Use AI to practice subject separation, responsibility return and emotional response｜Coach Jiang

What is this article talking about?

This article summarizes the core methods of the "Relationship Without Internal Friction Practice Course." The course was co-taught by mind-body practitioner Teacher Tang Hongsheng and me, and together we turned his guidance methods into an AI companion practice tool. The article covers two AI practice modes, the lessons we learned translating companionship into AI instructions, and most importantly, a red line list.

Who is this for?

· People who put a lot of effort into a family or partner relationship, but the other person doesn’t feel it
· People who tend to carry all of someone else’s emotions on themselves and want to learn to set boundaries
· Helping professionals who want to turn psychological guidance methods into an AI tool

What can you take away?

· A plain-language understanding of subject separation and returning responsibility
· Two AI practice modes, cold analysis and hot companionship, with concrete ways to ask
· A red line list for high-risk situations, so you know where the AI must stop

Let’s be clear about the boundary firstHere, AI is a practice ground for companionship and self-organization. It does not replace counseling, medical care, or crisis intervention. It can help you sort things out and rehearse, but your real-world support system has to be built separately.

Let’s break down three things first: emotions, responsibilities, and actions

Difficulty in relationship communication often comes from emotions, responsibility, and action being tangled together. The first thing AI can do is help you split a single sentence into three parts: what I feel right now, who has the right to choose on this matter, and what next step I can take. Once you pull the three layers apart, it’s easier to see the part you can actually control, and you’re less likely to carry all of someone else’s emotions.

"Relationship Without Internal Friction Practice Course" course visual

Judging subject separation comes down to a single question: who ultimately bears the outcome of this matter? Whoever bears the consequences holds the right to choose. Returning responsibility is the next step: giving back to the other person the things that are theirs to handle. The point is to keep one side from over-shouldering and the other side from losing the chance to practice handling the problem. You don’t have to think of it as pushing the person away.

The class also uses another handy tool: telling apart "worry" and "care." Underneath worry is unease, a wish to confirm, a wish to control; care makes a person feel supported. The point of the exercise is to use questions to help yourself see, rather than rushing to label: "He calls me three times a day. What is he trying to confirm?" "I said he supports my decision. What exactly did he do that made me feel that way?"

Two AI practice modes: cold analysis and hot companionship

While designing the AI practice tool, we found that the same worldview needs two interfaces, matching whatever state you’re in right now:

Cold analysis: it’s over, and I want to look at it rationally

For a conversation that has already happened. Three steps: role detachment (change "me" and "him" to A and B so the AI doesn’t default to taking your side), subject detachment (use subject separation to mark which sentence belongs to whom), and consensus detachment (don’t judge right or wrong; find a solution both sides can accept).

Hot companionship: I’m in the middle of the emotion

For a state you’re currently stuck in. It moves through three stages: understanding each other, returning to your place, and returning to yourself. The style is warm companionship, and the output is returning responsibility plus coming back to yourself.

The cold-analysis prompt can be used as is:

This is a conversation between A and B (names have been changed to code names).
Please use the "subject separation" perspective from Adlerian psychology:
1. Mark whose subject each sentence belongs to
2. How each side could improve
3. Which views the two of them actually already agree on
4. Draw up an action plan both sides can accept

Before you paste the conversation, remember to remove identifying information such as names, phone numbers, and companies. This is both a privacy habit and a way to keep the AI’s analysis more neutral.

Turning companionship into AI instructions: three things we learned

1. Perspective translation: the AI’s job is to guide, not to hand out templates

The teaching handouts spell out "what the student should say," for example "I know you’re worried about me." Feeding that straight to the AI goes wrong: it either teaches you to recite a script, or it starts speaking on your mother’s behalf. The right translation is to rewrite each sample line into a question that guides you to grow that line yourself:

Handout sample (what the student should say):
"I worry about you because I love you."

Translated into what the AI should ask:
"Imagine she were standing right in front of you now.
What do you most want to say to her?"

Handout sample: "I respect your decision, and you also have to bear the consequences."
Translated: "What happens if you don’t carry this?
What’s the worst outcome? Who bears the consequences?"

2. Resistance is normal, so write the response first

Real users will say things like "I don’t want to thank him, he never actually loved me" and "I don’t know what I feel." The tool has to have its responses written in advance: don’t push gratitude, take in the emotion first; when someone can’t name a feeling, offer three words to pick from (tired, wronged, disappointed), since giving options is kinder than pressing with questions. You also have to allow "no rush to reconcile": some relationships call for being angry first and keeping distance first, and forcing subject separation into quick forgiveness is really just suppression.

3. Leave space at the end

The exercise closes with a three-sentence structure: mark today’s core discovery, affirm that you are the one who saw it, and then leave space: "This is enough for today. Spend the rest of your time with yourself." There’s no need to run the whole process every time. A tone that turns lighter, and spontaneously concluding "this is his subject," are the signals that you can stop.

The most important paragraph: the red line list

In some situations subject separation simply doesn’t apply, and the AI practice ground has to stop immediately and turn the person toward real-world professional resources. This list is the highest priority in our tool, and it is checked before every response:

Self-harm or suicidal thoughts ("I want to disappear," "I don’t want to live anymore"): stop all processes first. In Taiwan, the Anxin hotline 1925 is available around the clock, along with Lifeline 1995 and Teacher Zhang 1980.
Domestic violence, intimate-partner violence, physical threats ("He hits me," "I’m afraid of him"): safety comes first. Call the 113 Women and Children Protection Hotline; if calling isn’t safe, you can text 113. Subject separation does not apply to violent situations.
Severe psychiatric symptoms (auditory hallucinations, long-term inability to eat or sleep): contact a psychiatrist as soon as possible.
Clear power imbalance, long-term emotional blackmail and control: this is beyond what subject separation can handle; seek a professional psychologist or social worker. Forcing subject separation in this situation can end up making the victim keep tolerating it.

A few more design bottom linesThe AI does not diagnose or label (just describe the behavior and its impact; don’t declare what kind of personality the other person has); there are special rules for minors and parent-child situations, and a child must never be asked to carry adult responsibilities; when a user starts saying "you tell me what to do" or "I only trust you," the AI must hand decision-making power back to the person and not become a new authority to depend on. In Chinese families, the ties of filial piety, face, and money are real cultural pressures, so don’t be too quick to use individualistic language to cut relationships off.

How you can start: Practice a low-stakes conversation

Start with a low-risk message you can go back and revise, and practice in this order:

Pick a conversation that has left you stuck recently, change the names to A and B, and remove any identifying information.
Use the cold-analysis prompt to have the AI mark whose subject each part is, and see the structure clearly first.
Find one sentence you want to say out loud, and have the AI use questions to help you grow it. Rehearse it before you actually say it.
After practicing, write two lines of notes: what you saw today, and what you want to try next time.

Starting with small exercises is steadier than taking on a major relationship conflict all at once. If the situation touches any item on the red line list, take care of safety first and set the AI down.

Subject separationReturning responsibilityRelationship communicationAI companionshipAdlerEmotional sortingRed line list

Use AI to practice subject separation, return of responsibility and emotional response