AI vs Human Writing—How To Tell?

I can usually spot AI-written content in about two paragraphs. Not because it is bad.

Most of the time it is perfectly fine. But there is a familiarity to it that creeps in quickly… a sense that I have read this before even when the topic is new.

I review content every day for B2B teams. Some of it is written entirely by humans, while some of it is drafted with AI and shaped carefully afterwards. Some of it has barely been touched since it was generated, and some of it has undergone heavy human-rewriting processes. So I know what I’m talking about when it comes to AI generated content.

But the industry is changing. A year ago, the question I kept hearing was whether AI could write content at all. Now the question is different. People want to know how to tell the difference.

AI is no longer novel, but it’s fast becoming an assumed part of business’ workflows. The jury is still out on whether people care if content was AI generated or not, but IMO one thing hasn’t changed: What people very much do care about is trust, clarity, and whether the content in front of them was actually thought through by someone who understood the context.

Nonetheless, as AI improves, the tells become subtler, but they do not disappear. So, if you’re reading this and looking for an answer once and for all on how to tell apart AI vs human writing, then below, we’re going to get into exactly how. 

In this article, I want to break down the markers I see every day that separate human writing from AI writing, so you can do the same in your own content.

Is It Even Worth Telling Them Apart?

Before we begin, I do want to ask an important question of you right now. Is it even worth it to tell AI and human written content apart?

There are many situations where it is important to tell if someone has used AI: if you’re a teacher, if you’re a newspaper reporter (this I do think is important – biassed AI has no place in reporting the news), if you’ve paid above the odds for ‘human’ written content.

But if you’re a business owner looking to tell the difference between AI and human content for your SEO blog strategy, can I suggest that you might be wasting your time?

“Holly!” I hear you cry, “Google penalizes AI content!”

Well, that’s not true. Google explicitly says that AI content is not a ranking factor in and of itself, and some of the best performing blogs I’ve written have utilized AI in the content writing process. AI will not impact the SEO of your website (small but important caveat: if used correctly).

So what do I mean by correctly?

Well, Google has been very clear – write and create content that serves the reader first and foremost. Don’t simply write a blog for the sake of writing one. Instead, I like to instruct my clients to think: ‘Am I writing content that is better than the content currently ranking in number one?’

Nothing about AI or not – but frankly assessing whether or not your content is helpful, adds to the discussion, and is worth the space on your site.

What Most People Look For (& Why They’re Wrong)

If you’re still interested in how to identify AI content, then it’s important to understand what people are looking for. When people first try to identify AI writing, they usually start with assumptions that no longer hold up.

They assume AI writing is always generic. They assume perfect grammar equals automation. They assume formal language or long words must mean a machine was involved. And my least fave, they think an em-dash is the ultimate tell (Hint: it’s not!)

Those shortcuts are outdated as the LLMs themselves adapt to how people use them and the content that they create.

For example, you may have noticed a distinct dearth of em-dashes on your LinkedIn feed recently, and that’s not because suddenly you’re following a load of copywriting experts. It’s because ChatGPT has itself moved away from that as a response to all the memes.

Your LLM tool is not static, it’s constantly evolving in accordance to the way you use it, so it makes sense that we’d see its output vary, too. 

AI can now vary tone. It can soften sentences. It can mimic conversational language. It can even introduce small imperfections if prompted carefully. What you generate with ChatGPT might be different to what you create with Claude, Manus, or Perplexity. Surface-level checks simply do not work anymore.

The deeper issue is that these assumptions focus on how writing looks, not how it behaves. But if you want to tell the difference, you have to stop scanning for polish and start paying attention to structure, reasoning, and judgement.

AI’s Favourite Crutches (The Words That Give It Away)

Having said all that, there are certain words and phrases that AI reaches for reflexively. These are the ultimate tell:

• Game-changer
• Cutting-edge
• Seamless
• Ensure
• Moves the needle
• Has you covered
• We’ve got you covered
• In today’s world (or variations of this for introductions)
• Ultimate 

Then there is minimising language:

• “It isn’t just XXX, it’s YYY”
• “Not XXX, but YYY”
• Something “just” does works / gets results / helps / achieves an outcome. For instance, “Writing that just works” 
• Something “quietly” happens (for example, you’re “quietly building an email list”
• Something “actually” does another thing, such as “writing that actually works”

And generic promises:

• Not to be perfect. But to be real.
• This is content that actually works.

Human writers use these occasionally but AI uses them constantly. They function as filler that sounds helpful while avoiding specificity, as they’re key phrases AI can use every time, for any subject.

Sentence Structures AI Defaults To

The same applies to sentence structures. AI defaults to neat oppositions and symmetrical phrasing because that is how most explanatory content online is written.

• Something is not one thing. It is another thing.
• From X to Y, statement.
• Whether you are X or Y, statement.
• Need help turning X into Y?
• The question is not whether X. It is whether Y.

Finally, a new tell I’m noticing is a triple, for example:

“Logos blur together. Features get copied. Pricing changes. But tone of voice sticks.”

This is a random sentence I ask ChatGPT to generate, notice the triple short sentences here: ‘Logos blur together. Features get copied. Pricing changes.’ before the negative. 

These constructions sound authoritative but commit to very little. They feel like insight without requiring a point of view and, most importantly, they can be used and applied to any situation. A product, a service, a news report, even understanding who you are and replying to you, these semantic devices are the ultimate tells of AI. 

Beyond individual phrases, AI writing often reveals itself through repetition at the sentence level. You will see the same rhythm over and over again, and unnaturally uniform structures crop up over and over again.

For example, you might ask for a list, and all items in that list are three words long. That in itself is unlikely from a human writer, who might write a list with random variations of line length, from 3 to 7 words. Alternatively, you might notice that each new paragraph starts with a broad statement and is then followed by a general explanation, before closing with a safe takeaway. Again and again!

Structural Giveaways

These are structural tells that are more about how the content looks on a page versus what is actually being said. The good news is, there are a couple of giveaways that less eager eyed content writers miss.

The first is random extra spacing, in a list like this:

Understanding X.
Why X matters.
How to use X.

Look closely. In this list, the first line is not indented, but then the second and third have a random additional space added before they begin. That’s a sign it was copied directly from an LLM.

Other giveaways here include:

1. Random emojis when a human writer would not include them, for example in ever blog heading, or within prose. Especially: 
2. Title case headings – this in itself isn’t a big giveaway, because humans also write like this, but LLMs will default to title case.
3. American spelling in places you’re surprised to find it (By British or Australian writers, for example.)
4. Structured lists, with each item being the same length and following the same structure.
5. Sentences beginning with ‘But’ and ‘And – again, not something I am against (you’ll find my blogs littered with this!) but it certainly is an AI trope.

Of course, again none of these things are major red flags in and of themselves (except from the extra spacing issue). A British writer could use American spelling (and I indeed do, too! I have a lot of customers in the US, so switch between them in my content). But put together, they are tells that this person has used AI.

The Same Old Emojis

I asked ChatGPT for its favourite emojis, and this is what it came up with:

Here are the emojis I most commonly reach for in everyday replies, grouped so it’s easy to skim.

Friendly / conversational
? ? ? ? ? ?

Agreement / confirmation
? ✅ ✔️ ? ?

Thinking / explaining
? ? ? ?

Encouraging / supportive
✨ ? ? ? ?

Light emphasis / transitions
? ⬇️ ⬆️ ➡️

Playful / casual
? ? ?‍♀️ ?‍♂️ ?

Warnings / emphasis
⚠️ ? ❗

Celebratory
? ? ?

Emojis themselves are not a red flag, but I never see ChatGPT use ? ⚜️ ☤ ? ?

Sorry to whoever’s IG this is, but this picture is a great example of a ChatGPT caption – it’s got it all! The drama, the short sentences, the generic promises to ‘be real’, the repeated use of the triple, the overall structure on the page.

The Drama of it All! ?

If there’s one thing that we can say about AI is that he’s a drama king! Take a look at the below paragraph I prompted ChatGPT to write for this blog. 

If you can, take the time to notice a few traits:

1. How hyperbolic it is – the image of these being ‘the stock photos of writing’ is dramatic, then the use of short sentences leads us to read it quickly, as if building to a crescendo. 
2. The frequent use of the triple, in this short paragraph, the triple is used four times.
3. The sweeping statements without concrete examples, such as “AI averages instead of choosing.”
4. The structural and formatting tells, such as the extra spacing in the list.

The Stock Photo Problem

There is a reason I think of these patterns as the stock photos of writing.

They look professional at a glance. They signal authority. But they could belong to almost anyone.

The same thing happens with AI-generated structure. Three points. Three examples. Three neatly packaged sections. Subheadings that sound like they came from the same template.

Understanding X.
Why X matters.
How to use X.

This kind of symmetry feels organised, but it rarely reflects how humans actually think or work.

Human writing is messier. It cuts corners. It sometimes spends far too long on one idea because that idea mattered to the writer.

AI averages instead of choosing.

Generic Examples / Storytelling

One of the hardest things to get AI to do effectively is tell a strong story or a convincing anecdote. 

There’s no wonder that big businesses are investing in storytellers as a result, and storytelling brands are hot news at the moment.

In general, AI tells examples, stories, and little hooks that struggle to feel convincing. That’s because AI examples tend to rely on common knowledge, so they feel like they could apply to any team, any business, any situation, or they’re unbelievably unlikely and too perfect.

Real, genuine case studies or stories have awkward details that make things sound natural. Not all things go well every time. There are things that have to be explained, or steps that you took that got you the result, just the long way round. 

Human writers pull from real experience, even unintentionally. They reference specific moments and dates, plus the issues and the problems that they had as well. In addition, the stories are unique, funny to listen to, more genuine and different. 

(If there’s one way you’d like to improve your own writing or an initial first draft from AI, then this is something to look for – add in your own expertise and your own, real life experiences.)

What This Means for Your Content Strategy

If you are creating content, understanding these patterns makes you a better editor. You will spot where AI defaults slipped through and correct them intentionally.

If you are evaluating content, though, my suggestion would be to stop relying on detection tools and instead go with your gut. Those tools focus on the wrong signals and are unreliable at best, whereas that uncanny feeling is usually right.

However, I’m still standing by the idea that the real question is not whether AI was used, rather it is whether the content does its job, is valuable and genuine, and outperforms the other top content in that keyword / niche. 

If you are competing for rankings, this matters even more than simply using the right tools to create your content. Search engines are increasingly sensitive to templated thinking and content that feels averaged will struggle to perform long term.

If you need help with your content marketing strategy, don’t hesitate to reach out and book a call with the team.