How to Tell If Something Was Written by ChatGPT or Other AI

What to Know

According to a study done by Graphite, a marketing company, more than 50% of articles on the web were generated using AI, and unfortunately, we have evidence that existing AI detectors are inconsistent. The best way to detect AI is to know the most common signs and learn to spot them. It can also be helpful to use generative AI, such as ChatGPT, to see firsthand examples of what a chatbot creates based on the prompt you give it. Let's dive into how to tell if writing is AI-generated.

Jump To:

• How to Tell If Something Was Written by ChatGPT
• Why Identifying AI Will Get Harder with Time
• FAQ

How to Tell If Something Was Written by ChatGPT or Another Text Generator

With text-generating apps getting better, it can be difficult to tell if something is AI-generated. Despite being able to recognize AI images pretty well, I really struggle when it comes to detecting AI writing. AI detectors don't work very well and can incorrectly flag an essay written by a neurodivergent person or someone who's trying to optimize for SEO (search engine optimization). The surefire way to spot AI-generated text is to learn the signs. I also recommend testing AI generators to see how they work and what sort of content they produce. The more experience you have interacting with AI-generated content, the easier it will be for you to distinguish it from words written by a human.

1. AI Favors Certain Words

While this alone cannot be enough evidence to determine that something was written by AI, there are certain words that could be the first sign of trouble. While AI's "word of the day" might change over time, it's usually quite easy to look up what words are trending with AI chatbots by doing a quick Google search.

The University of Helsinki conducted a study, published in April 2025, that detected the changes in words used in essays written one year before ChatGPT was released compared to one year after. The two words now used more than 10x as often as before are "delve" and "foster," followed closely by "crucial," then "critical," "leverage," "significant," and "potential."

Other words on this list include:

• Meticulous
• Comprehend
• Tapestry
• Landscape
• Multifaceted
• Essential
• Impressive
• Robust
• Valuable
• Vita
• Boost
• Boast
• Drive
• Enhance
• Ensure
• Implement
• Leverage Provide
• Unleash

Obviously, these words aren't anything crazy that a regular person wouldn't use on occasion, especially if they are trying to sound formal while writing an academic paper. However, something that they have in common is that many of them are buzzwords. AI likes to use buzzwords because they will catch your attention and they sound more dramatic than necessary. Another thing that AI likes to do is sprinkle in the buzzwords without providing any substance. For example, "To solve the problem effectively, we must delve into the underlying data and leverage our insights to make informed decisions." It sure sounds fancy and important, but it doesn't actually offer any valuable information. It's like a lazy word salad that a student might use to buff up a paragraph to meet their essay's word count requirement.

One thing that you won't see a lot of in AI is cursing. Of course, formal writing won't include any expletives either, but creative writing might. If you read a story for adults where, any time someone stubs their toe, they yell out "ouchy" instead of "*$#^," it might be a sign of AI. It could be that the character is opposed to swearing, but there are times when the most disciplined person will let a bad word slip out—it's only human, and AI isn't, so it's PG by default.

2. Excessive Em Dashes

Every seasoned writer has used an em dash here or there, so a single em dash in a 600+ word article isn't necessarily a red flag. When AI uses em dashes, it likes to hand them out like it's Oprah—every paragraph gets one… or five! Not only does AI sprinkle them into an article as generously as a child decorating a cupcake, but it will also put them where they don't belong.

The Merriam-Webster Dictionary states that the em dash mostly functions as an indication of interruption of speech, either due to confusion or a change in direction. It can also be used instead of a comma when it's not just a normal pause, but a more dramatic hesitation. Back in English class, I was taught to use the em dash when listing phrases rather than words. I was also taught to replace parentheses with them when the text between the dashes is an important fact that should not be missed, rather than simple clarifying information.

Since AI is all about creating drama where it doesn't belong, it will use em dashes in addition to extravagant language to make something mundane sound important. Basically, if you read something that sounds exciting but ends up being a nothingburger, it was likely written by AI. To learn more about AI and the latest Apple tech features, don't forget to sign up for our free Tip of the Day newsletter.

3. Sentence Formulas & Patterns

When humans write, we can get carried away with our feelings or get distracted, which leads to our writing sounding chaotic, unintentionally humorous, or making no sense. Since AI doesn't have emotions or ADHD that can veer it off course, it treats written words like a mathematical equation. We all learn certain sentence formulas and patterns when we start writing essays for school.

For example, a standard essay should have three main points. The first should be the most important, the last can be less significant but powerful enough to leave the reader remembering it, and the middle can be the weakest while still supporting your argument. This is part of the "rule of threes," which is the widely accepted belief that groupings of three are more memorable, satisfying, and interesting (See what I did there?) The issue with using this to identify AI is that many of us will do it automatically in writing. I couldn't help but notice myself doing it over and over again throughout this entire article. So again, on its own, it doesn't scream AI, but if it's overdone or seems forced or unnatural, that can be a big clue.

Another AI-favorite is the "it is not just X, it's also Y" sentence formula, sometimes written as "it's not X. It's Y" or other variations. This can be a good way to build up the drama, which AI loves to do, usually for no reason. When I asked ChatGPT to show me an example of a sentence with this structure, it gave me, "It's not about working harder—it's about working smarter." Notice the em dash it threw in, just to be extra. For some reason, AI will often generate text that is super dramatic or even sounds artificially sassy, and it usually serves no actual purpose. Again, this is not something ridiculous that a human wouldn't normally do, but it can be a red flag.

4. Either Vagueness or AI Hallucinations

A more subtle way that AI reveals itself is by being super vague. AI tends to do one of two things when asked about specific facts: it will either be vague even though the fact is obviously true, or it will come up with a complete lie, which is known as an AI hallucination. It can be hard to recognize an AI hallucination if you don't know the answer to the question you are asking, which is often the entire point of asking it. This is why AI chatbots themselves will warn you that all answers should be fact-checked.

So if you Google the answer to double-check that it is true and realize that it was completely wrong and made up, congratulations! You've just experienced an AI hallucination. Unfortunately, lots of people will skip the fact-checking step, and since AI takes the information it gathers from the internet, the more these AI hallucinations happen, the more likely the incorrect fact might be generated again.

On the complete opposite side of the spectrum is AI vagueness. Because AI is so literal, it tends to answer in ways that can't be proven wrong. For example, I asked ChatGPT if the sky is blue. Here's how it responded, "Yes—most of the time. The sky appears blue because of a phenomenon called Rayleigh scattering, where shorter blue wavelengths of sunlight are scattered in all directions by the gases in Earth's atmosphere. That scattered blue light is what you see when you look up. But it's not always blue—at sunrise or sunset it can turn red, orange, or pink because the light passes through more of the atmosphere, scattering the shorter wavelengths and leaving the longer ones behind." As you can see, the chatbot gave us a package of vagueness, em dashes, and the rule of threes, all in just three sentences. I'm just surprised that it didn't mention tapestries as the cherry on top of this obviously AI pie.

Speaking of vagueness, if you ask AI to write an article or an essay, it will spout a lot of facts at you, but it doesn't give you a natural conclusion. I have a travel blog and have lots of people pitching their own articles for me to post. The AI articles stand out because they will list a bunch of popular places for you to visit, but they won't include a strong conclusion. It kind of feels like watching someone take a deep breath and waiting for them to exhale… and the exhale never comes.

5. Clichés Galore

A cliché is a phrase that has been used so much that it's become trite or meaningless. Since AI trains by consuming hordes of content, it makes sense that it will hear certain clichés over and over again. The trouble is, humans are nostalgic creatures and are not immune to them either, even if we laugh at ourselves for using them.

I asked ChatGPT to give me a list of clichés that are often found in AI writing, and here's what it delivered:

• "In today's fast-paced world…"
• "At the end of the day…"
• "It's important to note that…"
• "A testament to…"
• "Needless to say…"
• "Delve into…"
• "Leverage the power of…"
• "Cutting-edge technology…"
• "Game-changer…"
• "Think outside the box…"
• "Rich tapestry of…"
• "A myriad of…"
• "In conclusion…" (especially when overused or unnecessary)

Reading just one of these in an article shouldn't necessarily ring a warning bell, but the combination of clichés and AI buzzwords should definitely raise the alarm. As someone who says "it's important to note" in every other how-to article I write, I will pull out my thesaurus and think of some alternative phrases so that an AI detector doesn't accidentally confuse me with a soulless robot.

6. Consider the Who & Why

Humans always write things for a specific reason. A lot of writing online is marketing material written by a copywriter in an attempt to sell you a product or a service. Then there are websites like iPhone Life that focus on teaching the reader how to do a specific task on their smart device. Any time you read something, it is important to consider who wrote it and why it was written.

You won't necessarily recognize a name in the byline, but one red flag is when a company says an article was written by "the (company name)'s staff," because writers usually want to be acknowledged for their work and are proud to put their names on it.

If you are reading an informational article about a mundane topic, a sign of AI could be that the text is overly enthusiastic without there being anything to purchase. Similarly, if you are reading a sales pitch, you might expect a first-person experience or advice as well as a clear conclusion that a particular product or service is the be-all and end-all. Knowing that AI doesn't usually write in first person and keeps things vague can make it obvious that an advertisement was AI-generated.

When it comes to how-to articles like the ones on iPhone Life, be suspicious if the steps are vague or open-ended. Step-by-step instructions in articles are always direct and concise, teaching the reader exactly which icon to tap to achieve the task at hand, whether it's troubleshooting a problem or tweaking a setting. If the steps are incorrect, don't make sense, or don't answer the question at hand, AI may be to blame.

7. Other AI Signs

When looking for AI, it is appropriate to use a common cliché and conclude that the devil is in the details. In addition to the signs above, a tip off that you're reading AI-generated text can be that the writing is:

• Exclusively in 2nd and 3rd person voice with no personalization
• Overly complimentary to the reader, sometimes even sounding flirty
• Full of lists and bullet points when they aren't necessary or helpful
• Stuffed with commas
• Packed with run-on sentences that use a lot of words with little content
• Perfectly spelled with textbook grammar
• Not reflective of the overall topic.

Why Identifying AI Will Get Harder with Time

Something that makes it harder to recognize AI writing is that using AI and reading AI-written text changes the way the general public speaks and writes. The University of Southern California did a study, published in February 2025, that showed evidence that AI has led to a "decrease in linguistic diversity." This can happen because people might be using AI tools such as Grammarly, which might suggest changing a word or phrase you used to a synonym that's on AI's list of preferences. Also, human beings are social animals prone to subconsciously mimicking those around them, known as the "chameleon effect" (aka monkey see, monkey do). This is why you might notice yourself using the same mannerisms as your friends, or using words that you haven't before, because you've been reading them frequently. Unfortunately, this phenomenon makes it harder for us to distinguish between AI and human writing, since we are all unintentionally talking and writing more like a Large Language Model (LLM) ourselves.

So now you know how to check if something is AI-generated! As you probably noticed, I included a lot of 'AI writing' signs, some on purpose, some accidentally. Since AI detectors aren't reliable, you'll have to spot the signs yourself. Now that you know what AI detectors look for, you can be your very own detector! Do keep in mind that some humans might sound like AI even though they don't use it, especially if they are not native English speakers or if they have autistic traits. In today's world, people can be quick to point at something and call it AI, and it very well might be.

According to a study done by Graphite, the percentage of AI-written content on the internet has plateaued since May 2024. Hopefully, new rules and regulations will help create platforms for human-only writing or will make it mandatory to flag something as written by AI. Until then, you'll need to rely on your own ability to recognize it. And if you are a writer like me, articles like this can help us learn what AI-isms in writing to avoid.

Top Image credit: Primakov / Shutterstock.com