AI & transparency
Last updated: 27 September 2026
I automate conversations with AI. Here is, concretely, how I do it transparently and under control — in line with the EU Artificial Intelligence Act, whose transparency obligations have applied since 2 August 2026.
On this website
The contact conversation is a guided, pre-written journey: it is not an AI, and no AI analyses your answers. The demonstrations (phone, inbox) are illustrative simulations.
In the assistants I set up for your customers
- Clear disclosure: from the first message, the assistant states that the person is talking to an automated assistant (Article 50 of the AI Act).
- Human in the loop: a draft phase approved by your team before any go-live; handover to a human on request and for sensitive topics (complaints, refunds, whatever you define).
- Immediate stop: switch back to draft mode or turn off automatic replies at any time, channel by channel.
- No automated decision with legal or similarly significant effects on a person (refusal, individual pricing) without human involvement.
- Logging: conversations remain available in your tools (ManyChat, messaging, dashboard).
- No practice prohibited by the Act (manipulation, social scoring, biometric recognition).
Models and data
Depending on the project, I use ManyChat's built-in AI or models from Anthropic (Claude) or OpenAI (GPT), via their business API offerings, which do not use this data to train their models by default. Messages are processed only to answer them, based on the knowledge base you have approved.
A question?
Write to me: jonathan@ajsrl.be. Details of the data processing are in the privacy policy.