HOW TO SPOT AI-GENERATED TEXT: A GUIDE FOR OLDER ADULTS
The telltale signs of AI writing, why it's getting harder to detect, and when it matters.
The two columns below show two versions of text that were created to answer the question "How to spot AI-Generated text?". The left column was pasted verbatim from the response given by Claude while the right column contains the same text after having been reviewed and edited by a human.
Pure AI Output
AI-generated text is everywhere now. Some of it is labelled but most of it isn't. While AI has gotten very good at mimicking human writing, it hasn't perfected it, and there are still tells - patterns that give it away.
Knowing how to spot AI writing matters because you should know when you're reading something a human wrote versus something a machine predicted, not because all AI text is bad.
Common patterns in AI writing
AI writing has a distinctive flavour, and once you've read enough of it, you start recognising certain patterns. Here are the most common giveaways that can help you identify AI-generated content.
Overly formal tone
It's worth noting that AI tends toward formality, even when it's trying to be casual. It often uses phrases like "It's worth noting that..." or "It's important to remember..." or "In today's world..." more frequently than humans typically do. Additionally, even when you ask it to write casually, it often sounds like someone trying to sound casual rather than actually being casual.
Repetitive structure
AI generally loves lists and bullet points. It frequently loves "Firstly... Secondly... Thirdly..." structures. Moreover, it loves starting paragraphs with "Additionally" or "Moreover" or "Furthermore." While humans do this too, they don't do it as reliably or as mechanically as AI does.
Hedging language
AI tends to be quite cautious in its assertions. It may say "may," "might," "could," "often," and "generally" more than humans typically do. This is partly because it was fine-tuned to avoid making overly confident claims since it can hallucinate, and partly because hedging language appears frequently in its training data. The result is text that often feels somewhat non-committal, even when it's trying to make a point.
Tell-tale phrases
It's important to understand that certain phrases can be considered AI signatures. "It's worth noting..." is perhaps the most significant one. "In conclusion..." at the end of something that isn't a formal essay is another indicator. "Delve into" is a phrase humans rarely use, but AI loves it. Similarly, "landscape" in the metaphorical sense, like "the digital landscape," appears frequently. "Robust" is often used to describe anything from arguments to systems. While none of these phrases definitively proves AI authorship (we use them too), when you see several of them in close proximity, it may be a clue.
Smooth but empty
AI is generally very good at producing text that flows well and sounds intelligent while saying almost nothing substantial. It can write multiple paragraphs that are grammatically perfect, logically structured, and yet completely devoid of original insight. It's essentially filler dressed up as meaningful content.
Human-Edited Version
AI-generated text appears almost everywhere now and while some of it is labelled, most of it isn't. AI has evolved to become very good at mimicking human writing, but it hasn't perfected it yet, and there are still tells, patterns and idiosyncrasies that give it away.
Knowing how to spot AI writing matters not because all AI text is bad but because you ought to know when you're reading something a human wrote rather than something a machine predicted.
Common patterns in AI writing
AI writing has nuances, styles, patterns and poor fluency and once you've read enough of it, you start recognising it. Here are the most common giveaways.
Overly formal tone
AI loves idioms. Turns of phrase like "It's important to remember", "In today's world", "It's worth noting" appear in practically every piece of AI writing I've ever read, and often multiple times. We don't talk like that unless we're writing academic papers or trying to sound important. Even when you ask AI to write casually, it sounds like someone trying to sound casual rather than actually being casual.
Repetitive structure
AI adores structures like lists, bullets and numbered bullets. You will see things like "Firstly, secondly, thirdly" everywhere. It seems obsessed with transition words like "Additionally", "Moreover", and "Furthermore", using them to join the many short sentences it produces in an effort to lengthen them and make them seem 'normal'. Yes, you and I might do this occasionally, but AI uses them like punctuation where seemingly every paragraph starts with one!
Hedging language
AI qualifies everything - "may", "might", "could", "often", "generally". It rarely commits to anything because it was trained to be cautious (probably because it makes so much up). The result reads like someone who's permanently unsure of themselves.
Tell-tale phrases
Certain phrases are dead giveaways. "Delve into" - nobody says this in real life. "The digital landscape" - corporate nonsense. "Robust" applied to everything from arguments to software. Any single one of these might be coincidence, but see three in a row and you're reading AI. Humans use these words too, but not with AI's mechanical frequency.
Smooth but empty
This is the worst tell and the hardest to describe. AI text flows beautifully, sounds professional, and says absolutely nothing a lot of the time. Perfect-looking grammar, logical structure but yet zero insight. It's protraced filler that looks like content and while humans might make messier mistakes because we ramble, contradict ourselves or use clunky phrasing, AI polishes all that away and leaves you with competent blandness.
I am not saying I am perfect, or better than AI, but what I am saying is that AI, when it comes to writing prose, benefits enormously from human input. And interestingly, the more I used Claude to provide me with the text for my "explanations", the more it learnt my style. Its grammar improved, its sentences got longer and the American spellings were all but eradicated.
Detection tools versus AI getting better
There are tools designed to detect AI-generated text by analysing patterns, sentence structure, word choice, and statistical markers. Some are better than others, but none are perfect.
The problem is that AI keeps getting better at mimicking human writing, and as detection tools improve, AI developers train their models to avoid the tells that detection tools look for. It's an arms race, and detection is losing.
The other problem is false positives where detection tools sometimes flag human writing as AI, especially if it's formal, well-structured, or written by someone whose first language isn't English. This has already caused issues in schools and workplaces where people have been wrongly accused of using AI. Cant say that it has ever happened to me though!
Detection tools can be useful, but they're not reliable enough to be definitive. You can't prove something was written by AI just because a tool says so, and you can't prove it wasn't just because a tool gives it a pass.
Why this matters
Students are using AI to write essays where some do it badly and get caught immediately because the AI hallucinates sources or the style doesn't match their previous work. Others do it well enough that it's hard to tell. This creates problems for teachers and for education more broadly because if students can outsource their thinking to AI, what are they actually learning?
AI makes it cheap and easy to flood the internet with plausible-sounding nonsense including fake news articles, fabricated reviews, and generated blog posts that exist only to game search engines. The volume of low-quality content is exploding, and much of it is AI-generated.
Even when AI-generated content isn't actively harmful, it's often just there - filling space and clogging up search results with bland, generic text that doesn't add value. Some people call this "slop", which is content created not because anyone wanted to say something but because content itself has value through ad revenue and SEO ranking. AI makes slop cheap to produce at scale.
What you can do to verify whether something was AI-written
Read the text carefully and look for the tells described above - overly formal tone, repetitive structure, hedging language, tell-tale phrases, and that smooth but empty feeling. None of these are proof, but they're clues.
Check the specifics because AI is good at generalities and bad at details. If a piece of writing includes verifiable facts like dates, names, or citations, check them. If they're accurate, it's more likely to be human or human-edited, and if they're invented, it's AI hallucinating.
Compare it to the author's usual style because if you know what someone's writing normally sounds like and this doesn't match, that's a red flag. AI tends to flatten style and make everyone sound a bit more like everyone else.
Tools like GPTZero or Originality.ai can help, but don't treat their verdict as final since they're better at spotting obvious AI text than subtle AI text and they make mistakes in both directions.
If you're in a position to ask whether something was AI-generated, just ask. Some people will lie, but many won't.
The bigger picture
We're at a point where AI can produce text that's good enough to fool most people most of the time, which is new and has consequences.
It means you can't assume that text was written by a human just because it's coherent, and it means "written by" is becoming blurry - is something AI-written if a human edited it afterward, like here? Or what if the human wrote the outline and AI filled it in?
It also means the internet is going to get worse before it gets better with more noise, more slop, and more text that exists not because anyone had something to say but because generating text is now trivially easy.
You can't stop this, but you can be aware of it and get better at spotting the tells. AI writing has a signature, so learn to recognise it.