZero is a software application used to analyze text to determine whether the text was written by a human or an AI application. When it analyzes the text, it’ll provide a simple assessment stating how it was written.
Additionally, it’ll provide a confidence assessment such as ‘highly confident.’ Finally, you’ll get a percentage stating how much of it was human-written and how much of it was AI written.
In this GPTZero review, you’ll get an in-depth breakdown of how it works, its accuracy, and its core features.
How does GPT Zero work
GPTZero is easy to use. You won’t have to spend time learning new software to get the hang of it.
Step 1: Take the text in question and paste it into the text input box available on GPTZero or upload a file.
Step 2: Click the ‘scan’ button and it will start to analyze the text. The time spent depends on the length of the text
Step 3: Review the results GPTZero produces.
GPTZero claims to use seven factors to determine whether content was written by AI. These include an education model aimed specifically at educators, internet text search, a sentence-by-sentence classification model, defense against common shortcomings of AI detectors, deep learning, perplexity, and burstiness.
Perplexity and burstiness are two major factors for detecting AI text. Perplexity is how common the sentence or phrase is in the language. Burstiness is the variability of the sentence structure and length. This is a simplification.
The more variability the text has in sentence structure, the more likely it is to be written by humans. The more uncommon phrase and word usage is, the less likely it is to be written by AI.
In addition to the core AI detection, it also has multiple other features such as plagiarism detection, authorship verification, source finder, and more. The features will be looked at in more detail later in this article.
Is GPTZero accurate – my test
I tested three pieces of text with GPTZero. One was written by me. One was written by ChatGPT. One was written by ChatGPT but I changed the prompt in a way that was specifically designed to make the writing more human-like.
The results of the one written by me were as expected. It was highly confident that it was written by a human. There was only a 2% probability that it was written by AI.
The results of the one written by ChatGPT without additional prompting were also unsurprising. It highlighted the entire text and said it was 100% AI written with high confidence.
When I started getting serious about bypassing GPTZero by altering my prompt structure, GPTZero started to struggle a bit. It was able to classify it as AI-generated but it wasn’t confident. In fact, it said it was uncertain.
When to use GPTZero
There are countless scenarios where it may make sense to use an AI detector like GPTZero. A few common use cases are below but don’t let these limit you. You can use it however you like.
Schools and educators
There’s a lot of controversy around using AI detectors in schools. That’s because there are inherent inaccuracies and biases in the tools – GPTZero included.
A study by Scribbr showed that the most accurate paid detector was correct 84% of the time. The most accurate free detector was correct 68% of the time. In a school setting, up to 32% of students may be falsely accused and receive real repercussions even though they did the work honestly.
Because of this challenge and the way AI detection tools work, many schools have banned the use of AI content detectors outright or frown upon its use.
If you’re using it in a school setting, take the results with a grain of salt. Don’t jump to conclusions and start to accuse students of cheating. Instead, it may be better to work with them, let them know why their writing was flagged, and try to find common ground.
Content editors
As a content editor, the last thing you want is content without a voice. You can use AI detectors to quickly sift out people who may be using AI to write all or part of their work.
Armed with this information, you can provide specific points for them to work on to enhance their work and make it more engaging. Conversely, you can use a tool like Humanize AI to improve AI written content.
HR and recruiters
Oftentimes, applicants are required to submit a cover letter that’s relevant to the position they’re applying for. You can use an AI detector to check if they’ve written it themselves or they’ve used AI.
Of course, you have to be skeptical and not base your assessment on the AI detector alone. There is a margin of error and you don’t want to chase away a good candidate because of a false positive.
Use the AI detector in conjunction with other assessment tools and strategies to narrow the applicant pool.
Community moderators
There’s a deluge of automated tools that will comment on popular posts for whatever reason. You can use AI detection tools to quickly remove AI-generated answers and comments. This will allow real engagement to move to the top of the pile.
GPTZero pricing
GPTZero has four pricing tiers.
- Free – 10,000 words per month and only the most basic features.
- Essential ($14.99/m)– 150,000 words per month and it comes with more advanced features and tools such as vocabulary checker and a Chrome extension.
- Premium ($23.99/m) – 300,000 words per month and comes with the advanced scan functionality as well as writing feedback.
- Professional ($45.99/m) – 500,000 words per month, greater security, team management, and support for overages.
Features
GPTZero has a wide range of features. Of course, it has the core AI content detection. It also comes with other writing aids that folks may find useful. Some of the most important features include the following:
AI content analysis
As mentioned earlier, it uses multiple factors to determine whether a piece of text has been written by AI or a human. The major factors used to determine whether it’s AI are perplexity and burstiness.
While not 100% accurate, it can point you in the right direction and help you create higher-quality content.
Writing feedback
It enlisted the help of a few professors to provide various types of feedback for a piece of content. The feedback centers on structural changes such as whether the sentence is too wordy, repetitive, etc.
If you’ve struggled with concise and clear writing, this can be helpful for you.
Plagiarism detection
Plagiarism needs no introduction. This feature compares your writing to a large database compiled by GPTZero and sources on the internet. It’ll quickly flag any content that has been published elsewhere.
Of course, just because a piece of content has been used elsewhere, does not mean it’s plagiarism. It’ll also tell you how much of the writing is plagiarism and the source of the content. With this info, you can make an educated decision about whether or not plagiarism occurred.
File upload
Instead of copying and pasting your text from one place to another, you can upload the file directly. You can choose a single file or multiple files. This can save you considerable time and energy if you’re checking a large collection of text.
Conclusion
GPTZero is a new tool launched in 2023 that has seen a lot of success in a short period of time. For simple text generated by AI, it does its job well. For more complex pieces that are meant to mimic a certain writing style or have been edited down, it may struggle.
No AI detection tool is perfect. There will be false positives and false negatives. For any AI detection tool you use, be sure to use it in conjunction with other methods to determine the authenticity.
