Generative AI has become part of how a lot of content gets made — including some of ours. We use it the way most serious publishers do now: to speed up research, keep formatting consistent across dozens of articles, and draft a first pass faster. What it doesn’t do is decide what we publish. Every piece of new content on FileZilla Pro is written and edited under direct human ownership, and nothing goes live without a named person reviewing and approving it first.
With the EU’s Artificial Intelligence Act now largely in force, we wanted to lay out plainly what the law actually requires around AI-generated content, and how our own process lines up with it. Not just as a compliance checkbox, but because we think it is the right way to publish.
What the EU AI Act actually requires
The AI Act (Regulation (EU) 2024/1689) regulates AI systems on a sliding scale of risk, ranging from practices it bans outright, to “high-risk” systems (things like credit scoring or biometric surveillance) that face strict controls, down to lighter-touch transparency rules for everyday tools like the ones behind most content-assist features. The obligations most relevant to a website like ours sit in Article 50, which sets transparency rules for AI-generated content. Those rules have applied since 2 August 2026.
The core rule for published text
Article 50(4) requires that AI-generated or AI-manipulated text be clearly labelled when it’s published to inform the public on matters of public interest — the European Commission’s own guidance points to things like politics, public administration, justice, public health, and consumer safety as the kind of subject matter this is aimed at. Product documentation and software comparisons sit in a genuine grey area or more likely just fell out; regulators haven’t spelled out whether ordinary commercial content counts. We’re not going to lean on that ambiguity as an excuse to skip transparency, as described in detail below.
The Act doesn’t require labelling when AI-generated text “has undergone a process of human review or editorial control” and a named person or organization holds editorial responsibility for what gets published. Crucially, the Commission’s guidance is explicit that this has to be real review, not a rubber stamp: a spell-check or a light copy-edit doesn’t count. It means someone with relevant knowledge actually examined the substance, and has the authority to change or reject it.
In other words, the law itself recognizes that AI-assisted content, properly supervised by a human editor, is a different thing from an unsupervised AI publishing to the world on its own, and treats it accordingly.
How we actually work
This isn’t new for us, and it isn’t a policy we wrote to match the law after the fact — it’s how our content process has run for a while, and the Act is a good reason to say so publicly:
- A person starts every piece. Every article — comparison pieces included — begins with a human writer deciding what we are covering and why, based on real product knowledge and, where relevant, direct testing run by that person of another team member, depending on the specific product.
- AI assists with format and research, not conclusions. We use AI tools to help keep structure and formatting consistent across articles, and to speed up first-pass research and data-gathering. It does not get the final word on what is stated or recommend.
- Every fact and claim gets checked. Pricing, feature claims, and technical details are verified against primary sources — vendor documentation, our own product, or direct testing — before publication.
- A named editor reviews and approves before anything goes live. This is the step the Act treats as decisive: substantive human review, with real authority to edit, hold back, or reject a piece, not a formality.
Why we are holding ourselves to this even where the law is unclear
Because whether or not “matters of public interest” turns out to formally cover a software comparison article, the underlying concern is one we share: readers deserve to know that what they are reading has a person standing behind it, who read it, checked it, and is willing to be accountable for it. That is a lower bar for trust than “an algorithm wrote this and nobody looked,” and it’s the bar we hold ourselves to across the board, not just where a regulator might eventually say we have to.
If anything changes about how we use AI in our content — the tools, the process, or the level of human review — we’ll update this page rather than let it go stale.