Chapter 1: Recognizing AI-Written Content

I can usually identify articles produced by AI around 80–90% of the time. There isn’t a single telltale sign; rather, it's a combination of indicators. AI-generated text often feels superficial, lacking depth.

AI tools like GPT-3 utilize predictive text algorithms, sifting through online content to determine what typically follows a given prompt. The issue is that many blogs are poorly crafted. For each well-written piece from reputable sources like The New York Times, there are dozens of low-quality SEO-driven articles—often lengthy, unnecessarily so.

For instance, I tested an AI writer with a prompt asking how to wake up earlier. The response was overly verbose:

"To wake up on time, establish a morning routine. Setting an alarm is essential, but creating a consistent routine is key to waking up early."

The redundancy is evident; this could easily be summarized as "A solid morning routine enhances your ability to wake up early and boosts productivity."

Daphne Ippolito, a senior researcher at Google Brain, points out that an easy indicator of AI text is the excessive use of the word "the." I've also noticed an over-reliance on demonstrative pronouns like "this," "that," "these," and "those." Skilled writers know that many of these words can often be omitted.

Another characteristic of AI writing is its absence of typographical errors. While AI writers may produce grammatically correct text, they can generate nonsensical ideas that pass spellcheck. From a design standpoint, it makes sense—why would anyone endorse a writing tool that produces errors?

However, GPT-3 may misplace words contextually and is prone to using cliched phrases. Having spent years in the blogging realm, I can easily recognize these tired expressions:

• "If you’re like me,"
• "What if there was another way?"
• "Everyone wants to ____"
• "It doesn’t happen overnight"
• "Anyone can live their dream."

These phrases often recur within a single post.

Additionally, AI-generated content lacks seamless transitions. When a writer abruptly shifts topics without explanation, it's a clear warning sign. These systems fail to capture the subtleties of human language, resulting in emotionally flat writing. Literary devices like metaphors and similes are often absent or poorly placed.

GPT-3 tends to produce longer sentences, even if the original input is concise. The output often expands unnecessarily, akin to an uncoiling hose.

Moreover, the voice in AI writing generally remains impersonal and static. While GPT-3 can generate a first-person perspective, many prompts default to a second or third person. Experienced human writers frequently break conventional writing norms intentionally, using various perspectives to engage readers and maintain interest.

AI-generated content tends to be dull. Both AI articles and poor-quality writing share one trait: they are unappealing. Attempting to produce something halfway decent with AI is a waste of time. Rather than relying on generated content, starting with a blank page can yield far superior results.

The main takeaway? I’m not concerned about AI writers. They diminish the intrinsic motivation to write, robbing creators of the satisfaction that comes from reader validation. Authentic writers will always shine amidst the noise of automated content.

Interestingly, many people dislike reading AI-generated text without being informed of its origin. A recent example on Quora involved the introduction of a Question Prompt Generator, which was met with backlash and ultimately removed due to its perceived lack of authenticity.

In the end, writing is often about editing, and without a keen editorial eye, utilizing AI-generated text can be challenging. It’s akin to working with a pre-shaped block of clay; though the initial structure is there, it requires excessive effort to mold it into something worthwhile.

Ultimately, AI writing is simply a manifestation of machine learning, and it’s no surprise that its output resembles mechanical creation.

Subscribe to my newsletter for more insights and content.

Read all stories from Sean Kernan (and thousands of other writers on Medium). Your membership directly supports their work.

Chapter 2: Techniques for Identifying AI Writing

In this video, "How to Detect AI Written Content," you’ll learn key techniques for identifying text generated by AI, including common pitfalls and indicators.

The video "How To Detect AI Writing in 2024" explores updated strategies for discerning AI-generated content, highlighting emerging trends and tools.