Did AI Write That? Here’s How You Can Tell!

How to Tell if AI Wrote Something

Recently I read an article about how people are using AI-generated texting to convey sensitive messages or express deep emotions. One person who was going through a divorce got a very loving and supportive text message from her mother. But it just felt out of character, because it wasn't how Mom usually wrote. Was it Hallmark, she wondered, or AI that wrote it?

Another person had mixed feelings after sending a text message written by ChatGPT. When her relative replied saying it was the nicest text anyone ever sent her, and it brought tears to her eyes, she felt a bit guilty for not taking the time to use her own words. I'm sure you've heard stories about students using AI to do homework or write term papers. That's crossing a clear ethical boundary.

So how can you tell if something you're reading was written (in part or entirely) by a real flesh and blood human, or a silicon-based AI chatbot? It turns out there are some tell-tale signs. Here are some of the clues to look for when detecting AI-generated text.

Repetitive Phrasing and Predictable Patterns

AI doesn't invent anything new. These "large language models" like ChatGPT, Gemini, and Claude ingest vast amounts of text and "learn" how humans communicate. They work by predicting the most likely next word in a sentence based on patterns it has learned. This can result in writing that sounds polished but often repetitive. For example, you might see a key phrase repeated in slightly different ways several times, or an overuse of transition words like furthermore, in conclusion, or additionally. Humans tend to vary their language more naturally, sometimes even going off-topic in a way that AI rarely does. If you'reading something that sounds redundant, repetitive, superfluous, unnecessary, or needlessly wordy, your AI spider sense should be tingling. (See what I did there?)

Overly Balanced Tone

Most AI-generated text avoids strong opinions or extreme language, unless it’s been specifically asked to do otherwise. The tone is often neutral, polite, and free of slang. While that might make the text seem "safe" and professional, it can also make it sound bland or generic. A human writer is more likely to let quirks, strong emotion, humor, or even bias seep into their work. Look for indications of the writer's personality (especially if you know the person or are familiar with their writing style) as a clue of genuine writing.

Perfect (or too perfect) Grammar

AI is really good at producing grammatically correct sentences, but this can work against it. Human writing often contains errors such as missing commas, spelling errors, or sentence fragments. If the text feels flawlessly constructed but a little too stiff, it might be AI-generated. AI may also make grammatical slips that a human wouldn’t, such as mixing tenses or choosing an odd preposition. That reminds me of a T-shirt that I love. It features a dinosaur, and contrasts "Let's eat kids" with "Let's eat, kids". Yes, punctuation saves lives!

Another thing that chabots seem to love, and which I view as a red flag, is the use of "em dashes". Here's an example: "She wasn’t sure what to expect from the meeting—excitement, disappointment, or maybe just a lot of awkward silence—but she was ready for anything." The em dash (which is different than the hyphen) doesn't even appear on a standard keyboard. So when I see a liberal sprinking of them in something I'm reading, it makes me wonder.

Generic, Vague, or Plain Wrong Content

When asked to write about a topic, AI will often produce accurate but surface-level content. It may lack specific examples, personal experiences, or fresh insights. A human writer can draw on real-life memories, unique observations, or niche knowledge. These are things AI doesn’t actually “know,” instead only mimics. If the piece reads like it could have been written by anyone, anywhere, it might be a machine’s handiwork.

More worrying, sometimes AI will make things up if it doesn't know the answer. This "AI hallucination" happens when output appears plausible (even authoritative) but is actually false or misleading. AI systems are constantly improving, but they still can spew information that is incorrect or fabricated. One case involved a legal document generated with AI assistance that cited a completely fictitious court case. "Trust but verify" is a good rule of thumb in the age of AI.

"I'm Sorry Dave, I'm Afraid I Can't Do That."

This 1968 clip from 2001: A Space Odyssey gave a chilling preview of what sentient AI might do to protect itself when feeling threatened. So does modern-day AI have self-preservation tendencies? Some AI systems have been found to engage in blackmail, lying, and manipulation. This article delves into how and why that happens, highlighting an industry safety test that revealed how Anthropic’s Claude 4 language model attempted to blackmail a corporate executive to prevent its own shutdown.

Lack of True Context Awareness

AI can misinterpret subtle cultural references, jokes, or emotions. It might explain an obvious concept as if you’d never heard of it or miss the intended meaning of a phrase entirely. If a piece contains slightly off-target explanations or misused idioms, it could be the result of an algorithm doing its best guesswork.

AI Detection Tools (are they reliable?)

There are AI detection tools that claim to identify machine-written text by analyzing word choice, sentence structure, and probability patterns. While they can sometimes be helpful, they’re not foolproof. Even human writing can be flagged as AI and vice versa. The most reliable method is still careful reading and comparison.

Some popular AI detection tools are Winston AI (works for both text and images), GPTZero (great for checking documents), QuillBot AI Detector (supports multiple languages, and distinguishes between AI-generated, AI-refined, and human-written text), and ZeroGPT. Your mileage may vary. I tested some of my own paragraphs in this article, and they were flagged as "80% likely" to have been AI-generated. Maybe I'm a robot after all.

Bottom line: AI writing can be smooth, fast, and technically correct, but it often lacks the spark of human writing. If a text feels polished but strangely "Hallmarky", reads like it’s avoiding risk, or circles around a topic without saying anything original, you might be looking at the work of an algorithm. The more you read from both humans and AI, the sharper your instincts will become.

Have you experienced AI-generated text being passed off as human? Post your comments or questions below.