"With" or "without AI" doesn't mean anything anymore
Rewording a sentence and generating a whole paragraph with a language model are not the same thing. And yet we keep lumping the two together: with AI, without AI, full stop.
The result: a few years of pointless friction. And worse, a bad answer to a question that deserved better.
The real problem is the vagueness
An Instagram post generated from scratch and a photo that's simply been upscaled with an AI tool have nothing in common. Yet today both end up under the same label: made with AI.
This vagueness doesn't just bother purists. Creatives have no words to describe their process. Clients don't know what they're paying for. And the public no longer really has a way to form an opinion.
Niels Ackermann, a photographer (Lundi 13), calls it an AI thermometer. You can see him lay out the idea in Décryptage Culture on RTS.
And the problem is far from theoretical. The show A bon entendeur (RTS) decided to test it themselves: in a single day, the team put together an illustrated children's book entirely with ChatGPT and listed it for sale without issue on a major online bookstore. The text held up reasonably well. The illustrations, much less so. And no buyer had any way of knowing what was really behind the book.
Why now
Generative AI is no longer a gadget people quietly experiment with. It's woven into the creative process at every stage, from idea to execution. The question isn't really whether it's being used anymore, but to what degree.
At the same time, Switzerland's copyright law reform is moving forward. And that kind of legal overhaul needs solid definitions, or it risks turning into a free-for-all.
There's also, more simply, a matter of trust. Without shared vocabulary to describe what was actually done, any mention of AI becomes suspect by default — even when human work still makes up the bulk of the effort. Conversely, a precise disclosure protects everyone: the public and the creators alike.
A scale, not a checkbox
DALIA emerged from this observation in late March 2026, driven by three people who don't even share an office: Niels Ackermann (Lundi13), Baptiste Lefebvre, aka Cetusss (Bureau culturel de Genève), and Alexandre Pugin (TWKS). RTS covered it in an article at the time of launch.
Their idea: stop drawing a line between two worlds and start measuring instead. DALIA has six levels, from D0 to D5, describing who came up with the idea, who executed it, and who judged the final result.
One detail that changes everything: DALIA doesn't judge quality. A D5 project, nearly autonomous, is neither better nor worse than a D0 project, done entirely by hand. The scale describes a process, nothing more.
And a single project can easily mix approaches. A traditional shoot with AI-generated visual effects added in post-production, for instance — in that case, you either apply different levels to different parts of the work, or use whichever level dominates.
What this actually changes
DALIA is free and open. No one owns it. Anyone can use it, adapt it, or fold it into their own practice — including to build a "without AI" claim when that's genuinely the case.
For an agency, a brand, or a freelancer, this means being able to explain their process with precise words, instead of hiding behind a vague phrase that tells no one anything — starting with the client.
Fear of AI isn't what makes transparency necessary. What matters is that we finally have the vocabulary to say what we actually did.
To go further
RTS — Dalia, the Geneva-made thermometer measuring AI involvement
A bon entendeur, RTS — AI makes its way into children's literature
Other sources: Bureau culturel de Genève, Le Temps (03/30/2026).
