How we use artificial intelligence
We sell AI Act consulting, so the first company that has to be in order is our own. This page sets out how artificial intelligence actually enters the work we deliver: where it is involved, where we do not use it, what never goes into a tool, and who answers for the result. It is written to be checked against the facts, not to reassure.
Last updated: 21 August 2026
Where we use it
AI is a working tool, like a development environment or a spreadsheet. We use it where it saves time on tasks we are able to judge.
- Writing code Generating and reviewing code for our own projects and for clients. Code is read, understood and tested before it reaches a product: what we do not understand, we do not ship.
- Research and writing First drafts, rewording, cross-checking. Every legal reference we publish is then verified against the original text, in the Italian Official Gazette or on EUR-Lex. Citations do not come out of a model: they come from the source.
- Design and training materials Drafts, illustrations and images generated or reworked with AI tools, for the website and for training materials.
- Document analysis Reading and summarising documents we receive, within the limits set out below.
- Software products In some of our products AI is a feature, not a support tool: there we declare it inside the product itself, with instructions for use and known limitations.
Where we do not use it
This part matters as much as the previous one, and in client questionnaires it is the one asked first.
- This website It contains no AI systems. Browsing, forms and the AI Act self-assessment send nothing to model providers: the self-assessment is calculated by rules we wrote, and the same answers always produce the same outcome.
- Decisions about people We do not use AI to evaluate, select, rank or profile people, neither for ourselves nor on behalf of clients. We do not produce scores about individuals.
- Professional judgement The substantive choices, recommendations and conclusions we sign are ours. A model does not decide what to advise a client.
- Unreviewed deliverables Nothing reaches a client without being read and verified by a person.
What we never upload
Rules that hold even when breaking them would be convenient, and that clients can test by asking us about a specific case.
- Credentials and access keys Passwords, tokens, API keys, payment data. Never, in any form.
- Special categories of data Health data, beliefs, trade union membership and the other categories under Article 9 GDPR: only on the written instruction of the client and with a legal basis established beforehand.
- Archives of personal data We do not upload entire lists of a client’s employees or customers. We work on an extract reduced to the minimum necessary, or on data that does not allow an identified person to be traced.
- Client production systems We do not connect to live databases to run analysis: we work on exports or on agreed test data.
- Restricted documents Material covered by confidentiality agreements that rule out processing with third-party tools.
The tools, one by one
Listing the tools means being able to answer a supplier questionnaire without having to think about it. If this list changes, this page changes.
Anthropic Claude and Claude Code
Software development, research, writing. It is the tool most of the work runs through.
Paid professional plan: content is not used to train the models. Data processing agreement in place, transfers based on standard contractual clauses.
Google Gemini
Notes, summaries and image generation for design materials.
Retention and training settings verified on the account in use.
OpenAI ChatGPT
Occasional use, mainly for images and to compare answers.
No client data is entered into this tool.
If we adopt other tools in future, this page will be updated before they enter work on client projects.
Who checks, and who answers
MAXYMIZE BUSINESS is a sole proprietorship. The person who writes the prompts, reads the output and signs the delivery is the same one: Romano Maximilian Giurastante. There is no chain of controls to display, and drawing one would be dishonest.
What does exist, and can be verified: nothing is delivered without review; legal references are checked against primary sources; every document we publish carries the date of its last verification; and for our software products we keep compliance documentation we can show to anyone who asks.
Responsibility for the result is ours, always, whatever tool was used to get there.
What you can ask of us
- Work without AI You can ask for a specific piece of work to be carried out without these tools. Just say so, even once the engagement has started.
- Know how something was produced On request we set out which part of the work had AI support and which did not.
- Binding instructions If your company has an internal policy on suppliers’ use of AI, we apply it: send it over and it becomes part of the engagement.
Questions about this page
If something is unclear, or you need a more formal statement for a supplier questionnaire, write to us: we answer with documents, not reassurance.