The short version: AI writing tends to be smooth, evenly paced, and confident in a way that flattens out the bumps that make writing feel human, while human writing is uneven, opinionated, and full of the specific detail nobody would bother inventing. The difference isn’t really about grammar or vocabulary, it’s about texture, and once you know what to look for you’ll see it everywhere, including in your own drafts.
What I noticed after reading hundreds of client drafts
I read a lot of blog drafts for a living. Not dozens a year, dozens a month, from clients across the UK, US and Israel who all want help sounding like themselves online. Somewhere in 2023 the drafts started changing shape. Not overnight, but within about six months I could open a Google Doc and know, before I’d read a full paragraph, whether a human had typed it or a chatbot had generated it.
Here’s the moment it clicked for me. A client, a financial advisor in Manchester, sent me a “first draft” of a piece about retirement planning. It opened with: “In today’s fast-paced world, planning for retirement has never been more important.” Every sentence after that was correct, balanced, and said absolutely nothing he would say out loud in a meeting. I asked him to just talk to me for ten minutes about what worries his clients. He mentioned a woman who cried in his office because she’d delayed her pension by eight years out of embarrassment. That sentence went in the piece. The AI-generated opener didn’t survive.
That’s the whole difference in one story. AI writing gives you the average of everything ever written on a topic. Human writing gives you the one thing that happened to one person, on one Tuesday, that nobody else would think to include.
Side by side: what shows up on the page
Strip away the debate about robots taking jobs and just look at sentences. Here’s what I see consistently when I put an AI draft next to a human one, line by line.
- Sentence length: AI writing sits in a narrow band, usually 15 to 22 words per sentence, over and over. Human writing swings wildly, a four-word sentence next to a forty-word one, because real people get excited or tired mid-thought.
- Openers: AI leans on “In today’s world,” “When it comes to,” “It’s important to note that.” Humans open with the thing that happened, the number that surprised them, or a flat statement of opinion.
- Specificity: AI says “many businesses report improved results.” A human says “Sarah’s bakery in Leeds went from 40 to 190 Instagram followers in six weeks after she started posting behind-the-scenes videos.”
- Hedging: AI writing hedges constantly, “it can be argued,” “this may vary,” “in many cases.” Real writers commit to a position and let readers disagree with them.
- Rhythm: AI content reads like it was written by someone with perfect posture the entire time. No stumbles, no asides, no “look, I’ll be honest with you here.”
- Lists inside lists: AI tends to over-structure, dropping in a bulleted list every three paragraphs whether the content needs it or not.
I wrote a longer breakdown of the exact tells, the phrases and rhythms, in a piece on the patterns that give AI writing away, and once you’ve trained your eye on three or four of them, you cannot unsee them.
What human writing does that AI still doesn’t
People assume the gap is about quality. It isn’t, not really. A well-prompted AI can produce a technically stronger paragraph than a tired human on a Friday afternoon. The gap is about stakes. Human writing carries the risk of being wrong, of embarrassing yourself, of taking a side that costs you a client. AI writing has none of that risk built in, so it never quite commits.
A few things I still only see from human writers:
- Correcting themselves mid-paragraph (“, that’s not quite right, what I mean is…”)
- Bringing up a memory that has nothing to do with the topic but somehow illustrates it perfectly
- Naming a real person, a real place, a real pound or dollar figure, without softening it into “a significant amount”
- Being rude about something, plainly, without three qualifying clauses first
- Getting tired near the end of a long piece and it showing in shorter, blunter sentences
That last one is small but telling. Human energy dips and rises across a piece of writing. AI output has almost no fatigue curve, it’s just as polished in paragraph one as paragraph twenty. Real writing sags a bit near the end, and readers trust that sag more than they’d admit.
The bit most people writing about this topic skip
Here’s where I’ll say something that makes some writers uncomfortable. A huge amount of “human writing” online right now is not human, it’s AI output with a light human edit, and the person editing it will happily tell you they “wrote every word” because they changed a sentence and added their name to the byline. I’ve done this myself, on days when a deadline mattered more than my pride. The line between “I used AI to help me write this” and “AI wrote this and I tidied it” is much blurrier in practice than any of the LinkedIn purists want to admit, including me some weeks.
The honest version of this is that the tool itself was never really the issue. What matters is whether a real person’s actual experience, opinion, or mistake made it into the final piece. You can write every single word yourself and still produce something as hollow as a bad AI draft if you never say anything you’d stake your name on. And you can run a first draft through AI, then rip out the generic openers and stuff it full of your own specifics, and end up with something more human than half the “authentic” content on the internet. I’ve watched writers who’d never touch AI on principle produce flatter, more generic prose than clients who use it as a first-draft engine and then heavily rewrite. The tool isn’t the tell. The absence of a real person taking a real risk is the tell.
A simple way to test any piece of writing
When I’m not sure whether a draft in front of me leans AI or human, I run it through five questions:
- Could this sentence have been written about literally any company in this industry? If yes, it’s probably AI or AI-adjacent.
- Is there a number, name, date or place in here that couldn’t be invented? Real writing tends to anchor itself to something checkable.
- Does the writer take a side anywhere? Not “some might argue,” an actual position they’d defend in an argument.
- Would this sentence embarrass the writer if it turned out to be wrong? AI rarely writes anything with that kind of exposure.
- Does the piece get tired, funnier, angrier or looser as it goes on? Sustained perfect evenness is a tell in itself.
Run any 800-word blog post through those five and you’ll get a clear read within about two minutes. I’ve done it with client content, competitor content, and my own drafts when I’ve had a rough week and want to check I haven’t gone soft.
Why this matters for your marketing, not just your ego
This isn’t a purity argument, it’s a commercial one. Google and the large language models that now sit between your business and your customer both reward the same thing: content that says something specific enough to be worth citing. Generic AI phrasing doesn’t get quoted by ChatGPT or Perplexity, and it doesn’t get shared by a real person either, because there’s nothing in it to point at and say “look at this.”
If you’re producing content for a business, whether that’s blog posts, LinkedIn updates, or the copy on a scheduling page, the instinct to smooth everything into safe, average sentences is the exact thing working against you. I see it in social content constantly, which is part of why the mechanics of what you’re posting and where matter as much as the words themselves, whether that’s picking the right app for scheduling posts or getting the booking flow right on an all-in-one booking and website tool. The channel doesn’t fix flat writing. Only specifics do.
There’s also a resource cost to all this that most content debates ignore entirely. Every AI-generated draft you throw away and rewrite from scratch is compute spent for nothing, which is one reason I’ve written before about what AI tools cost in water and energy for business use. If you’re going to use AI as a first draft, use the output, don’t discard three versions before you get to the human bit. That’s better for your time and it’s better for the planet, and honestly it’s a stronger habit either way.
What I do now, day to day
I use AI constantly, for research, for first drafts, for shaking loose a structure when I’m stuck. What I don’t do is publish the first version. My process is roughly: draft with AI or without it, doesn’t matter, then go through and ask “what did I see, hear, or experience this week that proves this point,” and force at least two of those into the piece by hand. Then I read it out loud, because AI writing reads fine silently and falls apart the second you try to say it in your own voice.
If you’re building a writing career right now rather than just marketing content, the same test applies when you’re hunting for paid work. Editors and clients can smell generic pitches from three inches away, and if you want to stand out, the advice in how to find genuine freelance writing jobs online comes down to the same thing this whole post comes down to: specifics get hired, averages get ignored.
For businesses trying to build this into how they operate, rather than a one-off blog experiment, this is often exactly the kind of thing worth getting outside eyes on, which is part of why I do hands-on AI implementation coaching with teams who want AI in their workflow without losing the human bits that make the content worth reading in the first place.
Frequently asked questions
Can you always tell if something was written by AI?
No. A skilled writer editing AI output heavily, or someone deliberately varying sentence length and adding specific detail, can make it indistinguishable from fully human writing. What you can reliably spot is unedited, first-draft AI output, because it clusters around the same sentence lengths, hedging phrases, and generic openers every time.
Is AI writing always worse than human writing?
Not on a sentence-by-sentence basis, no. AI can produce grammatically clean, well-organised paragraphs faster than most humans. Where it consistently loses is specificity and stakes, it rarely includes a real name, a real number, or an opinion the writer would defend in an argument.
What’s the fastest way to make AI-drafted content sound more human?
Add one specific detail per section that could not be invented, a real name, date, place, or number, and cut any sentence starting with “In today’s world” or “It’s important to note.” Those two changes alone fix most of the flatness.
Does it matter for SEO whether content is AI-written or human-written?
Search engines and AI answer tools don’t penalise content for being AI-assisted, but they do reward specificity, and generic AI phrasing rarely contains the kind of concrete detail that gets cited or ranked well. The label matters less than whether the content says something worth quoting.
