Since 2 August 2026, yes — but only some of it. The obligation covers images, audio or video generated or manipulated with AI that imitate real people, places or events and could pass for authentic. A clearly synthetic illustration, or a voice that imitates no one in particular, falls outside it.

Content that meets two conditions at once: it imitates something that exists, and it could look authentic to a person. If either is missing, it is not one. A clearly stylised render is not; a synthetic photograph of a real product is.

Whoever publishes. Normally the brand or the advertiser, not the provider of the tool. Penalties reach 15 million euros or 3% of worldwide annual turnover.

The most common cause is not the content: it is access. Many hosting providers block AI crawlers by default, so they receive an error and never get to read anything. Other frequent causes: content that only exists once JavaScript runs, or missing structured data.

We analyse your site the way ChatGPT, Perplexity and Google's AI answers see it: what they can read, what they understand and which questions could surface you. You get a scored report with evidence and a prioritised action plan, with the files ready to apply.

Above all, not sending the browser what it doesn't need. We build static sites: the server hands over pages that are already made instead of assembling them on every visit. That removes most of the wait before anything appears.

Static, by default: the server hands over pages that are already made instead of assembling them on every visit. It suits most of the sites we are asked for — corporate sites, catalogues, landing pages — because they load sooner and depend on nothing that updates itself. We have nothing against content management systems: when several people publish daily, or a shop carries thousands of references, they are the right tool and we say so. What we do not do is set up that machinery for a site that changes four times a year. In practice: HTML, CSS and JavaScript by hand, a generator of our own when there are several languages, and the code versioned in a repository.

Yes, with each language on its own address. Translating the text is not enough: if the versions only appear when you press a button, search engines never see them. Each language needs its URL and its reciprocal links between versions.

Where we work most is pharmaceutical and health, and wherever professional judgement is the product: architecture and building surveyors, legal and notarial. But we close no doors: as AI changes what production costs, work comes in that we could not have taken on two years ago — from a sports club to a small brand with a large catalogue. From Barcelona, for the whole of Spain.

Yes. Not the kind that patches a system degrading on its own — on a static site that does not exist — but the kind that watches what does change while nobody touches it: the domain and its certificate, the contact form that can stop sending in silence, who still has access, and the rules outside. This August a hosting provider began blocking AI crawlers by default: entire sites stopped existing for ChatGPT without anyone having done anything wrong. Each month we run the review with a tool of our own and read it ourselves: if something fails, you hear immediately. Each quarter you get it in writing whether anything changed — we don’t get in touch just to get in touch — and what can be done better today than a year ago.