AI Writes for Me. But Can It Do It Better Than I Can?

It's two in the morning. Outside my studio window in Caserta, the silence is broken only by the distant hum of a car on the bypass. On the desk, a long-cold espresso cup has left a dark ring on the wood — a small mark of a creative battle that refuses to end. I needed to finish a piece for an important client, one of those who's built an empire on craftsmanship and reads every word as though he's weighing gold. I had the structure, I had the data, but I was missing that spark — that flash of something that turns a line of text into a feeling.

In a moment of tiredness mixed with curiosity, I opened ChatGPT. I typed a quick prompt, almost guiltily, asking it to draft something on the value of time in slow production. In three seconds flat, the screen filled up. The text was correct. More than correct — it was perfect. The syntax was impeccable, the rhythm balanced, not a single comma out of place. And yet, as I read it, I felt a sudden chill. That wasn't me.

There were none of my professional scars in it. None of that smell of dust and workshop I'd breathed in just the day before, talking with the client. It was a text that tasted of wet tarmac: glossy, reflective, utterly flat. Anyone could have written it, which means nobody had.

That was the night I realised that the question "does AI write well?" is the wrong question. Of course it writes well — it writes better than the human average on a grammatical level. The real question, the one that should be keeping us up at night, is: "does AI know how to write like me?" And the answer, for purely structural reasons, is no.

Artificial intelligence has never lived in London in the pouring rain, wondering whether it made any sense to stay or go back to Naples. It doesn't know what it feels like when a Campanian entrepreneur shakes your hand and you sense the weight of three generations in that grip. AI has no body, no senses, no emotional memory.

That said, this isn't a Luddite manifesto. I'm not about to tell you to switch off your computer and go back to a typewriter. I use AI every single day. It's become my fastest assistant, my favourite sparring partner, my tireless researcher. But I've had to learn not to become its ghost. In this article, I want to take you behind the scenes of my method: how to integrate these tools without selling off your identity, how to turn an algorithm into an ally, and — crucially — when it's time to close the chat and pick up the keyboard again to put some soul back into it.

Because marketing, when it comes down to it, is a matter between human beings. And human beings recognise each other by their imperfections, not by the statistical perfection of pixels.

The uncomfortable truth nobody tells you about AI and writing

It seems the world has split into two camps: those who cry apocalypse, convinced that copywriters will end up flogging encyclopaedias door to door, and those who think a three-word prompt is all you need to become the next Seth Godin. Both are wrong. The truth is far more subtle: the problem isn't that AI will replace good copywriters. The problem is that AI is enormously amplifying the mediocre ones, turning the entire landscape of communication into an indistinguishable sea of texts that are "good enough" but utterly lifeless.

If you can't write, AI won't save you

We need to stop thinking of artificial intelligence as a magic wand. If you haven't got a clear idea, if you don't know your audience, if you haven't got an original point of view on the world, AI will give you back exactly what you gave it: nothing dressed up for a night out. AI produces form — it doesn't produce thought. Imagine ordering a bespoke suit online by providing only the height and weight of a mannequin: you'll get a perfect garment, but one that will never fit anyone naturally.

If your writing is empty of substance, ChatGPT will simply make that emptiness more elegant. It's the triumph of the box over the gift. If you're a copywriter who does nothing but summarise things you've found on Google, then yes, you've got reason to worry. But if you're a writer who challenges certainties, who digs into the mud of real problems, AI just becomes a faster way to dig.

If you can write, AI can multiply you

The real difference between those who use these tools well and those who get crushed by them isn't technical skill — it's conceptual ability. Those who already know how to write use AI as scaffolding. They know where they want to get to, they understand the weight of words, and they use the algorithm to generate variations, explore synonyms, and break through writer's block. Those who can't write use AI as a substitute for their brain. The result? Texts that all seem written by the same person, in the same calm, neutral, and infuriatingly reassuring tone.

Writing well means making choices. AI doesn't choose — AI calculates probabilities. Your job is to break those probabilities with the unpredictability of human genius.

The real problem: we're confusing tool and author

Think of a pizzeria in Naples. The wood-fired oven is an extraordinary piece of technology, the result of centuries of refinement. But the oven doesn't make the pizza — it bakes it. The pizzaiolo is the author: he's the one who feels the dough under his fingers, who understands whether the humidity in the air calls for a pinch more flour, who decides the exact moment to pull the margherita out. When the pizzaiolo decides to hand control to the oven, he gets something cooked — but nothing that has anything Neapolitan about it.

The same thing is happening in writing. We're delegating sensibility to the oven. We can use the most advanced oven in the world, but if we take the pizzaiolo's hands off the dough, what we serve our readers is nothing more than a flavourless industrial product.

Two markets, two different speeds

I've had the chance to observe this from both sides of the Channel. In London, where I work with clients through Insight ADV, the adoption of AI in writing is already well-established practice: British copywriters integrate it into their workflow naturally, almost with technical indifference. In Italy — and in Campania in particular — we're at a different stage. There's still a phase of discovery mixed with suspicion, the same suspicion Berto brings into the office when he talks about "all this digital stuff".

The irony is that this lag gives us a concrete advantage: we can learn from mistakes already made elsewhere. The best English copywriters I know aren't the ones who use AI the most — they're the ones who worked out sooner when not to use it.

What AI actually does when it writes: explained without the engineering degree

To use AI without being used by it, you need to understand what's going on under the bonnet. You don't need a computer science degree, but you do need awareness. Because if you understand how an algorithm thinks, you'll also understand why its texts sometimes seem so washed-out.

The prediction mechanism

A language model like ChatGPT doesn't "know" anything. It has no understanding of the world, no opinions of its own — only those that someone else stitched into it during training. It is, at its core, a gigantic probability calculator. When you ask it to write a text, it isn't "thinking" about what to say: it's trying to guess which word is statistically most likely to follow the previous one, based on the billions of texts it was trained on.

If you type "The sky is...", the AI knows that "blue" has a very high probability, "cloudy" a middling one, and "made of jam" an extremely low one. That's its superpower — but it's also its fatal limitation: AI produces what's expected. Good writing, though, is almost always made of the unexpected. It's the breaking of the pattern, not its confirmation, that creates value.

Why AI can't have your voice (for now)

Your voice isn't just a list of adjectives or a choice of punctuation. Your voice is the sediment of every single experience you've lived through. AI doesn't know what it's like to present a quote to an entrepreneur in Caserta who eyes you with suspicion because "my nephew does it for two hundred euros". It hasn't lain awake the night before an important deadline, hasn't felt its heart leap at a genuine compliment from a reader.

All of those lived experiences enter your writing invisibly: a dialect word used at just the right moment, a metaphor that springs from one of your passions, a broken rhythm that reflects your anxiety or enthusiasm. These things can't be encoded in a prompt. AI can mimic a style, but it can't inhabit it. As I explained in the article on brand voice, → "Your brand voice isn't your logo", an author's voice isn't their prompt: it's the sum of their human choices.

The paradox of plausibility

AI is a victim of its own success: it's so good at seeming credible that it often fools us. Its text almost always passes the grammatical correctness test. But it never passes the identity test. The risk is falling into the "dictatorship of the perfect"  → "The dictatorship of "perfect"": texts as smooth as marble, and just as cold. If a text is too perfect, the human brain unconsciously identifies it as background noise and stops paying attention. We're wired to look for humanity in others — the small cracks that make a story feel real.

The 5 mistakes that turn a good copywriter into AI's unpaid editor

Many professionals are falling into a subtle trap these days: they think they're saving time, but in reality they're working for free for the algorithm, correcting mediocre texts that should never have existed in the first place. Here are the five most common mistakes I see being made every day.

Mistake #1: Using AI as a first draft instead of a second

It's the greatest temptation: "I'll get ChatGPT to do the draft and tidy it up afterwards." Wrong. When you correct a text produced by AI, your thinking is inevitably channelled into the structure the machine chose for you. You become a prisoner of its logic. Always write your own first draft. Even if it's just three disjointed lines, even if it's a rough scrawl knocked out quickly. Only then will you stamp your own direction on it. Use AI to expand, to clean up, to find alternatives. Never to lay the foundations.

Mistake #2: The weak prompt (Garbage In, Garbage Out)

"Write an article about marketing for SMEs" isn't a prompt — it's a sentence to banality. A vague instruction produces a vague result. AI needs context, constraints, a soul. Ask it to write in general terms and it'll draw from the general average of the internet, which is, by definition, mediocre. An effective prompt must contain the who, the what, the how, and above all the why. It must define the tone and, where possible, include examples of your style. Without those coordinates, you're sailing blind in an ocean of the obvious.

Mistake #3: Copy-paste without voice

Taking AI output and publishing it as-is is a form of professional suicide. It's not just a matter of ethics or intellectual honesty — we touched on this when discussing ghostwriting → "Ghostwriting isn't a lie" — it's a matter of effectiveness. Regular readers feel the difference. They feel when the author's "breath" is missing between the lines. If your reader stops trusting your voice, you've lost the only thing that sets you apart from a free chatbot.

Mistake #4: Not briefing AI on who you are

AI knows nothing about you unless you tell it. It doesn't know that you love short sentences, that you can't stand pointless jargon, or that you like quoting Antonioni films. If you don't build an author-identity into your working session, every time you open the chat you're starting from scratch with a stranger. It's like hiring a new assistant every morning and expecting them to know all your preferences by heart without ever having had a conversation.

Mistake #5: Expecting AI to know your client

Artificial intelligence is formidable, genuinely. It knows everything about the statistical client, chews through American buyer persona frameworks like they're nothing, and talks you through the perfect funnel in three seconds. But there's one thing it doesn't know, and probably never will: it doesn't know Berto.

If you follow this blog, you'll know who I mean. If you're new here, let me introduce him: Berto is the owner of a small business in Caserta — a practical man, with hands that speak of hard work and a deep-rooted suspicion of anything digital, born out of a nasty fraud he suffered in 2012. Berto isn't a "market segment". He's a person who gets emotional when he talks about his workers and who needs to look you in the eye before he'll trust you with his communications.

Then there's Bobby, his counterpart in Manchester or London. Bobby's different in his manner — he'd sooner have a pint than an espresso — but at heart he's the same: he's after truth, not flashy slide decks. The names aren't accidental: they're both diminutives of my own name, Umberto. I created them as a personal reminder that behind every strategy there's a human being, not a demographic profile.

That's precisely why AI will never replace you entirely: that specificity, that closeness to local and human reality, can only enter the text by passing through your filter.

My method: how I use AI in my writing every day (with real examples)

I want to be straight with you: I'm no purist who writes exclusively with a quill. In my day-to-day work at Insight ADV, AI is a constant presence. But it has a specific seat at the table: it's not at the head of it — it's the one taking notes quickly. Here's my actual workflow, the one that lets me produce deep content in half the time.

The 4-phase framework: Brief, Human Draft, Amplification, Demolition

Phase 1 — Internal brief.

Before I even look at the screen, I write five to ten lines for myself in a notebook. Who needs to read this text? What should they feel? What's the specific anecdote I want to use? This is my sacred skeleton. I don't give it to the AI — it's my compass.

Phase 2 — Human draft (even if incomplete).

I write the opening paragraph and section headings by hand. It's essential that the opening is one hundred percent mine. That's where the tone is decided. If the beginning is human, everything else can be guided far more easily.

Phase 3 — AI amplification.

At this point, I ask AI to help. "I've got this idea for section X — can you find me three real examples of companies that have applied this concept?" or "Expand this technical point so it's accessible to someone who doesn't work in design." I use AI as a fast researcher and as a rhetorical sparring partner.

Phase 4 — Demolition and reconstruction.

This is the crucial phase — the one nobody ever teaches you. I take the AI output and I don't simply correct it: I demolish it. I keep the good ideas, but I rewrite every single sentence in my own rhythm. If the AI has written a three-line sentence, I break it into three short ones. If it's used a dull, generic word, I put a concrete image in its place.

A practical example. The AI gives me: "the implementation of content marketing strategies represents a fundamental element for sustainable business growth". I demolish it like this: I delete everything except the idea of "growth". I rewrite: "Write well or disappear. That's not a threat — it's just how the market works". Same idea, completely different life.

A concrete example: the same text before and after

Let's say we want to describe the importance of a logo for an SME.

Pure AI draft

"A company logo represents the fundamental visual identity of a business. It is essential that it be professional and recognisable in order to attract customers and build trust in today's competitive market." (Correct, but mind-numbingly dull.)

My method (rewritten)

"A logo isn't a design indulgence. It's not there to 'look nice'. It's the first point of contact. The first impression. The first glance a client gives you, before they even know who you really are. It's a silent handshake. If your logo is weak, confused, or thrown together… it's like turning up to an important meeting in your pyjamas. You might be brilliant at what you do. But nobody's going to bother finding out. And this isn't about Caserta or London. The rules are the same everywhere. You don't get a second chance at a first impression. Either it happens to you… or you make it happen."

See the difference? In the second text there's an opinion, there's a concrete image (the pyjamas), there's a geographical reference, there's a varied rhythm. AI provided the concept of "identity". I put the life into it.

Prompt engineering for copywriters: how to build instructions that actually work

Don't let the word "engineering" put you off. For a copywriter, writing a prompt is simply writing a good creative brief. If you know how to explain to a colleague what you want, you already know how to talk to an AI. The key is precision.

The structure of an effective prompt: Role, Context, Tone, Constraints

A prompt that works needs to be structured like a military order but written with a poet's sensibility. Here are the elements you can't leave out:

Role. Don't tell it "write a text". Say: "You are a senior copywriter with twenty years of experience in B2B marketing for Italian manufacturing companies."

Context. Explain who the reader is. "My reader is a small business owner who works twelve hours a day, is sceptical about marketing, and likes concrete facts."

Tone. Be specific. "Use a direct tone, occasionally ironic, and avoid the phrases 'industry leader' or 'end-to-end solutions' at all costs."

Constraints. "Use the first person singular. No bullet points. Keep the text to 300 words."

The "voice prompt": how to transfer your identity to AI

An advanced technique I use every day is having a personal "system prompt": a block of text that describes exactly how I write. It says things like: "Umberto loves metaphors rooted in Neapolitan cooking and cinema. He prefers short sentences. He often uses rhetorical questions. He never uses the corporate 'we'." By pasting it at the start of every session, I train the algorithm to tune into my frequency. It'll never be me, but at least it'll stop sounding like a washing machine instruction manual.

How to spot when AI has taken over: the warning signs

The greatest danger is habituation. The more we read AI-generated texts, the more our ear risks becoming accustomed to that monotonous cadence. We need to develop a diagnostic sense for knowing when it's time to reach for the editorial chainsaw.

The list of symptoms:

⚠️ WARNING SIGNS: YOUR TEXT SMELLS OF AI IF…

• It opens with "In today's dynamic world…" or "In an increasingly competitive landscape…"
• It closes with "In conclusion, it is essential to underline that…"
• Every paragraph is the same length, as though measured with a ruler
• There's no trace of the first person, anecdotes, or specific sensory details
• Not a single sentence under 12 words, not a single rhetorical question
• No concrete geographical, temporal, or emotional references
• You could swap the author's name for anyone else's and nothing would change

The read-aloud test

This is my secret weapon — the one I recommend to all my collaborators. Take the text you've written (or that AI generated for you) and read it out loud. If you find yourself running out of breath because the sentences are too long, cut them. If your voice sounds fake, like you're reading a station announcement, rewrite it.

It should sound as though you're talking to a client over a coffee in a Caserta café or a pint in a pub in Southwark or Salford. If it sounds like a tutorial, bin it.

The hybrid workflow that actually works: what to delegate and what to never give away

To thrive in the age of AI copywriting, we need to accept a new role: we're no longer just writers — we're editorial directors of our own output. We need to learn how to manage the fastest intern in the world.

AI yes: the tasks you can delegate without guilt

AI no: the tasks that are yours by right

The concept of editorial oversight

Writing with AI doesn't mean writing less — it means thinking more. Every output that comes out of the chat requires your editorial sign-off. Not a grammar check: an identity approval. You need to look at that text and ask yourself: can I stand behind every single word? Does this text add value, or is it just noise?

AI is the bricklayer who brings the materials. You're the architect. And the architect is the one who decides whether the building will be a masterpiece or a faceless block of flats on the outskirts of town.

Conclusion: your nan's ragù doesn't come out the same with the same oven

We've reached the end of this journey and I want to leave you with an image that, for me, sums everything up. Think about your nan's ragù. If you took the exact same ingredients — the same meat, the same tomatoes — and used the exact same state-of-the-art oven, your ragù would never turn out like hers. Why? Because in hers there's the "time" that only she knows how to wait for, there's that way of stirring the wooden spoon that she learnt over fifty years, there's the sensitivity to know when the flame is too high just from the sound of the pot.

Artificial intelligence is the oven. It's powerful, it's efficient, it heats up quickly. But if you think all you need to do is switch the oven on to get your nan's ragù, you're going to be disappointed. And your clients will be disappointed too.

Authentic writing in the digital age isn't writing that rejects technology — it's writing in which technology is invisible. A tool that doesn't show up in the final result, just as you don't see the oven when you're eating a divine pizza in Naples: all you taste is the flavour, the texture, and the love that someone put into it.

That evening in Caserta, after reading that cold, alien text, I deleted everything. I picked up the cold espresso, looked out of the window, and started writing about how the craftsman's time is circular, not linear. I used my own words — perhaps less "perfect" than the algorithm's, but loaded with everything I'd seen and breathed in that workshop. The client approved it first time, telling me: "Umberto, it's like you were sitting right next to me the whole time."

AI could never have sat next to him. You can.