Detecting AI Content: Essential Tools and Techniques, Two Key Methods

I'm an editor for two publications on Medium. I always check if the content is AI-generated when reviewing drafts.

There are two ways to check for AI content:

AI Detection Tools

I suggest using several common tools because they're not always accurate. It's best to use multiple tools and compare results.

• GPTZero was made by Edward Tian, a Princeton student. It uses advanced language processing to tell if text is written by humans or machines. It uses deep learning and statistics to spot AI-generated text with up to 99% accuracy.
• SEO.AI is a free tool that checks if AI-generated content can be detected. A score of 1% means human-made content. A score near 100% likely means AI-generated content.

• Stealthwriter can detect content from ChatGPT, GPT-4, Bard, Bing Chat, Claude, and other AI tools.
• ZeroGPT analyzed over 10 million texts, some AI-generated and some human-written. It's 98% accurate in detecting AI text. It uses DeepAnalyse™ technology to identify the source of your text.

• Contentatscale is one of the oldest AI detectors. It existed before ChatGPT and has been improving since then.

Detecting AI Content by Specific Words

Why can AI content be detected? AI-generated content often follows certain patterns.

What are these patterns like? Based on my experience with ChatGPT, I've noticed it often uses specific words.

Let's look at words that often appear in ChatGPT's responses:

Transition Words ChatGPT Often Uses

• Therefore
• Moreover
• It can be said
• Of course
• Thus
• However
• Indeed
• Furthermore
• Nevertheless
• Despite this
• Undoubtedly

Adjectives ChatGPT Often Uses

• Skilled
• Praiseworthy
• Dynamic
• Efficient
• Evolving
• Exciting
• Exemplary
• Innovative
• Invaluable
• Powerful
• Seamless
• Synergistic
• Thought-provoking
• Transformative
• Extremely
• Vibrant
• Crucial

Nouns ChatGPT Often Uses

• Efficiency
• Innovation
• Institution
• Integration
• Implementation
• Landscape
• Optimization
• Domain
• Fabric
• Transformation
• Verbs ChatGPT Often Uses
• Align
• Enhance
• Delve
• Initiate
• Facilitate
• Maximize
• Emphasize
• Leverage

Phrases ChatGPT Often Uses

• It demonstrates...
• Finally...
• To summarize...
• It's important to note/consider...
• It's worth noting...
• On the contrary...
• This is not an exhaustive list.

Data Analysis Phrases ChatGPT Often Uses

• "Provide actionable insights through deep data analysis"
• "Drive insightful data-driven decisions"
• "Leverage data-driven insights"
• "Extract meaningful insights from complex datasets"

Other Signs of ChatGPT-Generated Content

• Overly complex sentence structures
• Unusually formal tone in casual texts, or too casual in formal texts
• Unnecessary wordiness
• Vague statements

Final Thoughts

Why does ChatGPT's text seem poor? ChatGPT gives an "average" answer to everyone. Its output is general and vague.

ChatGPT doesn't use the best words for each situation. It uses the most common words.

There's no perfect way to detect AI writing today, even with tools like GPTZero. AI can always create text that avoids these tools.

As ChatGPT improves, avoiding the words and phrases listed here might make your text seem more real.

Using specific, meaningful words and examples can make your text sound more human.

For example, you can ask ChatGPT to write a draft. Then work with ChatGPT to edit the text to sound more real.