How to make ChatGPT sound like you, or any other celebrity (Updated for 2026)

Three years ago, we tried to teach ChatGPT to write like us.

Not like a generic brand voice, or some vaguely inspirational celebrity, but like actual humans with specific habits, preferences, and an unfortunate fondness for wordplay.

At the time, ChatGPT was impressive but awkward. It tried hard. With enough prompting, feeding, nudging, and cajoling, it would occasionally produce a sentence that might plausibly have been written by a conscious being. But mostly it sounded… not quite right. Polite, competent, and unmistakably AI.

We documented the whole experiment, false starts and all, and concluded, somewhat smugly, that the copywriter’s job was safe for now.

Fast forward to 2026, and things look very different. AI writing tools are faster, more flexible, and genuinely better at adapting to context. They can produce copy that would pass a quick human skim without raising too many eyebrows. What hasn’t changed is the nature of the challenge.

Writing like a real person isn’t just about hitting the right tone or following stylistic rules. It’s about the small, often unconscious decisions that make writing feel like it came from ‘someone’, not ‘something’.

So can AI write like a real person today?

In this updated version of our experiment, we’re revisiting what we got right, what aged badly, and how we’d approach training AI to sound human in 2026—without losing the very thing that makes human writing worth reading.

What we tried in 2023 (and what we learned the hard way)

Tone of voice is a slippery thing.

It’s not just the words you use, but how often you use them, when you decide not to, and all the tiny instinctive choices that shape a sentence as it’s being written. Two people can follow the same style guide and still sound nothing alike.

Back in 2023, our plan seemed simple: feed ChatGPT examples of our writing, get it to analyse the style, give that style a name, and see if it could reproduce it.

Sometimes, it almost did.

More often, the output landed in that frustrating middle ground where you could see what it was aiming for, but it wasn’t quite polished enough to ship.

The more we asked it to sound “friendly” or “understated” or “funny,” the clearer it became how vague those instructions actually were. Asking for more humour didn’t help when we couldn’t explain what ‘kind’ of humour, or when to dial it back.

We got ChatGPT closer to sounding like us, just not close enough. And definitely not consistently enough to hand over the keys. The whole thing was useful, sometimes fascinating, frequently exasperating.

Now, in 2026, this reads like ancient history. But it’s a useful reminder of how far we’ve come.

How we’d train ChatGPT to sound like us today

If we were starting this experiment again today, we’d do a few things very differently. Not because AI has magically learned what tone of voice means, but because we’ve learned where it helps, and where it doesn’t.

1. Just ask (no ritual required)

Gone are the days of elaborate instruction ceremonies. Now you can just say something like: "Here are some examples of my writing. Learn my style and write like me." The AI gets it. It's almost suspiciously easy, like discovering you've been taking the scenic route to work for years when there was a shortcut all along.

2. Show, don't tell (but less of it)

You still need to feed it examples, but you don't need to sacrifice your entire content archive. A handful of diverse well-chosen pieces—a punchy social post, a longer article, maybe that email where you were particularly on fire—and you're good. Quality over quantity.

3. Define boundaries, not just vibes

Asking for a tone to be “friendly” or “understated” only gets you so far. What actually helps is spelling out the edges. Be explicit about:

- what you avoid

- what you rarely do

- what you only do in specific situations

For instance: “Never use exclamation marks in body copy. Rarely use questions as headlines. Only break into fragments when making a deliberate point.” Those constraints are often more revealing than positive instructions, and they help prevent the writing from sliding into default AI cheerfulness.

4. Get specific feedback (and actually use it)

Ask the AI what makes your style distinct. Then react to its drafts the way an editor would: pointing out what works, what doesn’t, and why. You could ask: "What makes this style distinct? What would you change to make it more like this?" The AI will give you concrete observations about sentence rhythm, humour style, even quirks like your suspicious fondness for em dashes. Listen to it. Argue with it. Refine together.

5. Skip the naming ceremony (or don't, it's still fun)

You can still name your style "TOV 1" if you want the nostalgia, but modern AI remembers context throughout conversations without needing formal labels. It's like it actually has a memory now. Revolutionary stuff.

6. Test it immediately

Ask it to write something in your style right away. But here's the thing: it'll probably be... pretty decent on the first try? Not perfect, but recognizably in the ballpark. You might even do a double-take. This is where 2026 gets a bit uncanny.

7. Refine with actual nuance

Instead of "be funnier", you can say things like: "This is close, but I'd never use that phrase. I'd be more self-deprecating here. And I'd cut the last sentence entirely, I like ending on something slightly unexpected." The AI actually responds to this kind of nuanced feedback.

8. Iterate without losing your mind

The back-and-forth is faster, smarter, and less soul-crushing. You can even upload reference materials, point to specific phrases you love or hate, and watch as the AI adjusts in real-time. Within a few rounds, you’ll have something that might sound genuinely like you. Maybe even better than you on a Monday morning.

9. Accept that it will never quite finish the job

The biggest shift isn’t technical, but mental. Don’t expect ChatGPT to be the voice. Expect it to get you close enough to make the final decisions yourself. It speeds things up, offers alternatives, helps us get unstuck. But the last pass — the one that decides what stays and what goes — still needs a human hand.

Will AI replace the copywriter?

Short answer: no. Longer answer: not in the way people think.

AI has gotten scarily good at producing language that looks right. It follows instructions, mirrors tone, respects your constraints, ticks all the boxes. The results are all over the Internet: marketing content that's coherent, confident, technically fine... and weirdly bland. Nothing's actually wrong with it. It just lacks personality.

Here's what AI still can't nail: the unspecifiable stuff. The slightly odd phrasing you keep because it makes you smile. The joke you decide not to make. The sentence you hack down because it's gotten too impressed with itself. The bit where you ignore your own style guide because breaking the rule just works better.

AI can get you 85% of the way there, maybe 90% if the stars align. But that final stretch? That's human territory. Not because AI is bad at its job, but because good writing isn't just linguistic tricks. It's judgment calls that don't follow rules. It's knowing when to be clear instead of clever, and when your carefully crafted sentence needs to die for the greater good. Those decisions don't come from data, they come from experience, intuition, and understanding what you're actually trying to achieve.

Used well, AI makes copywriters faster, more consistent, and more focused on the interesting parts of the job. Used badly, it turns everything into the same smooth, pleasant, competent mush.

So no, copywriters aren't packing their bags for that beachside villa with the piña colada and laptop setup. Not yet, anyway. Back in 2023, we thought the gap was unbridgeable. Now it's just... a gap. A meaningful one, but smaller. The job is shifting: less time typing, more time thinking. Less cranking out words, more deciding which words deserve to exist. And a lot more responsibility for making sure what goes out still sounds like an actual person wrote it.

AI can help you sound like yourself. But you still have to know who that is.

TL;DR

Back in 2023, teaching ChatGPT to write like us was painful. Lots of prompting, vague results, and output that was almost right but not quite human.

In 2026, it's easier: feed it a few good examples, set clear boundaries (what you don't do matters more than vibes), give specific feedback, and iterate. AI will get you 85–90% there, fast.

But that last 10%? Still needs a human. AI can't make the judgment calls—the weird phrasing you keep, the joke you cut, the sentence you kill because it's too pleased with itself.

The copywriter's job isn't going anywhere. It's just shifting from typing to thinking, producing to directing. AI makes you faster. You make it sound like an actual person wrote it.

The bottom line: AI can help you sound like yourself. But you still have to know who that is.

Illustration: Midjourney