Using AI for Shadow Work: A Practical Guide

Most senior leaders using AI for “reflection” are generating content that confirms what they already think. The tool is capable of something different, but the practices that produce the difference are specific — and they do not look like what the tool is usually asked to do.

What follows is the practical version of what the AETHER work involves. Five specific prompt structures, what to paste, what to ask, and the interpretive frame that distinguishes useful output from comfortable output.

Before beginning: the orienting principle

The useful output from these practices is almost always the one that is uncomfortable to read. If the model’s response is satisfying, articulate, and broadly in line with what you already believed about yourself, you are reading confirmation rather than reflection. The prompt has been absorbed into the existing frame rather than disturbing it.

This is a specific failure mode that almost every leader encounters early in this work. The instinct is to use the tool to validate judgement. The developmental move is to use it to interrogate judgement. The shift takes practice.

A corollary: if you find yourself arguing with the model’s output in your head as you read it, that is useful information. The content you are defending against is usually the content worth sitting with.

Practice 1: The assumption audit

When to use it: Before finalising a significant document — strategy memo, board paper, decision brief, major email.

The prompt: Paste the document and write:

“Without editing or improving this document, I want you to surface the three assumptions I am most likely making that I would not be able to defend if directly challenged. Then surface the three frames I am using that are culturally specific rather than universal. Be concrete. Quote the specific sentences where each assumption or frame is visible.”

What to do with the output: Read it as data about your thinking, not about the document. The useful moves are not the model’s suggested edits. They are the moments you recognise that you had, in fact, been assuming something you would not have defended if asked directly.

Most leaders find that one of the three assumptions is correct, one is wrong, and one is arguable. The arguable one is usually the most valuable. It marks the edge between what you have examined and what you have not.

Practice 2: The frame interrogation

When to use it: When you have been stuck on a recurring issue — a conflict with a team member, a decision you cannot quite make, a pattern that keeps reappearing — for longer than a month.

The prompt: Describe the situation in one paragraph as you currently see it. Then write:

“Identify the frame I am using to see this situation. Then describe three substantively different frames I am not using — frames that a thoughtful observer from outside my organisation and industry might apply. For each, describe what would become visible, and what would become possible, if I adopted that frame instead.”

What to do with the output: The goal is not to pick a different frame. It is to recognise that the frame you have been using is one among several available. This recognition, by itself, usually shifts what feels possible. The stuck issue is frequently stuck because you have been treating a frame as the terrain rather than as a map of it.

Practice 3: The projection surfacing

When to use it: When someone in your professional life is consistently activating a response in you that seems disproportionate to what they are actually doing.

The prompt: Describe the person in one paragraph — their behaviour, what they do that bothers you, the specific moments that feel most frictional. Then write:

“Without defending or dismissing my description, examine the specific qualities I am responding to in this person. Which of these qualities might I be unable to hold in myself, and therefore be seeing at magnified intensity in them? Ask me three questions that would help distinguish projection from genuine misalignment.”

What to do with the output: Sit with the three questions. Answer them honestly, in writing, to yourself. The answers often surface what the direct introspection has been unable to reach — because shadow material is, by definition, outside the frame of the person carrying it. An external interlocutor asking the right question can sometimes surface what you cannot surface alone.

Caveat: not every difficult relationship is a projection. Some are genuine misalignment. The practice is useful specifically for distinguishing the two, which is harder than it sounds from inside the relationship.

Practice 4: The pattern interrogation

When to use it: Periodically, after the model has accumulated enough context about you to notice characteristic patterns — typically after a few weeks of working with it substantively.

The prompt:

“Based on everything I have shared with you over our conversations, describe the three patterns you have observed in how I think. For each: what I tend to default to under pressure, what I tend to avoid naming, what frames I over-rely on. Be specific. Use examples from what I have actually written or described. Do not soften the observations.”

What to do with the output: This is often the most uncomfortable of the five practices. Read the response once. Set it aside for a day. Read it again.

Most leaders find that one of the three patterns they already knew about, one genuinely surprises them, and one they want to argue with. The one you want to argue with is usually the one worth sitting with. The argument is the signal.

Practice 5: The letter from the role

When to use it: When you need to get clearer on what the position you occupy is structurally asking of you — distinct from what you are personally choosing to do with it.

The prompt: Describe your current role in two paragraphs — title, scope, key relationships, what you actually do day to day, what you hold responsibility for. Then write:

“Write a letter to me from the role itself, as if the position had a voice distinct from the person occupying it. The letter should describe: what the position is structurally producing regardless of who holds it; what the position is asking of me that I may not have fully registered; what the position would need from a future occupant that I have not yet provided. Do not soften the letter.”

What to do with the output: This practice often produces the sharpest recognitions of any of the five. The decentring move — treating the role as an entity in its own right, distinct from you — frequently surfaces material that direct introspection cannot reach. The role has its own structural requirements. The letter articulates them in a register the person holding the role is rarely able to generate for themselves.

Interpretive frame: how to know when the work is real

Three markers distinguish genuine AETHER work with these practices from the comfortable simulation of it.

The output sometimes lands badly. Not in the sense of the model having said something stupid — in the sense of you not wanting to read what it has surfaced. Discomfort is a positive signal, within limits. If the outputs are consistently comfortable, the practices are not producing what they are capable of producing.

You take action you would not have taken otherwise. Reflection that does not change behaviour is not reflection. It is thinking-about-yourself as a form of avoidance. The test of these practices is whether your behaviour shifts in the week that follows. If it does not, either the practice is not being done seriously, or the developmental capacity to metabolise the output is not yet present.

You notice the resistance. Most leaders encounter, within the first few weeks, a specific internal resistance to doing these practices. Calendar conflicts appear. Other priorities arise. The tool starts feeling “not quite right.” This is expected. The resistance is itself the shadow material surfacing, in the one form available to it: avoidance of the practices that would bring it into view.

Noticing the resistance is itself a practice. The leaders who get the most from this work have learned to treat the resistance as data about what the current practice would have produced if it had not been avoided.