How to tell if something is AI-generated: potential indicators
In today's digital age, distinguishing between human-written and AI-generated content has become increasingly challenging. As artificial intelligence (AI) continues to evolve, its ability to mimic human writing is more convincing than ever, leaving readers questioning the authenticity of what they read.
ChatGPT wrote the two sentences above. Could you tell?
If this tricked you, read on to learn how to tell if something is written by AI. We'll explore the indicators of AI writing, strategies to identify AI-generated content, and the top AI content detectors you can use to simplify the detection process.
Spotting AI-generated content isn't always easy, but there are some potential indicators to watch for when reading a piece of content. Remember, just because content meets one or more criteria doesn't confirm that someone created it with an AI tool. (Be careful with accusations!)
Most notably, one of the common themes of AI-written content is that it's high-level and doesn't explore the topic beyond surface-level explanations.
If the content concerns a business concept, it often uses a lot of fluffy business jargon (e.g., bandwidth, leverage, synergy). It might also provide basic advice without specific information or steps to back it up. While the advice might make sense, it may not be tangible or applicable to the reader.
Although this advice technically answers my question, it's so vague that I have follow-up questions to help me apply it, including:
AI-generated content also lacks uniqueness and a human perspective. If there is no point of view, opinion, quotes from sources, specific examples, advice based on personal experience, or other anecdote to add a layer of personalization, the content could be AI-generated.
Many prompts provide outputs in numbered or bullet-point lists, which can sometimes seem repetitive.
All of these points might be true, but "increased productivity" and "enhanced focus and concentration" are similar points worded differently, and both descriptions start with "taking breaks..." This duplication is especially noticeable when it tries to achieve a numbered list.
Content with a robotic tone, a lack of personality, or a distinct voice may be AI-generated. In many cases, without a specific prompt indicating otherwise, AI-generated content is dry and flat. It doesn't necessarily read with a defined tone or personality.
Here's an example from ChatGPT using the prompt: Write a paragraph about the benefits of AI for knowledge workers.
A paragraph about the benefits of AI knowledge workers with a distinguishable tone and voice might instead sound something like:
Knowledge workers can reap the numerous benefits of AI tools in the workplace. The right AI tools can reduce the time workers spend doing the tasks they dread most from their to-do lists: writing and editing an excess of emails, drafting reports that (almost) no one reads, and analyzing large amounts of data without going cross-eyed at their laptops. Additionally, AI tools can help knowledge workers kickstart pattern and trend analysis, allowing them to identify valuable insights that can impress their boss and drive faster decisions.
In comparison, some AI-generated content might have too much flair.
Below is an example of the prompt above with tone specifications. Here's the prompt: Write a paragraph about the benefits of AI for knowledge workers. Be bold, inspirational, and educational.
Phrases like "the beacon illuminating the path," "transcend the boundaries of time and space," and "labyrinth of information overload" might seem too intense or displaced.
Large language models used in AI tools have a knowledge cutoff, which refers to the date when the data used for training the model was last updated. This means an AI tool can't pull information after its training cutoff, so current events, breaking news, and other recent information are unavailable. At the time of writing, most advanced and popular tools like Gemini and GPT4 have knowledge cutoff around mid-2023, meaning they can't produce accurate details beyond that.
Sometimes, detecting AI writing requires more thorough analysis. While there aren't any hard and fast rules for detecting AI-generated text, there are some strategies you can deploy.
One way to test content to see if it was either copied exactly or edited slightly from an AI tool is to experiment with prompts related to the content and see what the AI tools output in return. You may not be able to generate the exact content, especially if the author edited it before publishing. However, if you submit a prompt and receive an output that's formatted similarly, uses the same phrasing and keywords, or follows the outline of the published work, it's possible that the author used an AI tool.
As discussed before, repetitive words, phrases, and sentences are potentially obvious signs of AI-generated content. Machine learning models observe patterns and replicate them, which means they are relatively good at repeating the same information (or a pattern). For example, according to Plus AI, some of the words ChatGPT overuses include "delve," "embark," and "landscape."
Assuming the author is transparent, the most obvious way to confirm whether content is AI-generated is to look for a disclosure statement. Using AI tools to support the content process isn't inherently wrong. There are plenty of benefits to using these tools for brainstorming, outlining, and structuring the flow of content. However, as a best practice, anyone who uses an AI tool to generate content (whether entirely or as a supporting tool) should reveal and be transparent about it.
Following the surge in the popularity of AI tools, many people are now turning to AI content detector software to verify whether a piece of content or written assignment is AI-generated. Schools and teachers are using AI-detection tools to help differentiate between AI-generated and human-written work. However, the software is, at times, flawed. While there's some debate around whether or not AI content detectors are accurate, these tools are on the rise.
AI content detectors help businesses determine whether the content has been generated using AI, also known as synthetic media. These tools are similar to plagiarism checkers; although they focus specifically on detecting whether the content is AI-generated instead of written by another human being and stolen.
To qualify for inclusion in the AI content detectors category, a product must:
*Below are the top five leading AI content detector platforms based on G2's Fall 2024 Grid Report. Some reviews may be edited for clarity.
GPTZero is an AI detector that calculates and displays the probability that a text input was AI-generated by large language models like ChatGPT. You can scan up to 50,000 characters for free in their web-based platform. Some additional feature that the tool offers are multi-file upload, Chrome extension, integration with Google Docs, and suggestions for improvement in the content.
"GPTZero quickly and effectively analyzes text, emails, etc. to determine if they were generated or enhanced through the use of artificial intelligence. GPTZero logs your entries and provides you with a detailed understanding of how the texts were modified by AI."
- GPTZero Review, Theren B.
"While GPTZero can be helpful, it sometimes gives false positives or negatives, making it difficult to fully rely on the results. There are times when human-written content is flagged as AI-generated, and vice versa. This inconsistency can be frustrating, particularly when accuracy is critical, such as for academic or professional content verification."
- GPTZero Review, Ashutosh J.
Writer AI Content Detector offers two options for users. The first is a free AI detector tool that allows users to review up to 5,000 words. Through the second, subscribers can check up to 500,000 words monthly through the API. The Writer's AI detector assesses the likelihood that written content is AI-generated.
"The best thing I like about it is it gives the percentage of human-generated content for checking and analyzing the text, which gives more ideas to detect the AI content."
- Writer AI Content Detector Review, Ambika M.
"The one and only thing I didn't like is its character limit. But I don't think it's a major issue."
- Writer AI Content Detector Review, Manish A.
Undetectable AI offers both a free AI detector, which assesses the likelihood of AI-generated content, and an AI humanizer, which transforms AI content into a more human-like output. Users simply input AI-generated content into these tools and, in a matter of seconds, receive humanized AI content in return. It's designed to be undetectable by AI detection tools and reads as if it were written by a human.
"It's got a dead simple UI and processes things quickly without a need to register an account."
- Undetectable AI Review, Zack S.
"One additional helpful feature would be if it showed where it shows AI content based on the tool."
- Undetectable AI Review, Anwita D.
Content At Scale allows users to check for plagiarism and AI content by simply pasting their content or uploading their files in the application. It then quickly scans them and then shares probability results whether something reads like human or AI. For content that sounds robotic, it even offers an option to humanize it.
"The sleek and user-friendly interface, regular updates, and instant results with the highlighted AI text are my favorites."
- Content At Scale Review, Matheen H.
"The tool is highly complex to set up and requires time to learn and get comfortable using it."
- Content At Scale Review, Amelia G.
Copyleaks Plagiarism Checker is an AI-powered text analysis platform offering a range of content-checking features, including plagiarism detection, an error-free writing assistant, generative AI governance enforcement, an AI grader, and AI content detection.
"Copyleaks Plagiarism Checker is an online handy tool to quickly and easily detect if the submitted content is human-written or AI-based and copied from the internet or originally written. You can easily register, implement, and start using the tool. Multiple settings help in customizing omit settings, sources to be checked, internal databases, and frequency of scans."
- Copyleaks Plagiarism Checker Review, Pranav K.
"Limited Free Tier: While other competitors offer a free trial, the Copyleaks functionality of the free tier may be limited in terms of the number of scans allowed or the depth of analysis provided, restricting full exploration of its capabilities."
Spotting AI-generated writing is not an easy feat, specially with all the advances in technology. However, you can look for signs like overly polished tones, repetitive phrasing, or a lack of personal touch to spot AI involvement. Tools such as plagiarism checkers and AI detectors also help with AI detection. And while no approach is perfect, these methods offer a solid starting point.