Using a tree with AI

Why a tree is good input for an AI model, and a few places to start.

A Tree of Influence ends at understanding, and everything that comes after takes the tree as input. An AI model is one more thing that can take it as input. This page covers why a tree works well there, what to hand over, and a few places to start. This guidance should help you find uses that fit your own. It contains no prompts and names no products, because both change faster than a page can follow. The reasons survive much longer.

Whenever you are about to explain a situation to an AI model, and a tree of that situation exists, hand over the tree. The people who developed the Tree of Influence have done this for many kinds of work.

A tree of influence as input

These 5 properties do not depend on any particular model:

• A tree gives the model context it may lack. A model was not in the room. It knows your situation only through what it is given. A tree with a legend can be read by a stranger, and a model is a stranger.
• In prose, relationships hide inside sentences, and the model has to guess at them. In a tree, each one is an arrow with a direction.
• A tree puts understanding before solutions. Models rush to solutions similarly to the way groups under pressure do. With the tree in hand, a model can aim at the causes the group found, in addition to the problem as you happened to word it.
• A tree carries the group's judgment. Node types and votes record what people decided together. It represents the problems, the desires, the intentions, and what matters most.
• A tree is small. The whole notation is 4 node types and 3 kinds of arrow. A few lines of legend explain it, and even a large tree is short as text.

What to hand over

Most models are capable of interpreting a complete Tree of Influence from a legible image, such as an export from a digital board, or even a screenshot. Arrows on a crowded board are easy to misread, so check what the model saw.

These alternatives are extremely effective as well:

• List the nodes by type, then the arrows as sentences. It is the same text version that makes a tree readable without the board.
• Don't forget the legend. Include the context, the tree scope, who built it, and when.
• Make the parking lot clear. Say it in words if needed.

Then ask for what you need.

Here is the finished tree from the Quick Start, followed by the same tree as text.

Tree of Influence: why new hires take months to become productive
Built by the platform team, October 2026

Legend
[P] Root / Problem
[A] Action / Intervention
[O] Desired State / Outcome
Every arrow means "influences"
Votes show what the group wants to focus on

Nodes
[P] Each team onboards differently
[P] Documentation is out of date (2 votes)
[P] Docs are stale (duplicate of "Documentation is out of date")
[P] Setting up a laptop takes a week
[P] Senior people have no time to mentor (3 votes)
[P] New hires are afraid to ask questions (2 votes)
[A] Automate laptop setup
[A] Assign each new hire a buddy (1 vote)
[O] New hires ship something in their first month (4 votes)
[O] New hires feel welcome and confident

Parking lot
Buy a new wiki tool (no type yet)

Arrows
"Each team onboards differently" influences "Documentation is out of date"
"Setting up a laptop takes a week" influences "Automate laptop setup"
"Senior people have no time to mentor" influences "New hires are afraid to ask questions"
"Senior people have no time to mentor" influences "Assign each new hire a buddy"
"Documentation is out of date" influences "New hires ship something in their first month"
"Automate laptop setup" influences "New hires ship something in their first month"
"New hires are afraid to ask questions" influences "New hires feel welcome and confident"
"Assign each new hire a buddy" influences "New hires feel welcome and confident"

Places to start

Read this list the way participants read a pre-seeded board. Each entry is a hypothesis to be used, reworded, or ignored.

• Carry the understanding into the next tool. A tree feeds plans, backlogs, and decision records. A model can draft any of them from the tree, with the reasons attached.
• Ask the tree questions. Which problem reaches the most outcomes? What does every path to a given outcome pass through? Which nodes have no arrows yet?
• Ask for a challenge. Which arrows look weakest? What would a newcomer expect to find and not see? Take the answers back to the group as questions, not as edits.
• Compare versions. Hand over 2 saved versions and ask what persisted, what was resolved, and what emerged.
• Explain the tree to people who were not there. A one-page summary for leadership, a walkthrough for a new team member, or the same tree in another language.
• Keep it beside ongoing work. When a model helps with daily work in the same problem space, the tree tells it what that work is for.

In addition to the list above, in general anytime you find yourself doing these two things:

• You are writing paragraphs to explain a situation to a model. If a tree of that situation exists, hand it over.
• A model keeps answering the wrong problem. The understanding it lacks belongs on a tree. If no tree exists, consider building one first.

Beyond the algorithm

The conversation is the product. A model can read a tree, but it was not part of the conversation that built it.

• The group builds the tree. A tree written by a model and presented to a group skips the part that matters most, like a tree drawn alone.
• A model can help you prepare. It can pull candidate seed nodes out of interviews, incident reports, or survey results. Present them like any other seed, as hypotheses, and say where they came from.
• What a model says about a tree is a contribution. The people who own the tree decide what changes on the board.

Still being explored

This page is about a subject that changes faster than any other on this website. If you hand a tree to a model and something useful or surprising comes back, please share what you learned.