From a confidence gap to a step-by-step curriculum, from personal goals to a life-priorities catalogue, from trainer-led sessions to chatbot-mediated practice: how 97 conversations became a training methodology
In June, we reported on what 97 participants told project partners during the collaborative actions: 58 people with intellectual disabilities and 39 support workers, across six countries. Those conversations were never meant to sit in a report. They were always meant to feed directly into the next phase of work: the design of the training itself. They have now been translated into the Training Scheme (R2.2), the methodology that will guide every training activity the project delivers, from the first session to the last.
Here you can appreciate how the feedback received during the collaborative actions became three specific design choices.
From a confidence gap to a step-by-step progression
The gap between how useful people found AI and how confident they felt using it alone couldn't stay a passing observation. It had to become a design constraint. The Training Scheme uses a learning-by-doing approach: participants use AI tools in real or simulated scenarios rather than watching demonstrations. Confidence, the collaborative actions made clear, is not built by watching someone else use a tool well; it is built by using it yourself, with support close at hand for when it goes wrong.
Every training path opens with a recognition moment: realising that familiar tools are themselves forms of AI. Confirmed as one of the strongest engagement triggers during the collaborative actions, this moment now serves as the structural entry point of every activity. It reframes what a session is about before it has properly started — not "here is a new, unfamiliar technology," but "here is something you already do, and here is what has been doing it with you all along." Training starts from what people already know how to do, not from what the tool can do.
That ordering matters. It means the step-by-step progression that follows is never a climb from zero. It is a climb from a foothold participants already have, which is precisely what a confidence-first methodology has to protect.
From personal goals to a catalogue organised by life priorities
Participants engage most when AI serves something they genuinely want to do, and the Training Activities are built on that principle, in particular by being organised around the independent-living domains participants themselves prioritised: entertainment, communication and mobility, with health and learning calibrated to individual context. The order is not incidental — it mirrors the order in which participants raised these domains during the collaborative actions.
Content is sequenced starting from the areas where people already feel confident, using AI's power to reduce dependence on other people as the leverage to take on the challenge of new AI-based tools. A participant who already trusts themselves to plan an evening's entertainment, in other words, is a participant with a foothold from which to approach a harder domain — one where the stakes, or the unfamiliarity, are greater. The catalogue does not ask participants to prove themselves on the project's terms first; it meets them on their own.
From trainer-led support to chatbot-mediated autonomy
That same confidence, once built, needs somewhere to go. Scaffolding is how the methodology carries it forward: repeated cycles of doing with support followed by doing alone, so autonomy grows step by step rather than being assumed from the start.What makes its application here distinctive is what "support" turns into as it fades: not its disappearance, but its transfer from the trainer to the AI tool itself. Participants move from learning with a person to learning independently with the chatbot.
This transfer happens at two scales at once. Within a single session, the trainer leads the opening steps and the chatbot takes over during hands-on practice — the handover is immediate, visible, and repeated every time. Across the whole programme, the same handover happens more slowly: trainer presence is strongest early on and steadily gives way to chatbot-mediated practice as participants advance. The two rhythms reinforce each other — a participant experiences the shift session by session long before they experience it across the arc of the whole course.
By the end, independence does not mean doing without a tool. It means doing without needing a person to mediate it — a distinction the collaborative actions made unmistakably clear was the one that mattered to participants themselves.
Looking ahead
Three findings, three choices: a confidence gap that became a step-by-step curriculum built on recognition; personal goals that became a catalogue ordered by what participants themselves prioritised; and a fading dependence on trainers that became a deliberate, two-speed handover to the chatbot itself. None of these was self-evident before the collaborative actions. Each is now written into the Training Scheme that the whole project will build on.
The next step is delivery: the same throughline — start from what people already know, follow their priorities, and hand independence over deliberately rather than assume it — now moves from methodology into the training activities themselves, across all six partner countries.
AI-ASSISTANTS 4PID is co-funded by the European Union under the Erasmus+ Programme (KA220-ADU), project no. 2025-1-DE02-KA220-ADU-000360510. The consortium brings together IAT (Germany, coordinator), AIAS Bologna (Italy), CECD Mira Sintra (Portugal), INESC-ID (Portugal), Universitat Politècnica de València (Spain), Fundación Amica (Spain), ARFIE (Belgium), and HURT (Croatia). More information is available at ai4pid.eu.
Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Education and Culture Executive Agency (EACEA). Neither the European Union nor EACEA can be held responsible for them.