How Finrato uses AI

Last updated: 5 September 2026

Finrato uses generative AI to help write the explanatory text on this site. It does not use AI to calculate your results. This page sets out exactly where the line falls, who checks the content before it is published, and who is answerable for it. It is published to meet the transparency duty in Article 50 of the EU Artificial Intelligence Act (Regulation (EU) 2024/1689), which has applied since 2 August 2026.

1. Where AI is used

Generative AI models are used as a drafting and editing tool for the written parts of this site: the guides beneath each calculator, the FAQ answers, the explanations of how a rule works, the reference tables and the descriptive text you are reading now. They are also used to draft the Polish and Latvian wording, and as a coding assistant while the site is built. In every one of those cases a model produces a draft; a person decides whether it is published, corrects it, or throws it away.

2. Where AI is not used

The calculators themselves contain no AI. Every result is produced by ordinary deterministic code running in your browser, implementing the formula published by the authority that issues the rule — HMRC and GOV.UK, ZUS and the Polish tax administration, VID and VSAA in Latvia. The same figures in always produce the same figures out, and no model sits anywhere in that path. Nothing you type is sent to an AI system, or to any server at all. The share images used when a page is posted to social media are drawn programmatically from the site's own design tokens, not generated by an image model. There is no chatbot, no AI assistant and no automated decision made about you anywhere on this site.

3. Human review and editorial responsibility

Editorial responsibility for everything published on finrato.com is held by Łukasz Wójcik, the operator of the Site, as a natural person. Nothing on this site is published automatically. A draft becomes a page only after a person has read it, checked its figures and thresholds against the primary official source listed on that page, and verified the calculator behind it against an independent calculation. Where an automated watchdog detects that an official source has changed, it writes the change into a review queue and notifies the operator — it publishes nothing and edits nothing by itself. Every calculator page carries the date its figures were last reviewed.

4. The limits of that review

There is one reviewer, and being straight about that matters more than sounding institutional. Łukasz Wójcik is not a licensed financial adviser, accountant or solicitor, and there is no separate editorial board or second independent reviewer. Review is substantive on the things that can be checked against a source — rates, thresholds, qualifying periods, worked examples — and it is thinner on general explanatory prose. That is precisely why this disclosure is published in addition to relying on the human-review exemption in Article 50(4), rather than instead of it.

5. Machine-readable marking

Pages whose prose is AI-drafted carry a machine-readable marker in their HTML, using the IPTC DigitalSourceType vocabulary that the European AI Office's Code of Practice on Transparency of AI-Generated Content builds on. Under the AI Act that technical marking is a duty on the providers of the models rather than on a site that deploys them, so it is offered here voluntarily, alongside — never instead of — the plain visible notice at the foot of every page.

6. If something here is wrong

An AI-drafted sentence that survived review and is still wrong is an editorial failure, not an excuse. Report it to info@finrato.com with the page address and what is wrong. Corrections are the fastest thing acted on here, and the review date on the page changes when one lands. How corrections are handled in general is set out in the editorial policy on the Legal page.

7. Illustrations

The illustrations on this site (the banner image on the home pages and the picture at the top of each calculator) were generated with Google Gemini from prompts written by Łukasz Wójcik, who chose and reviewed them. They are abstract and decorative: they show no real people, places, products or brands and contain no figures, so nothing in them is information to rely on. Each image file is also marked as AI-generated in its embedded metadata (IPTC DigitalSourceType trainedAlgorithmicMedia).

For the disclaimer, terms of use, funding disclosure and editorial policy, see the Legal page.