People first, then AI.

In 2025, I worked on sessions for Mental Health Network Gelderland and Mental Health Network Amsterdam. The facilitators had already designed the overall process. As the work progressed, I contributed to the session design. Together, we sharpened the purpose and worked backwards to the questions for participants and what we could bring back to them during the session. I aligned the use of AI with that: what it could reflect back, what material it needed and where a person had to make the judgment.

Start with purpose

The published example Prompt the people first uses a simple sequence. Start with the desired picture, then ask what is crucial, what different behaviour would be visible and what steps would make it real. The order matters. People speak from experience and aspiration first; only then does the conversation become analytical.

Seeing your own words again

People saw their own words in a draft on the screen. They could respond aloud: is this right, what is missing and what needs refining?

See the structure I gave the draft

This is the output structure from the published prompt, with spaces for what the group has said.

Identified theme
Social map / consultation table / exploratory conversation

Draft vision for this theme
A narrative summary in the required writing style, for validation.

Crucial elements
The elements participants say need to be part of the vision.

Read the full prompt →

The draft gives the group something to work on together. In the approach to developing an implementation plan, feedback feeds into the next version: first the vision, then the behaviour and steps it calls for. People continue to assess and decide.

A reflection at the chosen moment

In another live session, I prepared a prompt behind dembrane’s echo button. The facilitator chose when to bring that reflection into the conversation. What the group needed at that moment shaped the prompt. Read about the preparation.

The prompt had to learn from practice

My concrete judgment came in the prompt for the transformation plan. An earlier version referred to “the reference transformation plan”, even though AI would not have that document during the session. The prompt asked for something that was not available.

I therefore took the style characteristics out of the source document and put them literally into the prompt: language, tone, perspective, terminology and structure. That was one of several corrections in a process of twelve rounds. The prompt became usable because it matched the session plan, the purpose of each moment and the material that would actually be available in the room.

Read Prompt the people first · Read how the style entered the prompt · Read about iteration as dialogue

What I take from this

A good prompt begins with a good human question. A recognisable draft invites assessment; it is not evidence of agreement. And when a source is unavailable, the judgment you need from it has to be made explicit.

If you are preparing a session and wondering where AI could help, I like to start with what the people in the conversation need.