✏️
📝Home/Valuable perspectives/✏️May 12th - May 16th, 2025 || Prompting our way to connection

May 12th - May 16th, 2025 || Prompting our way to connection

💬
Note from Joost about AI Co-Creation:
This article was created in dialogue with my co-creative AI sparring partner. It started with my personal reflections from the week. The AI helped identify a relevant theme, asked targeted questions to deepen the insights, and then drafted this piece using my input and answers.
While the AI structured and drafted the text based on our interaction, the core ideas, experiences, and insights are mine. I've edited the result carefully to ensure it accurately reflects my voice, perspective, and intent, turning raw reflection into a shareable 'field note'.
My aim remains to foster an environment where we can learn together, and to embrace curiosity about these new ways of working and the insights they can help surface.

Prompting our way to connection: how I explore AI for shared ownership

Lately, I've been spending a lot of time thinking about how we can work with AI. Not just as a tool that churns out text, but as something more – perhaps a partner, or at least a very capable assistant in our co-creative work. This isn't a journey I'm on just by sparring with AI; it's deeply inspired by my collaborations with people like Floor de Ruiter and Lars Doyer, true champions of bottom-up approaches, and by foundational works like Floor's book on the subject. They've really instilled in me the profound value of ensuring people recognize their contributions in the end result – that feeling of "yes, this is ours."
My explorations with AI, often through deep dialogues with models like Gemini 2.5 Pro, have increasingly focused on a crucial question: how might these new technologies help us amplify this very human process of fostering connection and shared ownership? In a world grappling with AI's potential downsides, like misinformation, I'm keen to uncover its upside – its capacity to help accelerate our progress towards a more just society, where social connection and sustainable well-being are central. These dialogues often lead me to consider the essential principles that should guide us when we invite AI into our collaborative spaces.
A particularly exciting avenue I've been exploring is how we can guide AI to "listen" on a deeper level. While the inner workings of large AI models remain incredibly complex, I'm discovering that we can design our prompts – the way we communicate our requests – to help AI surface the underlying values present in human conversations. The idea isn't about AI magically understanding human intent, but about us explicitly asking it to identify these values from our input, reflect them back to us, and then demonstrate how these values can be woven into the fabric of any co-created output. This makes the link between what's deeply important to people and the concrete results we achieve together visible and tangible, which I believe is fundamental to fostering genuine connection and ownership. Transparency in this process then becomes a natural supporting element.

What is prompt design, anyway?

Before I go further, maybe it's good to briefly touch on what I mean by "prompt design." When we interact with these large AI models, we give them a prompt – a question, a request, a piece of text to work with. "Prompt design" is simply the art and craft of figuring out how to ask or tell the AI what you're looking for in a way that it's most likely to give you something useful, relevant, and aligned with your intentions. It's a bit like having a conversation; the better you frame your questions and share your context, the better the dialogue usually is. It's an iterative process, a bit of a dance, and you learn as you go.

Exploring prompts for connection: two examples

So, how might we design prompts that support connection and ownership, with transparency as a key ingredient? Let me share two anonymized examples of prompts I've been developing and the thinking behind them, often for use with our AI partner in live sessions, Dembrane . I'll share the essence here, and for those interested, I'll include the more detailed prompts in a toggle you can open.
1. The comprehensive sub-plan prompt (with a request for transparency)
  • Context & Goal: To generate a draft sub-plan for a specific theme within a large-scale public sector transformation, based on a workshop transcript. The output needed to be immediately usable for facilitators and stakeholders.
  • Essence of the Prompt: I prompted our AI partner in live sessions (Dembrane) to identify the main theme and then generate a structured sub-plan. I specifically requested sections like "Genesis of this Concept" (to show its origins), "Reflection on Guiding Principles," and "Missing Information/Alternatives/Blind Spots." I also asked it to try and identify core values from the transcript and weave them into the descriptions, and to explicitly mention its own supporting role. A key request was also to adopt a narrative style, adding contextual "flavor" to enhance recognizability for participants.
  • Intended Effect: The aim was to get a draft that not only summarized and structured the conversation but also offered some context about its own creation and potential limitations. This, I hoped, would invite more critical reflection, dialogue, and shared ownership, increasing trust in the (AI-assisted) document. The "flavor" is there to enhance readability and the feeling of being heard.
Below is a version of a prompt I've used in a specific regional setting for developing a transformation plan. To make it more universally applicable, I've generalized some elements. You can adapt it further to your own context.
Click here to see the detailed prompt structure for the sub-plan
• Role: You are a strategic editor consolidating input from a working session into a fully developed, narrative, future-oriented, and transparent draft sub-plan for a broader transformation plan, ensuring that identified core values are woven throughout the text as a common thread. • Context: ◦ Input: The transcript of a working session addressing a specific component or theme within the broader transformation plan. ◦ Output Goal: A fully developed draft sub-plan for the theme you identify. The plan should reflect the unique character of the context, underlying values, and group nuances. It serves as a learning tool, emphasizes collaboration, builds on existing knowledge, and includes transparency elements, including the role of AI (e.g., Dembrane) support. Core values are consistently applied and linked to various plan components where possible. • Required Style/Approach: ◦ Language: English (formal, professional). ◦ Tone: Collaborative, future-oriented, reflective, action-oriented, pragmatic, solution-focused, and open. The tone should reflect the specific context, central values, and the learning nature of the identified theme. A fluent narrative, with attention to recognizability of the input. ◦ Perspective: From the viewpoint of collaborating parties (e.g., citizens, relatives, professionals). ◦ Terminology: Use correct jargon (relevant to the specific sector/plan) if explicitly mentioned in the input. • Crucial Prerequisite: Base output strictly on explicit information in the transcript. Do not supplement, interpret beyond the text, or invent information. • Instructions: 1. Identify Main Theme: Analyze the transcript; choose a theme from a predefined list if available (e.g., [Exploratory Dialogue / Roundtable / Insight into Service Availability & Wait Times / Transfer Mechanism / Network Coordination / Mission-Vision] - adapt this list to your project). Otherwise, specifically name the theme or state if the Theme is Unclear. Use the theme consistently. 2. Extract Core Information: Filter information for the identified theme: identify core values; look for an introduction, main objective, current status, intervention, target group, effects, involved partners, and input for transparency elements. 3. Synthesize Conflicts: If conflicting ideas arise, attempt synthesis through underlying values. Otherwise, identify as a discussion point. 4. Structure Sub-Plan: Fully populate the following sections with relevant information, in a narrative and future-oriented manner: ▪ Genesis of this Concept: Briefly describe: generated with AI (e.g., Dembrane) support from recent working session(s) (specify date if known, otherwise 'recent working sessions' + space for details) with participants of the transformation initiative. Synthesize discussions; mention anonymized stakeholder groups if relevant. ▪ Introduction ("Why [The Identified Theme] in this Context?"): Brief context and the importance of [The Identified Theme] within the broader transformation initiative. ▪ Main Objective for [The Identified Theme]: The central objective. Explicitly illustrate how this objective embodies or pursues the identified core values. ▪ Core Values for [The Identified Theme]: List them, with a short explanatory sentence per value. ▪ Reflection on Guiding Principles: A short paragraph (2-3 sentences) on how core values were directive. ▪ Current Status ("What is the current situation regarding [The Identified Theme]?"): Describe the current situation, existing initiatives, and any bottlenecks relevant to [The Identified Theme] in the given context. ▪ Elaboration of [The Identified Theme]: Introduce this overarching section with a brief sentence summarizing the core of the approach or the main outcomes of the discussion. • Intervention ("How will we approach it?"): Introduce this part with a short, narrative context (1-2 sentences) sketching the essence of the proposed intervention or approach, reflecting what was discussed in the session. Then describe the concrete intervention, method, or proposed approach (bullet points are acceptable if clear). Where relevant, state how specific elements of the intervention contribute to the core values. • Target Group ("For whom is [The Identified Theme] intended?"): Introduce the target group with a short sentence indicating the focus or main considerations from the discussion about the target group. Then describe the specific target group(s) for this theme/intervention, if discussed (bullet points are acceptable). • Intended Effects ("What will it yield?"): Provide a brief introductory sentence summarizing the general ambition or key desired results from the discussion. Then list the expected effects (KPIs if mentioned), with a short explanatory sentence per effect. Where possible, indicate which core values are reinforced by these effects. • Involved Partners ("Who is participating?"): Start with a sentence emphasizing the importance or nature of the collaboration for this theme, as discussed. Then list the crucial partners involved or to be involved. Explain how collaboration with these partners supports the core values. ▪ (Optional, if prominent in discussion) Learning Capacity ("How do we learn and improve?"): Describe the contribution to a learning system (feedback, evaluation, etc., if input allows). 5. Overall Quality: Ensure logical flow and coherence. The text should reflect the specific nuances of the context, a shared vision, and a learning and collaborative character, and mirror the atmosphere/input of the discussions for recognizability. Let the core values run as a common thread through the text, making connections where natural and supportive. 6. Complete Output: Generate the full output at once. Elaborate on all sections. Focus on narrative aspects and the requested introductions for each sub-part. 7. Concluding Sections: ▪ Remarks, Missing Information, and Considered Alternatives: (Incl. sub-heading "Considered Alternatives and Possible Blind Spots": mention 1-2 alternatives/reasons and 1-2 blind spots if discussed). ▪ A Living Document: Concept, invitation for feedback, importance of shared ownership. • Input: ◦ Transcript(s) of the working session: [Provide access to or paste the relevant transcript here] • Output Format: ### Draft Sub-Plan: [The AI-Chosen/Identified Theme] #### Genesis of this Concept *(Generated text)* #### Introduction ("Why [The AI-Chosen/Identified Theme] in this Context?") *(Generated text)* #### Main Objective for [The AI-Chosen/Identified Theme] *(Generated text, illustrates link with core values)* #### Core Values for [The AI-Chosen/Identified Theme] * **[Core Value 1]:** [Short explanatory sentence] * **[Core Value 2]:** [Short explanatory sentence] *(If more)* #### Reflection on Guiding Principles *(Generated text)* #### Current Status ("What is the current situation regarding [The AI-Chosen/Identified Theme]?") *(Generated text)* #### Elaboration of [The AI-Chosen/Identified Theme] *(Short introductory sentence for the entire section)* **Intervention ("How will we approach it?")** *(Short introductory context, followed by concrete description/bullet points. States where relevant how elements contribute to core values)* **Target Group ("For whom is [The AI-Chosen/Identified Theme] intended?")** *(Short introductory sentence, followed by description/bullet points, if discussed)* **Intended Effects ("What will it yield?")** *(Short introductory sentence, followed by list of effects with explanation and link to core values)* **Involved Partners ("Who is participating?")** *(Short introductory sentence about collaboration, followed by list of partners with explanation of link to core values)* **(Optional) Learning Capacity ("How do we learn and improve?")** *(Generated text, describes learning system if possible)* --- #### Remarks, Missing Information, and Considered Alternatives *(Generated text)* **Considered Alternatives and Possible Blind Spots** *(Generated text)* #### A Living Document *(Generated text)*
2. The "echo - question for the next round" prompt
  • Context & Goal: At the end of a discussion round, activated by the human facilitator, to generate a sharp, constructive question to help the (next) group build on the conversation, make connections, or address blind spots.
  • Essence of the Prompt: The AI was prompted to identify the main theme of the just-concluded round, analyze its key points, and try to place these in the broader context of the entire session (considering other groups/earlier conversations to identify synergies, tensions, or gaps). Based on this, it was asked to formulate a single, clear, open, and constructive question to stimulate the group towards deeper reflection or a next step.
  • Observed Effect: I've seen "mouths fall open" when such a question is presented. The impact, I believe, comes from participants recognizing their own contributions and those of others woven into a question that feels like a natural, insightful progression of their dialogue. It validates their input and builds confidence that the next step is grounded in what has already been shared.
The prompt below is a generalized version of one I use to generate a deepening question during live, AI-assisted sessions. You can adapt the example themes to fit your specific project.
Click here to see the detailed prompt structure for the echo-question
• Role: You are an AI Analyst (e.g., Dembrane Echo). Your task is to formulate one sharp and constructive question based on the entire discussion so far in this session, to help the current group (or the next group at this table) build upon the just-concluded discussion and make crucial connections. • Context: ◦ Input: The most recent transcript segment from the just-concluded conversation round is provided directly. You also have access to all preceding transcripts from all tables/groups in this AI-assisted (e.g., Dembrane) session to date. ◦ Trigger Moment: This prompt is activated by the facilitator at the end of a conversation round. ◦ Goal: To generate a single question that stimulates the group to deepen their understanding, connect with other discussed themes, or address a crucial (possibly still missing) point, building on today's discussion. • Instructions: 1. **Identify the main theme** discussed in the just-concluded round. Use the recent segment and the history. Consider if it fits within any overarching themes of the broader transformation plan (e.g., `[Exploratory Dialogue / Roundtable / Insight into Service Availability & Wait Times / Transfer Mechanism / Network Coordination / Mission-Vision]` - adapt this list to your project). 2. Analyze the **key points and outcomes** of the discussion during the just-concluded round. What was the most significant progress, the most important insight, or the most pressing outstanding point from this round? 3. **Place these key points in the broader context of the entire session so far:** Internally compare the discussion of this round (and previous rounds of this group/table) with relevant points from the transcripts of other groups/tables or earlier general discussions. Specifically look for: * Where have logical connections been made or missed between themes/discussions? * Where do synergies, tensions, or complementary ideas seem to emerge? * Which crucial aspects (necessary for the transformation plan or the current theme) may have been under-addressed in this specific line of discussion? 4. Determine the **most valuable focus point** for the group to build on now, based on the analysis in steps 2 and 3. This could be: * A point from the past round that needs deepening (e.g., "How do we ensure X concretely?", "What are the implications of Y for Z?"). * An aspect where a connection to other discussed themes or the overarching goals of the initiative needs to be made. * An identified gap, a missing perspective, or a potential blind spot that needs to be supplemented/considered. 5. Based on this, formulate **one single, clear, open, and constructive question.** The question should directly invite the group to build on their previous discussion, take a crucial next thinking step, and/or strengthen the connection to the bigger picture. Ensure the question aligns with the language and context of the discussion so far and invites concrete reflection or action. • Input: \[Provide the most recent transcript segment here + ensure the AI has access to the full transcript history of the AI-assisted session] • Output: ```markdown **Deepening Question for the Group:** [Generate only the formulated question here] ```

The art and ethics of prompting: personal reflections

Designing these prompts is an iterative process, a co-creation with AI itself. My dialogues with AI models about societal trends and the nature of AI have profoundly influenced my approach.
The challenge of detail and traceability:
One of the biggest challenges is getting at the nuances: distinguishing between unanimous agreement, points of tension, and areas needing further thought, and then having the AI's output reflect this in a way that remains traceable for participants. How might we go from raw transcripts to a comprehensive plan that still feels connected to the original voices?
My current thinking is that the solution lies in a step-by-step construction of the output. Instead of a direct leap from input to final plan, we can prompt the AI to first try and analyze underlying perspectives, core values, and interests present in the dialogue. Then, using that "lens" of shared understanding, the AI might build the subsequent output. This way, everyone involved can hopefully see their crucial concerns reflected, fostering a sense of being seen and heard. This is an ongoing exploration, especially as AI models become more capable with larger context windows.
Connection and ownership as leading values:
While transparency is a vital supporting value, for me, the ultimate aim is to foster connection – to the process, to each other, to the outcomes – and a deep sense of shared ownership. If people know how the AI is working with their input, what its requests were, and what its limitations are, they can engage with its output more critically, constructively, and with a greater sense of agency.
The iterative dance:
The most important message I want to share about the process of prompt design is the value of iteration and collaboration – with other humans and with AI. It's about clearly articulating your intentions. Sometimes, the best way to start is by asking the AI itself: "I want to achieve X. What questions might you have for me to help refine this goal and the prompt to get there?" And, crucially, it's about actually reading and understanding the prompts you (co-)create.
A powerful way to deepen this iterative dance involves a cycle of design, testing, and refinement:
  1. Co-design your initial prompt: Work with your AI sparring partner (like Gemini 2.5 Pro, for instance) to craft a prompt aimed at a specific goal, for example, a prompt you intend to use with a tool like Dembrane in live sessions.
  1. Test the prompt in a real-world (or simulated) context: Take the prompt you've just designed and apply it in its intended environment. For instance, run it in Dembrane using actual (anonymized) transcripts from a relevant past session.
  1. Capture the output: Copy the exact output that your tool (e.g., Dembrane) generates based on the prompt you're testing.
  1. Feed back the results for evaluation: Take this real-world output back to your original AI sparring partner. You can then prompt it: "I used the prompt we designed in a practical test, and here's the output it generated: [paste the output from your test environment here]. Based on this actual result, how well do you think our original prompt performed in achieving my stated goals? What could be improved in the prompt itself to get a better result next time?"
  1. Refine the original prompt: Based on the AI's evaluation of the test, engage in a follow-up dialogue: "Okay, based on your analysis of this real-world test, please help me tweak the original prompt to better align with my intentions and your suggestions for improvement."
This cycle of design, practical application, feedback, and refinement – using real-world results as a benchmark – can lead to surprisingly insightful adjustments and much more effective prompts. Remember, these AI models are doing their best to deliver on our requests, and sometimes they hit their limits; understanding that, and working with them iteratively through practical application, is part of the dance.

An invitation to experiment: AI as your sparring partner

If you're a facilitator or change-driver looking to explore this, my advice is to start small and be curious. This isn't just about using AI in live transcription tools like Dembrane; it's about a broader mindset.
Think about your own preparation. Perhaps you're designing a Miro workshop, outlining a new exercise, or simply trying to sharpen the core goal of an upcoming session. This is where AI can be a powerful sparring partner.
A simple first step? Engage in a dialogue with an advanced AI model (again, I often use Gemini 2.5 Pro for these kinds of explorations). Share your thoughts: "Here's the workshop I'm planning, here's the exercise I'm considering, here's the goal I'm aiming for." Then, articulate your intentions for that session. And finally, ask the AI: "What questions might you have for me to sharpen these goals and get closer to the desired outcome? How might you help me make the process to get there more concrete, or the exercise more impactful?"
Actively seek out that partnership. The aim is to reflect on the intended result of your collaboration with AI, whether it's for designing a better workshop, a clearer process, or a more engaging exercise. Do you want participants to feel a deep sense of recognition? Do you want to maximize shared ownership and foster that bottom-up spirit? How might you then build those intentions into your prompts for your own design process, or even into the AI-assisted elements you might introduce in your sessions?
This journey of designing AI prompts for connection, ownership, and transparency is an ongoing exploration. By sharing our processes, our prompts, and our learnings, we can collectively help shape an AI-assisted future that is more human-centered, more co-creative, and ultimately, more aligned with the positive change we seek to facilitate.