Transparency: how we use AI

Blog articles, AI Strategy Papers and the copy on this site are written with the support of Claude, the language model developed by Anthropic. We use it to research sources, produce first drafts and run formal checks on the text, including the style review we apply to every piece before publication.

The substantive review, including the sources and figures cited, is carried out by Fabio Lalli, founder of ZeroFive.AI, who approves every piece before it goes live and holds editorial responsibility for it together with ICONICO SRL, the company that publishes this site. These are the rules we follow:

• nothing goes live without Fabio Lalli's approval of the final version;
• if an approved text is modified with AI, even just to rephrase a passage, it goes back into review before publication;
• every figure cited names the institution that produced it and its reference date;
• automated checks on the text support the review but do not replace it.

English versions are translated with the support of Claude and go through the same review as the Italian ones.

For every publication we keep an editorial log recording the content, the tool used, who reviewed it, the approval date and the approved version. At the end of each article you will find the same information in short form, with the date the text was approved.

The images on the blog are diagrams and infographics, not photographs. If we publish realistic images generated or modified with AI in the future, they will carry the label required by EU rules, embedded in the file so that it stays visible when the image is downloaded or shared.

In our projects we use language models to analyse documents, draft deliverables and build prototypes. Before uploading a client's documents to a third-party model, we agree with the client which data may leave the company and on which tools, and we use accounts configured so that content is not used to train the models. When data must not leave the client's perimeter, we work with locally run models.

Every deliverable is reviewed by a member of the team before it is handed over. In the AI Rating, models help collect and organise the evidence, while the final class is assigned by whoever led the assessment: the estimates produced by the online self-assessment tool are indicative and never assign class A, which is reserved for assessments verified through structured interviews.

ZeroFive.AI does not resell third-party software licences and receives no commissions or incentives from the technology vendors it evaluates for clients. When we compare tools from different vendors in a project, the choice starts from the client's requirements.

Our use of Claude in our internal work is not a recommendation to clients: in every project the model is chosen based on that project's data, costs and constraints. If we ever enter a commercial agreement with a vendor, we will disclose it on this page and in the projects where that vendor is among the options assessed.

If you find an error in our content, write to hello@zerofive.ai with the page and the passage. We correct factual errors in the text, and when a correction changes the meaning of a passage we note it at the end of the article with the date of the change.

From 2 August 2026, Article 50 of Regulation (EU) 2024/1689 requires those who publish AI-generated text on matters of public interest to disclose it to readers. The obligation does not apply when the text has undergone human review or editorial control and a natural or legal person holds editorial responsibility for its publication, as is the case for our content.

For this reason our articles do not carry the EU "AI" icon, which marks content subject to the labelling obligation. This page and the line at the end of each article are a voluntary transparency choice. The full picture is in our article on the EU icons for AI-generated content.