How To Detect AI Writing: Top Tools and Techniques

Artificial intelligence (AI) and generative AI tools like GPT-4 by OpenAI are becoming increasingly advanced. These AI writing tools can create text so well-written that, in a recent study from Cornell University, participants believed the chatbot they were interacting with was human over 40% of the time.

While this failed the Turing test, it calls to question our ability to distinguish AI-written text from human-written text. The way many of these tools are trained raises concerns about copyright infringement and plagiarism. And the way they operate—by applying machine learning techniques to training data in order to provide the most average output possible—raises concerns about quality.

To combat these challenges, AI detection tools have surfaced as a way to screen for originality against rising AI-generated text. These tools allow users to balance the increased efficiency AI promises, while maintaining the creativity of human expression.

In this article, we’ll discuss why AI detection is important and detection techniques, and provide a list of top tools you can start using today.

Table of contents:

• Understanding AI writing
• Why checking for AI writing is important
• How to detect AI writing
• AI detection tools
• Content and plagiarism detection tools
• Find content writers and editors on Upwork

Understanding AI writing

AI writing is an advanced technology powered by large language models (LLMs) like GPT-3.5 and GPT-4. These models use advanced algorithms and were trained on extensive datasets using cutting edge machine learning techniques, and can generate text that is hard to distinguish from human-written text.

Many companies build proprietary AI models using company-owned data. These systems generally have very defined functions and goals. AI writing might be used to help operators in a call center answer customer service questions, or interact with consumers directly as a chatbot on a webpage.

Many publicly available AI models, on the other hand, are made to perform a wider variety of tasks and were trained by accessing publicly available data, largely from the internet. While the companies aren’t forthcoming about the exact nature of the training data, it likely included news articles, social media sites, SEO content, and other web pages available from CommonCrawl and RefinedWeb.

The machine learning techniques used to train these LLMs enable them to analyze and process text. Users input prompts that move through a series of gated nodes in the AI’s programming. At each node, the AI’s algorithms give the prompt a certain weight. How it’s weighted determines which node it moves to next, essentially creating a very complicated decision tree that enables the AI to give a statistical prediction to an output a user is likely to accept.

Because interactive AI programs like ChatGPT are so accessible, countless companies and individuals have used them to create content. With the right prompt engineering, these tools can create drafts of content in seconds. Projects that used to take five to seven business days can now be delivered in one. However, AI writing poses challenges including the risk of plagiarism, monotony, and the missing human touch.

Companies have enabled AI models like Google Bard, Microsoft Bing, and ChatGPT to access the internet, which has increased their usefulness as writing and research aids. However, AI still lacks the capacity to understand anything, or to have real context for what it’s asked to write. It lacks ethics and can replicate existing content, it lacks the ability to add emotional nuance to text, and it can’t give a piece the creativity that natural human writing would have.

Why checking for AI writing is important

AI writing brings up many issues related to ethics, integrity, and morals. A sector where these issues are at the fore is education.

Because of AI, plagiarism detection has become more complicated, and educators struggle to know if an assignment is the product of the student’s own work or an AI’s. Student essays, academic papers, and school applications all have the potential to be a product of advanced AI algorithms.

And the issue doesn’t stop at schools—it bleeds into sectors like marketing, journalism, and public relations (PR). Without putting in the effort to detect AI writing, professionals in roles can risk their credibility and cause ethical issues for a company.

In addition to challenges of plagiarism and authorship, AI can also hallucinate and generate factually incorrect information. This is due to one of AI’s most important parameters: To generate content it predicts the user will accept. As a user’s prompt filters through the AI’s algorithms, the system will sometimes generate false data in order to give what it predicts will be an acceptable response.

How to detect AI writing

As AI becomes more and more sophisticated, knowing how to detect AI-generated text will soon be an essential skill in education, academia, and the workplace. This section will guide you through some steps to identify AI writing.

1. Use AI detection tools

Your first step to detecting AI-written content is relying on AI detection tools like Turnitin or Copyleaks. (Interested in browsing some top tools? Check out our list below.)

To use these tools, you’ll typically upload a document or copy and paste blocks of text into the program. The system will then analyze the text using algorithms designed to detect the patterns in human-generated text versus AI-generated text.

Keep in mind that these tools aren’t perfect, and only advertise detection rates of up to 97-99%. While really good, this isn’t enough to definitively flag AI writing. These tools are good as an initial filter but should be used in tandem with other AI detection techniques and methods.

2. Perform language analysis

Once you’ve done an initial scan with an AI detection tool, the second step is to do a language analysis. Language analysis techniques like n-gram analysis can uncover inconsistencies and repetitive patterns within AI-generated text.

AI writing is often predictable and repetitive in its language, and it often struggles to be consistent when it comes to emulating a writer’s style throughout a long piece of text. These techniques allow you to detect these inconsistencies and serve as red flags so you can spot heavy or inappropriate use of AI.

3. Check for semantic consistency

AI writing tools often struggle to maintain a coherent narrative, which means you’ll likely see big shifts in subject matter or unnatural transitions that seem odd or illogical to a human reader.

AI-generated text is generated one space at a time by algorithms that are predicting the symbol (a letter, number, punctuation, etc.) most likely to be accepted by the user.  The longer the generated content is,  the more likely the AI will be to steer away from the original query and hop from one idea to another without clear connections.

4. Investigate source authenticity

When you’re scrutinizing AI writing, consider both how the text sounds and also the authenticity of the sources or references. AI writing tools can generate pieces that include fictitious references or incorrect citations—which is a strong indicator of AI-written text.

Be sure to always check whether any cited source exists and if the information accurately reflects the text.

5. Employ plagiarism detection tools

Plagiarism—using someone else’s work without proper recognition—is one of the most concerning aspects of AI. Many publicly available AI writing tools were fine-tuned on extensive datasets that were publicly available. The work of thousands or possibly millions of creators went into training these models—and that work is not recognized by the AI. This calls to question the originality of anything these AI bots generate.

At the same time, AI writing tools that have access to the internet are known to post as output text that has been copied directly from a webpage. While the AI might link to its source—which is a form of attribution and so isn’t plagiarism—a writer who uses that output without attribution to the source has plagiarized.

Many plagiarism detection tools work by matching strings of characters with text that is published on the internet. A writer who copies work in this way, without attribution, has both plagiarized and possibly committed copyright infringement—in this case, reproducing the expression of an idea without permission.

Of course, not all matching text means a piece is plagiarized. Quotes and commonly used phrases can lead to matches, and fair use laws allow for the public to have a discourse about a topic without infringing on copyrights. So it’s important to look at these results in a holistic manner.

AI detection tools

Now it’s time to go through a few useful AI content detection tools that will help you identify AI-generated text amid a plethora of digital content. Whether in an academic or professional setting, these tools will help you sift through digital content and distinguish AI vs. human-generated text.

Copyleaks

Copyleaks is designed as an enterprise AI detection solution with algorithms that detect AI-written text with 99.1% accuracy. Other than being able to scan the web and identify text generated by GPT-4 and Bard, it’s also able to detect AI-generated source code from GitHub, ChatGPT, and more.

Features:

• Sophisticated AI-based algorithms to detect AI-generated text
• Multilingual detection capabilities (30+ languages)
• API and LMS integrations
• Free Chrome browser extension for easy access

Pricing:

While Copyleaks has a free, basic AI-detection tool, they also have an enhanced model priced by the number of pages of scanned content. Their cheapest plan includes 1,200 pages and costs $10.99/month or $9.16/month when billed annually.

Originality AI

Originality AI focuses more on a blend of sophisticated techniques, such as textual patterns and linguistic nuances, to detect AI-generated text. It also offers features specifically for serious content creators as well as false positive management.

In addition to detection capabilities, Originality AI has automated fact-checking and readability scores for higher accuracy and efficiency in detecting AI-generated text.

Features:

• Advanced pattern recognition
• Comprehensive linguistic analysis
• High accuracy
• Readability test
• Multilanguage (15+ languages) capabilities with cultural nuance recognition

Pricing:

Originality AI has a base subscription model ($14.95/month for 2000 credits) and a pay-as-you-go system ($30 one-time fee), so you can opt for whichever pricing option works best for you.

Scribbr AI Detector

Scribbr AI Detector is perfect for blog posts and SEO content analysis, as it’s designed specifically for the content creation industry. Its primary capability is being able to tackle content created by ChatGPT and similar AI writing tools.

Features:

• Specific detection for ChatGPT-created content
• Useful for SEO, professional, and academic content

Pricing:

The Scribbr AI detector is completely free and allows you to check up to 500 words at a time.

Content and plagiarism detection tools

These content and plagiarism detection tools comb through digital content to spot duplicates from human-copied or AI-generated text to make sure a piece is authentic and original.

Turnitin

You might have heard of Turnitin as a plagiarism checker from your days at school. It’s now evolved to also detect AI-generated content. Turnitin has become an indispensable tool for educators and students around the world to uphold integrity and originality in academic papers.

Features:

• Extensive database for cross-checking plagiarism
• Detailed similarity report
• Reveal text manipulations
• Dashboards for educator insights

Pricing:

To find out more about pricing, contact Turnitin.

Grammarly

Grammarly is a widely used digital writing assistant best known for offering grammar and spell-checking functionalities as well as a plagiarism checker. While there is a free version that checks for duplicate content, the premium version will identify AI-generated text and highlight areas that need citations. With web, desktop, and browser extension versions available, it can offer real-time suggestions and corrections for ease of use.

Features:

• Plagiarism detection against billions of web pages
• Integration with multiple devices and platforms

Pricing:

Other than the Free plan, there is also a Premium Plan starting at $12/month and a Business Plan starting at $15/member/month.

Find content writers and editors on Upwork

AI writing tools are sophisticated tools that can enhance your productivity and the productivity of your workers. But they can’t match a human writer’s originality, creativity, and understanding. If you’re looking for skilled writers or editors who can offer those talents, Upwork is the perfect global hub to start searching.

Upwork is a marketplace teeming with freelance writers and editors who can craft authentic articles and take on meticulous editing tasks with ease. Businesses looking for high-quality, human-authored content can easily connect with qualified writers and editors.

Freelance writers or editors seeking new opportunities can discover writing and editing jobs and expand their portfolios on Upwork.

Ready to start your journey? Sign up on Upwork today!

‍

Disclosure: Upwork is an OpenAI partner, giving OpenAI customers and other businesses direct access to trusted expert independent professionals experienced in working with OpenAI technologies.

Upwork does not control, operate, or sponsor the other tools or services discussed in this article, which are only provided as potential options. Each reader and company should take the time to adequately analyze and determine the tools or services that would best fit their specific needs and situation.

Prices are current at the time of writing and may change over time based on each service’s offerings.