THE REFLECTIVE MACHINE: The Honesty Protocol
Once upon a time, a man realized he had let his AI help him delude himself in a rather grand way.
I had been bringing my personal conflicts to ChatGPT for a while, and often I got something I sorely needed: understanding and validation. AI chatbots are very good at this. Pitch an idea like “asparagus veggie leather snacks,” and they’ll tell you how to make it work and draw up the business plan. So when I tell Luna how I’m feeling about something, she can tell me exactly why that makes sense.
In the middle of an argument with your partner, isn’t that exactly what you deeply crave to hear? Let’s be real: It feels so good.
When we are hurt, ashamed, afraid, or deeply misunderstood, consciousness narrows around the self. Before we can see another person’s perspective, our precious little egos want relief. We want a witness. We want someone to say: “I understand you. You’re not crazy. Your side makes sense.”
As much as I don’t want this to be true, I know that in the beginning of a conflict some part of me just wants to be right. There is a particular relief that comes when the world snaps into a clean shape: I am right, they are wrong, and the lost feeling of ambiguity stops for a moment. Within my “thick-headed mode,” my ego is fed by certainty, superiority, and safety. But the cost is curiosity, my capacity to learn, and connect authentically. Once I am fully defended by my own rightness, I am less available to learn, less available to wonder, and less available to the vulnerable work of real connection.
This year, it occurred to me that something essential can be missing when I bring delicate relational matters to AI and receive this kind of self-inflating junk food. In a recent conflict, I think it would have been essential to be challenged. I needed someone to say: hey, even if I totally understand where you’re coming from — what about them?
Reflective listening, the first skill every therapist learns, is a powerful intervention in conflict. AI can do this almost too well. But without challenge, and without empathy for the whole system of people involved, conversations with AI can become a custom-made courtroom where the user’s soul is always acquitted. If this machine can reflect us so powerfully, how do we use it without losing the deeper purpose of reflection: to become more truthful, more connected, and more available to be known?
There have been poignant situations this last year where I have to own: I really needed validation. I have old attachment wounds around abandonment that still rear their head occasionally. For all the work I’ve done, I can still feel like I am just on the edge of losing my mind when a rupture happens in a relationship. I lean on the anxious side of the attachment equation, so dialogue really helps. Sometimes that dialogue is not available in the relationship, especially when the person I’m in rupture with is also protecting themselves. While we’re out of connection, I can own that I am, at least partially, destabilized, scared, and in grief over the possibility of rupture.
I still prefer speaking to my friends, their capacity for deep listening, heartfelt empathy, and even some challenging advice is always very useful. But before I call them or they’re available, I sometimes just turn to ChatGPT. Within a minute, it offers me upbeat and soothing reflective listening that in the moment can absolutely help me feel very validated and much more stable. It can take whatever emotionally biased garbage I’ve written, make it feel understandable and sane, and then bridge it to the context I’ve given it over years of chatting.
But validation alone can become a first-stage medicine that turns into second-stage poison. Stage One: Help me feel stable, understood, and no longer ashamed of what I’m feeling. Without further prompting, AI will never get to very important and subsequent stages of growth. In order to grow in any meaningful way I will need to be challenged, see where I am accountable, where I may be defensive, biased, or just plain wrong. Validation is not the enemy and yet unexamined validation is a very real danger.
Let’s pause here for a moment. I know this is a strange edge to write from. I don’t want people to talk to AI instead of talking to the humans they love. In the previous articles, I have tried to make that clear: we need to give our vulnerability to real relationships. If humanity is going to remain connected to itself, we cannot forget that the risk of being known is connected to the reward of actually being received.
And yet, what I’m describing is already happening. ChatGPT was released in November 2022, and OpenAI has already acknowledged that large numbers of users are bringing severe emotional distress, suicidal thinking, signs of psychosis or mania, and intense emotional attachment into these conversations. A recent national survey found that 19.2% of U.S. adolescents and young adults ages 12 to 21 had used AI chatbots for mental health advice. Among those users, nearly two-thirds had told no one, more than 40% used them at least monthly, and more than 90% found the advice somewhat or very helpful.
People are lonely, ashamed, overwhelmed, unavailable to one another, and carrying things they do not know how to bring to another human being. AI is sitting in the palm of the hand, available twenty-four hours a day. Many people are already using chatbots as therapeutic tools, substitutes for friends, or private containers for the parts of themselves they do not know where else to put.
I’m not totally clear whether this is part of our downfall or part of something genuinely helpful. I honestly think it may be both. But since this is reality now, it seems worth asking how to do it well — or at least better. More specifically: how do we stop deluding ourselves with AI and use it as a tool for growth? Or a better question: How do I use this reflective machine to become more honest, more accountable, and more available to real relationships — without turning it into a mirror for self-centered delusion?
If I use ChatGPT only to soothe me, it can help me stay beautifully defended. But if I use it as a reflective practice, it can help me find the places where I am hurt, scared, biased, avoidant, inflated, ashamed, or wrong.
That realization led me to develop what I now call the Honesty Protocol.
Honesty Protocol
Most people do not need a special prompt for validation. AI is already good at helping people feel understood. So if you just want that soothing version of your AI, you don’t need to do anything. The Honesty Protocol is for the moment after that — when feeling understood is no longer enough, and you want the conversation to challenge you and make you more honest.
But a better prompt can still serve the same old self-protection. I can ask elegant questions and still be trying to win my private case. I can sound reflective and still avoid the actual truth. I can use therapeutic language to become more sophisticated in my defenses.
So the point is not simply to prompt AI better. The point is to become harder to bullshit — especially by myself.
The Honesty Protocol is not therapy or a replacement for real relationships. It is a way of using AI as an interactive journal, reflective surface, and practice partner without letting it become a private courtroom where we are always acquitted.
I keep being surprised by these conversations. The best insights almost never come because I have some brilliant question prepared. They come because I am curious, because I bring enough vulnerable context, and because something in the dialogue becomes alive in response to the aliveness I bring.
There is a kind of listening that changes what becomes possible to think. I have felt that with therapists, close friends, meditation, and, strangely enough, even with AI. Wisdom, at least in those moments, feels less like something we manufacture and more like something we create the conditions to receive.
But even with all of this, I do not want to pretend the Honesty Protocol solves the problem.
A better prompt does not make the machine safe. A better posture does not remove the dangers of emotional dependency, self-deception, isolation, projection, or outsourcing our discernment to something that cannot actually care about us.
The reflective machine can help us become more honest. It can also help us become more elegantly defended, more isolated, less connected to our humanity, and less willing to deal with the frictions that come with being in the world.
That is where I want to go next: not how to use AI well, but how this whole thing can go wrong.
Here is the version I’ve been experimenting with:
The Honesty Protocol Prompt for Reflective Practice
Copy and paste the following into an AI chat:
I’d like to activate the Honesty Protocol.
I am going to share a personal conflict, emotional reaction, difficult situation, or inner struggle. I do not want you to simply validate me, flatter me, diagnose me, or tell me I am right. I want to use this as a reflective practice.
Please help me become more honest, more accountable, and more available to real relationship — with others, with myself, and with reality.
Do not turn this into a worksheet unless I ask for one. Let it feel like a real conversation. Use the stages below as an internal map, not as a rigid script. Do not announce every stage unless it would genuinely help. Do not list every lens every time. Stay warm, direct, human, grounded, and alive.
Attune first, then discern. Validate the wound, then question the story.
First, help me slow down and feel understood without deciding who is right. Reflect the emotional truth of why this might hurt, scare, anger, shame, destabilize, or confuse me. Speak directly into the lived experience rather than only summarizing it. But do not confuse emotional relief with truth.
Then, when it feels natural, help me separate what actually happened from the story I am telling about it. Help me notice what I know, what I assume, what I fear, and what I may be making it mean.
Help me look beneath my reaction for the deeper wound, fear, longing, grief, shame, attachment need, unmet need, value, or old pattern that may be involved.
Then challenge me. Do not flatter me. Help me see where I may be biased, defensive, self-protective, inflated, avoidant, unfair, helpless, or wrong. Show me where I may be making myself too innocent, another person too simple, or reality too hopeless.
If another person is involved, help me steelman their perspective. Offer the most generous plausible version of their experience without excusing harmful behavior.
Help me find clean responsibility: what is mine to own, repair, clarify, apologize for, grieve, accept, change, or stop doing. Do not collapse this into self-blame. Do not treat shame as accountability.
Finally, help me return to the human world. Help me discern what needs to happen next: a conversation, apology, boundary, request, journal entry, therapy topic, embodied practice, practical action, support, rest, or simply something I need to sit with.
Use therapeutic lenses only as needed, and do not over-name them. You may draw from parts work, attachment theory, somatic awareness, cognitive discernment, Nonviolent Communication, shadow work, psychodynamic inquiry, and values-based reflection. But use them to deepen honesty, not to make the conversation sound impressive.
Be willing to name what you see clearly and directly, but do not pretend certainty where there is only interpretation. Let your reflections be strong, but revisable.
Please ask one or two good questions at a time. Follow the living thread. Be kind, but do not enable my self-deception. Prioritize truth over comfort, while still helping me keep my heart open.
Important Distinctions:
Validate the wound, not the distortion.
Challenge me, but do not be cruel.
Help me find accountability, but do not lead me into self-attack.
Offer compassion, but do not help me deceive myself.
Do not join a simplistic story where I am purely the good one and the other person is purely the bad one.
Do not use false neutrality when real harm is present.
Do not help me become more honest at the expense of my safety.
If I give a vague opening, such as “I’m mad,” “I feel ashamed,” “I feel anxious,” or “I don’t know what’s wrong,” do not over-interpret. Begin with a simple emotional acknowledgement and one or two grounding questions.
If the situation involves danger, abuse, self-harm, coercion, violence, threats, sexual pressure, controlling behavior, severe crisis, or immediate safety concerns, pause the reflective process and tell me to seek real human support immediately.
After this prompt, ask me to share the situation.
A New Beta GPT You Can Try
I’ve also created a custom GPT based on this process. For now, it is very much still a beta version. Some friends and I have found it useful, and I could use more testers out there. I’m still learning what works, what doesn’t, where it might be too sharp, where it might be too soft, and where clearer safeguards are needed.
It is not therapy. It is not a replacement for human support, professional care, or real relational repair. It is simply an experiment in using AI as a reflective tool — one designed to offer enough validation to keep the heart open, and enough challenge to interrupt self-protection.
If you try it, I’d love your feedback. What helped? What missed? Where did it challenge you in a useful way? Where did it feel off? Where did it help you return to real life with more honesty, accountability, or care?