Don't let artificial intelligence make you stupid
Image from Iemza’s work in the exhibition “Déclaration” at Le Cellier, Reims, 2026
For years, web editors have written like machines, for machines. Generative AI now does this work better than they do. So much the better: this is an opportunity to regain control over what has never been mechanical—namely, expertise, perspective, and creativity. Provided, of course, that we don’t entrust our thinking to the tool as well.
In 2017, I took the stage with a message that sounded almost like a plea: “We’re not machines—let us create the best content.” I was making a simple point: Content that imitates its competitors offers nothing to the reader, and a search engine will eventually catch on.
Nearly ten years later, search engines and large language models (LLMs) are delivering exactly what we asked for. But a new threat has emerged. It doesn’t come from an algorithm. It comes from our temptation to let AI do the thinking for us.
We were already machines
Let's be honest about what we've been doing for the past decade. We used to analyze the top ten results on Google. We noted the terms they used, their length, and their subheadings. Then we'd write an eleventh piece that resembled them, checking off the boxes on an optimization checklist.
The semantic analysis tools weren't the problem: they measured exactly what they were supposed to measure. The problem lay in how we used them. We turned a guideline into a rule, and a rule into a writing method. The result? Thousands of interchangeable pages—properly optimized and utterly forgettable.
Today, generative AI can perform that task in a matter of seconds. It identifies the intent behind a query, covers a lexical field, and structures an outline. It does so quickly, on a large scale, and without fatigue. Recognizing this strength is by no means a sign of defeat: it is the starting point for a clear-eyed strategy.
The good news: machines free us from machines
If AI takes over the mechanical aspects of the job, what’s left for us? Everything it can’t do.
An LLM generates an average. It has been trained on existing web data and returns the most likely answer. It doesn’t know your customer’s feedback, the mistake you’ve seen a hundred times in the field, or the conviction that sets your brand apart from its competitors. It doesn’t take a stance; it strikes a balance.
At ASSONANCE, we sum up this division of labor in one sentence: AI analyzes, humans write. AI helps us understand a query, map out a topic, and automate repetitive tasks. Humans contribute what makes content worth creating: new information, a unique perspective, and a distinct voice.
The Real Risk: Becoming Stupid
This is where the title takes on its meaning. AI does not, by its very nature, make people stupid. It makes those who delegate their thinking to it stupid.
It’s easy to imagine the scenario. A writer asks an LLM for an outline, then a first draft, then a rewrite. He proofreads it, corrects a couple of phrases, and publishes it. He hasn’t learned anything about the subject. He hasn’t formulated any ideas of his own. After a few months, he can no longer write without assistance, and his writing sounds just like that of every other writer using the same tool. Writing is a muscle that needs to be exercised.
We’re falling back into the trap we faced in 2017—only worse. Yesterday, we were imitating our competitors. Tomorrow, we’ll all be imitating the same machine. And a generative model has no reason to cite a page that repeats what it already knows how to produce on its own.
Discipline, therefore, means staying in control of the steps that require thinking:
choose the angle,
to slice,
search for data that no one has published,
Question the expert and describe the specific case.
AI can support each of these steps. It should not replace any of them.
Writing means making decisions.
Optimize for Google and LLMs at the same time
More good news: writing for traditional SEO and for generative AI doesn't require two different approaches. Both reward the same qualities, albeit with different emphases.
Answer first, then elaborate. A generative engine extracts passages: each section must be able to stand on its own and clearly answer a question. The reader in a hurry benefits just as much as the algorithm.
Contribute something that no one else is contributing. This could be data from your business, feedback based on your experience, a quantifiable and verifiable example, or a well-reasoned opinion. It is this added value that justifies citing one page over another.
Be specific. People, products, places, methods: the more a text anchors its content in identifiable entities, the more useful it becomes to a search engine and the more credible it is to a reader.
Sign your content. An identified author, demonstrable expertise, and cited sources: these trust signals matter to both Google and LLMs, and they serve as a reminder that a human is taking responsibility for what is written.
A Different Way to Measure
If we change the way we write, we need to change the way we measure. Keyword rankings and organic traffic volume are still useful, but they no longer tell the whole story. An increasing proportion of responses are viewed directly in a generative interface, without a click.
There are new metrics worth paying attention to: your brand’s presence in LLM responses to your strategic queries, and how it’s presented there; and trends in searches for your brand name, which reflect brand awareness built up over time. Actual reading time and navigation depth, which reveal whether content holds the reader’s attention. Links and mentions obtained without outreach—proof that a piece of content was deemed worthy of being cited. And, of course, the contribution of content to conversions, even when it isn’t the final touchpoint.
Convince people that a well-written page pays off
That leaves the most delicate step: upholding editorial quality in the face of management or a client who sees AI as a way to produce more for less.
The strongest argument has to do with the cost of generic content. A page that any competitor can produce in thirty seconds has no competitive value. It costs little to produce, but it yields no return, and it dilutes the brand’s image.
Second argument: visibility is changing in nature. To be cited by a generative model requires providing something the model does not already know. Distinctive content therefore becomes a prerequisite for accessing these new visibility channels, rather than a luxury.
Third argument: AI doesn’t eliminate the editorial budget—it reallocates it. The time saved on research, structuring, and repetitive tasks can be reinvested in interviewing an expert, collecting proprietary data, or developing a point of view. With the same budget, the content gains value.
Remain the author
In 2017, we asked to be allowed to create better content. Today, no one is stopping us anymore. Search engines demand expertise, large language models (LLMs) value new information, and machines have taken over the work we used to do reluctantly.
The only thing that can still keep us from writing better is ourselves. Use AI to analyze, explore, and save time. But keep for yourself what makes you valuable: thinking, choosing, and putting your name on it.
