Just so we are aligned: when I write with artificial intelligence, I am not looking for “a text”, but for an ordered precipitation within my Gaussian Semiotic Cloud. Copy is not a point appearing from nowhere: it is the local effect of a distribution of possibilities – my Gaussian distribution of meaning – which collapses into a bubble when the context is clear, the audience is defined and the voice is mine. If even one of those three elements is missing, the bubble expands, meaning blurs and the result becomes vanilla content that anyone could have written.

That is why, before opening the prompt, I define who is speaking (invariants), how far I can deviate (tolerances) and what is excluded (banned elements). My I·T·B is not an obsession with control: it is the form that lets the bubble remain coherent without becoming rigid. The machine then accelerates the variations; I decide which one makes meaning collapse at the right point.

There is another check I consider non-negotiable: the difference between intention and interpretation. I may intend X, but the reader may read Y. I measure that alignment between intention and perception, even if only mentally, as a Semantic Adherence Index. I think of the text’s ability to make different readings converge on the same core as a Semantic Convergence Index. If AI accelerates too much and I lose focus, those two indices collapse, and the text, however “perfect”, no longer sustains the identity.

In practice, my rule is simple: voice before prompt; form before speed. A small difference at the start, an enormous one at the finish. Within the Cloud, the bubble stays dense and readable: the text precipitates where it should, not wherever it happens to land. If I ask the machine for three variations, it is to choose clearly which trajectory of meaning I want to make real here and now, not to delegate my profession to an algorithm.

When people ask me, “But does AI really write well?”, I answer: it writes fast. It writes well when we teach it to.

Here is what I focus on when I use generative AI for writing, or help others use it:

• I define the voice before the prompt. “If you do not know who you are, AI cannot know either.” That was one of the first lessons I learned, the hard way. In short, stylistic clarity must come from me, not the model.
  
  
• I choose a precise task. A LinkedIn post? A client email? A descriptive paragraph? AI performs best when I ask for one clearly defined thing at a time.
  
  
• I ask for constructive alternatives and reasons, not absolute truths on the first attempt. AI cannot always formulate truths, least of all at the first try. Output is a starting point: I treat it as a proposal to assess, not a verdict to execute.
  
  
• I always carry out a considered review. Always!!! Every line. Every number and every “truth” must pass through my knowledge, know-how, experience, sensitivity and direct validation. I do not just mean the obvious risk of saying something wrong, but a subtler risk that is hard to detect without real attention: “correct banalities”. Their effect is quieter and more insidious: invisibility.
  
  
• I document the process. Keeping track of prompts, iterations and changes helps me build a real methodology rather than creative roulette.

• ToV (Tone of Voice)
  The tone, rhythm, vocabulary and taboos that make a company recognisable in its writing. Without them, even excellent copy feels anonymous.
  
  
• Prompt contract
  The implicit or explicit agreement between you and AI: what you want to achieve, in what style and within which constraints. Without a clear brief, the output is random.
  
  
• Constructive alternatives
  Asking for alternatives, different perspectives and narrative angles. The aim is to practise plurality, not find “the truth”.
  
  
• Truth vs. validation
  AI can sound authoritative even when it is wrong. The final check is always ours, using experience and common sense.
  
  
• Correct banalities
  Sentences that are grammatically and logically perfect but empty of meaning and read a thousand times before. The number one enemy of differentiation.
  
  
• Narrative boomerang
  When you delegate too much to the machine and end up with a text that rebounds on you: impersonal, flat and damaging to your identity.
  
  
• Considered review
  Editing is not spellchecking. It is an act of responsibility: deciding what to keep, cut and rewrite. With judgement.
  
  
• Vanilla content
  Generic content that is correct but lacks personality, tone, depth or distinctiveness. It could work for anyone, so it strikes no one. Texts that never really get anything wrong but never leave a mark. Perfectly neutral, and therefore invisible.

If there is one thing generative AI excels at, it is speed. Tasks that once took hours – drafting a blog post, preparing five headline variations, translating a text – can now be completed in minutes with tools such as ChatGPT. For a small business with limited resources, that is gold: it means publishing content faster and freeing the team’s time for other creative or strategic activities. In practice, an article that once took eight hours to draft and revise can now be ready in two or three hours of actual human work, because AI has accelerated the more mechanical parts.

And speed goes hand in hand with scale: the same team can now produce more content than before. I was curious to find figures showing how widespread this awareness was, and found some very interesting statistics confirming that I am far from the only person to have realised it.

Generative AI is no longer science fiction: globally, 78% of organisations report using it in 2024 [1]. In marketing, around 70% of companies have already started using it [2] for activities such as personalised content creation and audience segmentation. Yet adoption among small and medium-sized enterprises remains in its infancy: in the EU, only around 11% of firms with fewer than 250 employees used AI systems in 2024 [3]. Even across marketing functions more broadly, only 10–14% of companies report [4] using generative AI on a genuinely ongoing basis in marketing and sales processes – a sign of the distance still separating theoretical enthusiasm from everyday practice.

Beyond doing more, faster, AI can also help us do better, at least in some areas. Properly integrated and guided, generative AI can genuinely strengthen the human copywriter’s work. This is not just theory: in an experiment involving 758 consultants, Harvard and BCG found that groups supported by GPT-4 completed tasks 25% faster, performed 12% more tasks and achieved final quality over 40% higher than the group without AI [5]. In other words, AI accelerates drafts and improves results, but only when it serves the professional, not the other way round.

Precisely because it “accelerates”, generative AI can quickly lead to glaring errors if used without oversight. In its responsible AI framework, NIST highlights the risk of “confabulation”: generative models tend to invent false but plausible content and present it with complete confidence [6]. Without a human filter, users risk accepting these hallucinations as fact, publishing them or acting on them. That is why the NIST guidelines explicitly recommend additional human review and greater oversight in workflows incorporating generative AI [6].

Another often underestimated advantage is personalisation. Properly instructed, generative models can convincingly adapt the same message to different audiences. Take email campaigns: AI can help rewrite the same content in different tones for different customer segments, or translate it into several languages while retaining its style. Where an SME might previously have abandoned localisation for lack of budget, it can now try, perhaps using AI for a first draft in English or German and then refining it with a native speaker. The result? Broader, more targeted communication without soaring costs. Recent studies confirm that generative AI in advertising brings lower costs and higher profits precisely through these production efficiencies.

In short, generative AI today is a little like the smartphone in photography. It has democratised content creation. Where once only a professional copywriter could produce good advertising copy, now anyone – an entrepreneur, assistant or intern – can get decent copy for a webpage or social post with ChatGPT’s help. BUT… that does not mean we have suddenly all become professional photographers, artists, filmmakers or writers. NO! It does not automatically give us excellent texts, just as taking a phone photo does not make us Annie Leibovitz. It certainly raises the minimum standard, though: emails, posts and product descriptions can have fewer typos, cleaner structures and even unexpected creative touches. For many SMEs, that alone is a step up in quality. Above all, it raises the bar for those who already did this professionally. It accelerates, improves, optimises, verifies, extends and deepens what the professional was already doing.

This is exactly the catch: the professional. I do not use that word casually. I mean people who were already photographers and can now do it better. People who were already waiters and can now do their job even better with AI. But if you could not write about Dumas before, perhaps because you do not even know who Dumas is, and now attempt a literary critique of The Three Musketeers generated from a hastily written prompt, the result may seem like a masterpiece to you. To someone in the profession, someone who knows the subject, who has studied, written, made mistakes, corrected and rewritten hundreds of times, that text will sound off-key, artificial and insincere. Those were diplomatically impeccable words. What I really wanted to write was: stupid, anti-cultural and false.

That is the point. AI can elevate the work of someone with expertise. It cannot magically fill a void in knowledge, experience or sensitivity. What initially seems a brilliant shortcut can become a boomerang: invisibility, lost credibility, an indistinct voice among thousands. AI is not a qualification. It is an amplifier. And if there is nothing to amplify, all you hear is noise.

And what I have seen is that, after a while, people give up… I no longer see literary criticism of The Three Musketeers or Dumas written by people who previously dealt with investment funds in Malta, or in-depth AI articles written merely because someone activated a ChatGPT PRO subscription. Or I see people persisting – that happens too – but anonymously, without evidence, without success, amid collective indifference and complete silence. It is like singing in the shower: there we are all Pavarotti or Mina. Thank goodness nobody hears us. Perhaps only the neighbours. But they are used to it. And they know the shower lasts only a few minutes.

Generative AI is not yet our “new copywriter” in the strict sense: it is not waiting in the office in place of the writer you fired. In fact, call that person back immediately, apologise and rehire them with a bonus! AI has instead become the new super-apprentice: incredibly fast, tireless, able to write in ten seconds what takes us thirty minutes, yet still needing a senior’s guidance.

For SMEs, this is a concrete opportunity: those who integrate AI into their creative processes can produce content faster, in greater volume and potentially more effectively without distorting their own voice. Those who delegate everything and “let the robot handle it” risk soulless, inaccurate or damaging communication.

The history of technological innovation shows that the greatest benefit comes when technology empowers people rather than replaces them. And nobody is being replaced here, whatever poorly informed journalists chasing audiences with sensational headlines, or clickbait-obsessed influencers, may claim.

And as Marshall McLuhan said: “It was not people who shaped tools, but tools that shaped people.”

Here I agree with McLuhan. But the process is not immediate, instantaneous or sudden. It requires time, assimilation, acquired knowledge and awareness. To be truly effective, useful and above all “healthy”, it needs attention, method and intelligence – human intelligence.

There is a typo in this article. Just one.

In the Italian original, the typo survived two rounds of AI review unscathed, but a friend spotted it instantly. A human. For this version, I have moved the stumbling block to a different passage.

Why leave it here? To remind myself and all of us that writing is not just speed: it is attention. That fraction of a second when the human eye senses the wrong note and meaning precipitates in the wrong place.

If you spot the typo, or already have, you have experienced the idea running through this piece in real time: AI accelerates, but we remain in charge. And sometimes the difference between a “correct” text and a “living” one is a single letter.