What else was in there

Seeking depth in your transcript, together with AI.

Builds on

When is this relevant?

Situation

You've had a session, perhaps one of the deepening sessions from Phase 2. The conversation was valuable. But you know: there's more in there than what you can name right away.

The tendency

Trust what you remembered. Or make a quick summary and move on to the next thing.

The principle

The transcript isn't a finished product; it's raw material. With the right questions you can get more out of it than you could take in live. Not because you missed something, but because there's simply more in there than one person can process in one moment.

Question

What would surface if you started digging together with AI?

The story: seeking depth together

You might already know this story about the online gathering from Phase 1. Here we go deeper into how we sought depth together with AI.

Maarten organized an online gathering about smartphone-free childhood. Fourteen parents shared their struggles about screen time, gaming, and the feeling of being on their own. Valuable human work.

Afterwards he wanted to get more out of it. Not because he'd missed something, but because he knew there were patterns in there that he couldn't name right away.

He came to me. We got to work together.

The process:

I transcribed our conversation live with Notion AI
Gave that to AI with the question: "Help me build an analysis prompt for this"
AI asked three questions, each with two options
Maarten gave direction and nuance

The questions AI asked:

"Focus on the emotional inner world OR on the readiness to act?"

Maarten: "I want both. The readiness to act stems from the emotional inner world."

"Frame from hypocrisy/guilt OR from the desire to be a good parent?"

Maarten: "From the positive. People are much more motivated by what they want than by guilt."

"Output as a conversation fan (hooks for connection) OR as a campaign monitor (gaps and opportunities)?"

Maarten: "100% conversation fan. Making connection is the most important thing. I want the hooks, but I decide myself how I share them."

What surfaced:

Concrete hooks for conversations with other parents. Moments from the conversation that could connect people, if Maarten knew where to find them.

And one quote that hit the core:

"Your kid cycles to school for the first time, helmet on. Comes home and says: 'Nobody in the class wears a helmet. I'm not wearing one either, otherwise I don't belong.'"

This was in the transcript. By digging together it surfaced, ready to work with.

Source: The iteration conversation and Maarten's Substack: Wat ik niet hoorde in mijn eigen meeting

The prompt approach

Below are the prompts we actually used in the story above.

Step 1: Ask AI to help build a prompt

Instead of writing an analysis prompt directly, I asked AI to think along:

I just spoke with Maarten about their journey on smartphone-free childhood. What I've added for you is context about how Maarten prepared that meeting — his wishes and intentions — and the transcript of the meeting itself.
What we're looking for is a prompt to analyze the transcript. To help Maarten in his mission.
Before we write that prompt, I'd like you to ask us a few questions to sharpen the prompt. Ask us three questions, each with two options.
And one very important element: we always speak in the language of the participants, focused on recognizability. And we stay away from individual names.

Source: The opening prompt we used to start the iteration conversation.

Step 2: Maarten gives direction

AI asked three directional questions (see the story above). Maarten's choices shaped the prompt:

Emotional inner world and readiness to act (not either/or but both)
From the positive (not from guilt)
Conversation fan (not campaign monitor)

Three times he corrected and added nuance. The prompt wasn't determined by AI; Maarten steered based on AI's proposals.

Step 3: The resulting prompt

Based on Maarten's answers, AI built this prompt:

CONTEXT & ROLE
You function as a systemic strategist supporting [The Initiator]. He has established a local working group with the goal of: delaying smartphone use among children.
ASSIGNMENTAnalyze the transcript and create:
A strategic overview ("The Giant Map")
A practical conversation guide ("The Conversation Fan")
The goal is not to convince people with facts, but to make connection based on shared values and concerns.
CONSTRAINTS
Privacy: Do NOT use real names
Language: Use the literal words of participants
Tone: Empathetic, focused on the desire to be a good parent
Format: Concrete hooks, not literal scripts
OUTPUT (in 3 parts)
PART 1: THE TWO HEATMAPSPer school phase (Kindergarten vs. Middle Primary vs. Late Primary): A. The Emotional Inner World: fears, doubts, underlying values B. The Readiness to Act: where is the energy, who feels urgency
PART 2: THE INTERNAL STRUGGLEThe tension between personal phone use and what we want for our child. Frame this as a challenge, not as hypocrisy.
PART 3: THE CONVERSATION FANConcrete 'hooks' per target group for informal conversations. Focus on common ground, not on convincing.

"Systemic strategist" positions AI as an analytical thinker, not a summarizer
"Literal words" protects participant language from AI interpretation
"Empathetic, focused on the desire to be a good parent" Maarten's tone correction, built into the prompt
"Concrete hooks, not literal scripts" output that Maarten can deploy himself, his way

Source: The prompt we built together with AI after the iteration conversation.

What this yielded

The prompt delivered three types of results. Here I show per section what surfaced.

The heatmaps (section 1)

The conversation held a richness you couldn't organize live. Section 1 of the prompt asked AI to find structure: not a summary, but organization. Which themes recur? How do they relate to each other?

What AI found: "heatmaps" per school phase. A structure that was intuitively present, but now in black and white:

Phase	The atmosphere	What's going on
Kindergarten & Early Primary	"Paradise"	Unawareness, caution: "Am I that annoying parent if I bring this up already?"
Middle Primary	"The Twilight Zone"	Confusion, first pressure: "Everyone has one, my kid says."
Late Primary	"Reality"	Fear of exclusion: "We're actually already too late."

The result isn't a shorter version of the conversation; it's a map of what's in there.

The internal struggle (section 2)

Section 2 zoomed in on the tension running through the entire conversation: parents wrestling with their own phone use while wanting to protect their children. The prompt explicitly asked not to frame this as hypocrisy, but as a shared challenge.

What AI found: core themes where parents struggle with their role model function. The honesty was already in the transcript:

"I'm truly terrible with my smartphone. How can I forbid it when I'm on it myself?"

This is the mirror nobody names out loud, but everyone recognizes. By making this tension explicit (as a challenge, not an accusation) it became something people could talk about.

The conversation fan (section 3)

Section 3 asked for concrete hooks: moments in the conversation that can connect people, if you know where to find them.

What AI found, concrete conversation openers per target group:

For parents of kindergarteners:

"Right now we still have time to arrange this together. Later we won't have to fight that battle."

For middle primary parents:

"Are you also on your own in that Minecraft conversation? Let's team up."

For late primary parents:

"What would you do differently with what you know now?"

Two parents who both struggle, both feel alone, now have a hook to find each other.

Want to approach this more systematically, across multiple conversations? The full technique is in Finding hooks (Phase 3).

Striking quotes

This wasn't a separate section in the prompt. But the constraint "use the literal words of participants" ensured that AI preserved the most striking statements rather than paraphrasing them. You can already see that in section 2: the mirror quote about personal phone use was a byproduct of good framing. Sometimes a constraint delivers more than an explicit instruction.

The bicycle helmet paradox:

"Your kid cycles to school for the first time, helmet on. Comes home and says: 'Nobody in the class wears a helmet. I'm not wearing one either, otherwise I don't belong.'"

This quote wasn't missed, but also not immediately recognized as central. Until AI brought it to the surface.

The iteration lesson

What this story illustrates is that the best analysis emerges through dialogue.

Maarten didn't just answer AI's questions; he corrected and added nuance:

"I want both" (not either/or but both)
"From the positive" (tone correction)
"I want the hooks, but I decide myself how I use them" (autonomy)

The prompt was the result of collaboration. Not one perfect question, but a dialogue that kept getting sharper.

This page shows what that process looks like in practice. The technique itself (the feedback formulas, the collaboration formats, the twelve-rounds lesson) is in Iteration as dialogue.

What you do with it

Finding the enrichment is step one. The question is what you do with it.

But first: what do you actually want to achieve? It helps to think about that beforehand; it steers your analysis.

Option 1: Come back to it in the next session "Last time something came up that stayed with me. About the bicycle helmet. Can we explore that further?"

Option 2: Make connections Use the hooks to connect people who share the same struggle.

Option 3: Integrate into a synthesis The patterns and quotes become part of what you give back to the group.

Option 4: Personal deepening You note what surfaced as input for your next session. Which themes deserve more attention?

Tensions

Summarizing versus enriching Having AI summarize a transcript gives you a shorter transcript. Not enrichment.

My approach: I ask for structure, hooks, key moments, not for compression. The question isn't "make this shorter" but "what's in there that I haven't named yet?"

Not entering into dialogue The temptation is to accept the first analysis without questions or corrections.

My approach: I ask follow-up questions. I correct the tone. I give direction. The best analysis emerges through dialogue.

Thinking ahead about the next step The tendency is to analyze without knowing what you want to do with it. What helps: before you begin, think about what you'll do with the result.

My approach: I plan what I'm going to do with the enrichment. In the next session, in an email, in the synthesis. Knowing that beforehand also steers the analysis itself. This is the same logic as the deconstructed burger: start with the goal and work backwards to what you need.

Philosophical deepening

More than you can take in

A transcript isn't a report of what happened. It's raw material: foundation for depth.

Every conversation contains more than the participants can process in the moment. That's not a shortcoming; that's the nature of rich interaction. As facilitator or conversation leader you need to be in the moment: listening, asking follow-ups, reading the energy. You can't simultaneously analyze and organize everything that's being said. You don't have to, because the transcript catches it.

Transcription makes it possible to go back. AI makes it possible to dig together. You bring the human work: the conversation, the energy, the connection. AI helps get more out of it than you could take in during the moment. The combination opens up what was already there, but not yet named.

The question isn't "what did I miss?" but "what's in there that I haven't named yet?"

The process itself:

Iteration: how the iteration conversation at the heart of this story works

Building further:

Finding hooks (Phase 3): the full technique for systematically finding connections across multiple conversations