Skip to main content

AI transparency

What we use AI for, who reviews, and who is responsible.

What we use AI for

AI systems support our research, outlining and first text draft. They suggest wording, summarise sources and check texts against our own rules.

What they do not do: decide what gets published. No article, no image and no statement goes live without a human sign-off.

Who reviews

Before publication, every article is read in full by a qualified person, verified against the original sources and approved. Responsible for this is Visar, Founder & CEO, business information systems engineer.

A spell check or a blanket approval is not enough. What is required is a review of the substance, and a fact check is part of the minimum.

Sign-off comes last – the notice comes first

After human sign-off, no AI system runs substantively over the text again. This is not a matter of style but the condition under which the exemption from the labelling duty applies at all.

The notice, by contrast, comes first: the European Commission’s Code of Practice of 10 June 2026 requires a disclosure that is “clear, accessible and visible at the time of first exposure”. A note at the foot of the page does not meet that – anyone who stops reading after the third paragraph never sees it. Every affected article therefore carries the short form in its header and the full version, with reviewer and date, below.

If an article is revised later, the sequence starts over: draft, review, sign-off. The date of the last review is shown on the article.

Why we disclose this although we need not

As of August 2026 there is no statutory labelling duty for AI-generated content in Switzerland. On 12 February 2025 the Federal Council decided to ratify the Council of Europe’s AI Convention and to adapt the law sector by sector rather than create cross-cutting regulation; the consultation draft is to be prepared by the end of 2026.

Article 50 of Regulation (EU) 2024/1689 has applied since 2 August 2026 and can reach Swiss businesses where content is discernibly also intended for an EU audience. Mere accessibility of a website from the EU is not sufficient on its own.

The Digital Omnibus does not change this: Regulation (EU) 2026/1744, in force since 27 July 2026, postpones the obligations for high-risk systems to December 2027 and August 2028 respectively – it leaves Article 50 untouched. Anyone concluding from the headline “the EU delays the AI Act” that the transparency duty is deferred is reading the wrong half.

We disclose it anyway. Anyone who wants to know how a text came about should be able to read it without having to ask.

What this means for client projects

For client projects the same separation applies as above – and it is not the same across all divisions.

Brand and design materials from the Studio division are created by hand by a professional graphic designer, without generative AI. The IT division's services do without it too. What this page says about AI use does not concern those two areas.

For content, software and automation – Growth, Labs and AI – we work in a mixed way: drafting with AI support is permitted, delivery without human sign-off is not.

Clients who want to carry the disclosure in their own name receive the details from us – which systems were used, to what extent, and who reviewed. Since August 2026 that includes the technical marking described in the next section. This is governed bindingly in clause 16 of our terms.

The technical mark – and what it does not mean

Since 2 August 2026, Anthropic embeds an invisible statistical mark in text generated by newer Claude models. It does not change meaning, quality or readability, it survives copy and paste, and it withstands light editing. Generated image files additionally carry signed provenance data under the C2PA standard.

We state this here because it applies to our content, and because otherwise you would only learn it when somebody else measured it.

What matters is what such a mark says – and what it does not. Anthropic writes it plainly: a detection shows that a text was processed by Claude, and is not fully conclusive. It expressly does not establish authorship, because Claude is also used to proofread, translate and summarise. Conversely, the absence of a mark does not mean no AI system was involved.

It changes nothing about rights. Whoever wrote a text remains its author (Swiss Copyright Act) – a mark is a provenance signal, not a claim of ownership.

Nor does the mark replace our disclosure. Article 50 of the EU AI Act imposes duties in two different places: paragraph 2 obliges providers of AI systems to mark machine-readably, paragraph 4 obliges deployers to disclose to people. Anthropic covers one half, this page covers the other.

Legal basis

We value your privacy

We use cookies to understand how this website is used, to show and measure our ads on Google and Meta, and to tailor the emails we send via Klaviyo to your interests. Necessary cookies are always active; everything else only with your consent. More in our Privacy Policy