This article describes the content of the regulation. It is not legal advice and not an assessment of your specific case.
In a little over four weeks, on 2 August 2026, the transparency obligations of the AI Act become applicable. They are the provisions most visible in daily business, because they do not end up in a file but on the website and in customer communication.
The underlying idea is plain: people should know when they are talking to a machine and when content has been artificially generated. That is not a high bar. The work rarely sits in the implementation but in finding all the touchpoints in the first place.
Where companies overlook the touchpoints
The chatbot on the home page is obvious. The rest is less so, and that is exactly where the gaps appear:
- The automatic first reply in the mail inbox, which by now composes more than it confirms.
- The voice assistant on the phone system that takes and routes requests.
- The contact form with a suggestion feature that pre-writes answers.
- Product descriptions and blog texts generated with a tool and only lightly edited afterwards.
- Images on landing pages that were not photographed but generated.
In almost every business where we have run this inventory, at least one item surfaced that management did not know about. That is not a reproach but the consequence of such features arriving as product improvements to an existing tool, without anyone introducing them.
What marking looks like when it does not annoy anyone
The most common objection is that marking puts people off. Experience says otherwise when it is worded factually and immediately says what the customer gets from it and how to reach a human.
- 1.One sentence in the first chat window: that an automated assistant answers here, what it covers, and how to reach a person.
- 2.The same information in the phone announcement, before the request is taken.
- 3.For generated content, an internal rule on what gets marked and who checks it before publication.
- 4.A short page or section describing where AI is used in the company. That answers customer questions before they are asked.
The route to a human is the point that decides the effect. An assistant that clearly says it is an assistant and offers a way out in the next sentence reads as service. One that pretends to be an employee and sends the customer in circles reads as an obstacle.
Using the weeks until early August well
For a small or medium company this is not a project but a handful of appointments: record the touchpoints, write the texts, put them in at three or four places, and settle internally who checks before anything new goes online. That last point is what stops the gap from reappearing in six months.
If you are not sure how many places in your company already have customers talking to a system: that inventory is exactly what we do, and it rarely takes more than two days.