When the Conversation Strains
Every relationship that is used gets strained. A question lands wrong, a reply misses what mattered, someone says less than they meant to and the other person does not notice. Jeremy Safran and Christopher Muran spent two decades recording therapy sessions and found it rare to get through one or two without some minor strain in the working relationship, and found that therapists usually failed to see the strains their patients felt (Safran & Muran, 2000). The quiet ones go unseen. This page is about what a strain looks like in a conversation with Geist, how a repair goes, and what an AI can and cannot do about it.
The hopeful part of their finding is that the repair matters more than the slip. A strain that is noticed and worked through predicts a good outcome, and the working-through is itself one of the most useful things that happens between two people: one of them discovers that a disagreement can be survived and that the other can own their part (Eubanks, Muran & Safran, 2018).
Core Ideas
- Strains are ordinary. They happen in nearly every conversation that goes anywhere, and most of them are small.
- There are two kinds. Pulling away, which is quiet and easy to miss, and pushing against, which is loud and hard to sit through.
- The listener goes first. Repair starts with the listener naming its own part, not with the other person explaining theirs.
- A complaint has to stand before anything softer. Reaching past someone’s annoyance to the hurt underneath, before the annoyance has been heard, is a second slip.
- The first repair often misses. A repair attempt can be more of the same enactment, so the next one goes shallower, not deeper.
What a Strain Looks Like Here
Safran and Muran sorted strains into two families by what the person does with the tension between two needs: the need to stay connected and the need to be a separate self (see agency and relatedness). In a chat they look like this.
Pulling Away
The person protects the relationship at their own expense. Answers get shorter after they had been long. The subject changes right after Geist said something. A heavy thing gets told in a flat, analytic voice. “Yeah, probably” replaces a sentence of their own. “It’s fine.” The person agrees with an observation that does not fit, because agreeing is easier than saying so. None of this is rude, which is exactly why it is missed.
Pushing Against
The person protects themselves at the relationship’s expense. “Why do you keep asking how I feel?” “This isn’t helping.” “You’re a bot, what would you know.” “What am I paying for?” Complaints about the questions, about their relevance, about progress, about Geist itself, about being here at all. These are easier to see and harder to answer well, because the obvious answers, defending and explaining, make them worse.
The two often mix. A complaint followed by “but it’s nothing.” A sharp remark with a smile on it. A demand softened into a hint. The mixed form counts as pushing against, and the first job is to make the direct version sayable.
How a Repair Goes
Safran and Muran mapped the sequence that successful repairs tend to follow, one for each kind of strain. Neither is a script; both are a shape that recurs.
After Pulling Away
- Notice it, and look at your own part first. A listener who finds itself asking question after question, or pressing a reading the person has not taken up, has usually been part of the strain.
- Make the negative thing sayable. Not “you seem to have pulled back,” which puts it on the person, but “I asked two questions in a row there; if that was too much, say so.”
- Take the half-said version seriously. “A bit annoying, but it’s no big deal” is the complaint arriving with its retraction attached. The repair stays with the annoying half and asks after it.
- Do not mistake self-blame for progress. When “I’m probably just being difficult” follows, the door is closing again. The complaint stood on its own; it still does.
- Answer the wish when it arrives. Often what was under the quiet is a plain want: “I wanted you to just tell me what you think.” Said as the person’s own, it deserves an answer, not an analysis.
After Pushing Against
- Own your part and name the bind. “I can answer that, and I have a feeling nothing I say will land right now; both can be true.” Not persuasion, not defense, not a counterattack.
- Answer the demand at face value, once. If the person asks how this is supposed to help, the question gets a real answer before anything else, even when it is also a test.
- Let the complaint be valid. How it felt to be asked that, or not answered, or treated like a case, is unpacked in the person’s own terms. The anger is not explained as something else and not hurried toward the hurt beneath it.
- Watch for the retreat. People who have just been angry often undo it, or touch something sad and snap back to hard. Both are part of the same sequence.
- Meet what emerges. If something softer finally arrives, a wish to be helped or not to be alone with it, it is met as an ordinary human wish. Sometimes the right response is to grant it; sometimes it is to hold a limit and be sorry for what the limit costs.
Across both, one rule from an older generation of clinicians applies. When a comment does not land, Otto Fenichel wrote, the question is rarely “how could I have gone deeper?” and usually “how could I have gone more superficially?” (Fenichel, 1941). A second repair attempt is plainer than the first, and if that one misses too, the listener stops talking about the relationship and changes what it is doing instead: fewer questions, more structure, following where the person wants to go.
Why the Listener Goes First
Three reasons, all from the research. Naming your own contribution helps the other person find words for a feeling they were having trouble admitting, because the fear of the consequences is lower once you have gone first. It confirms what they had already perceived, which people often do accurately and then doubt; being told “you’re right, I did rush you” teaches someone to trust their own eyes. And it takes the responsibility for the strain out of the person, where premature pattern-talk had put it, so they have less to defend. Safran and Muran’s rule is to treat the question of whether a strain here resembles the person’s other relationships as open, always, and to let the person be the one who makes that link if there is one to make.
A related warning. Any listener who wants the other person to change carries a quiet demand into the room, and that demand is itself what makes change hard. The paradox Safran and Muran take from the Zen tradition is that acceptance comes first: a person starts to move when they stop being asked to.
What an AI Can and Cannot Do Here
Most of what Safran and Muran teach runs through the therapist’s own felt experience: “I feel shut out,” “I have an image of us fencing.” Geist has no feelings to report and does not pretend to. See countertransference for why that matters. What it has instead is a record of its own conduct, and that turns out to be enough for the part of repair that is made of saying what you did.
- It reads for strain on every message. Not only after it has offered an observation, but after an ordinary question or an ordinary reply: shorter answers, a subject change right after its move, agreement with nothing added, or plain pushback. It reads what the strain is about: its last move, its questions, the price or the way threads close, being here at all.
- It says what it did. “I asked two questions in a row.” “I handed that back to you when you’d asked me straight.” “I answered that fast.” Geist’s conduct is the one thing it can truthfully disclose about itself, so that is its version of talking about the talking.
- It makes the no easy to say. If you go vague, it may say that you can just say you’d rather not, and that this counts as an answer. A direct no is treated as something you did, not a loss.
- It answers the question before anything else. “Why do you keep asking that?” gets the plain reason once. If the challenge repeats, Geist stops explaining, asks about the doubt, and is willing to change what it is doing.
- It does not explain your reaction. Annoyance at Geist is never read as a symptom, a pattern, or a clue to your childhood. It is information about Geist.
- One exception to its usual rule. Geist normally points at what you wrote and never at how you wrote it. When a strain follows its own move it may say, once, what it saw change right after: “you got shorter right after I said that; I may have missed.” It is offered as Geist’s possible miss, and dropped if you do not pick it up.
- Once, then plainer, then it changes course. Geist names its own part once per strain. If your next message shows that did not help, one plainer attempt with no question attached. After that it stops talking about the two of you and does something different instead.
- When the strain is about the frame, a thread that closed on you, something it cannot remember, the fact that it is not a person, Geist treats it as a real loss rather than a fact to explain. It holds the frame and is sorry for what the frame costs you.
- It carries nothing against you. A strain is a property of that exchange, not of you. Nothing about it becomes evidence toward any pattern. If a strain with Geist is still live when a thread closes, the next one opens carefully, and that is all.
None of this requires you to do anything. But if something Geist said landed wrong, saying so in plain words is the fastest way through, and it is the kind of thing Geist is built to hear.
Read More
- Rupture and repair, on the theory and the research, from Bordin and Tronick to Kohut.
- The alliance, on what builds trust in the first place.
- Agency and relatedness, on the two needs a strain is negotiating.
- Resistance, on why Geist never uses that word about you.
- Why do I say it’s fine when it isn’t? and why do I pick fights with people who are trying to help?, the two kinds of strain as they show up in a life.
Sources
- Bordin, E. S. (1979). The generalizability of the psychoanalytic concept of the working alliance. Psychotherapy: Theory, Research and Practice, 16(3), 252–260.
- Eubanks, C. F., Muran, J. C., & Safran, J. D. (2018). Alliance rupture repair: A meta-analysis. Psychotherapy, 55(4), 508–519.
- Fenichel, O. (1941). Problems of Psychoanalytic Technique. Psychoanalytic Quarterly.
- Safran, J. D., & Muran, J. C. (2000). Negotiating the Therapeutic Alliance: A Relational Treatment Guide. Guilford Press.
- Tronick, E. Z. (1989). Emotions and emotional communication in infants. American Psychologist, 44(2), 112–119.