A prospective buyer evaluating an EV moves through three turns: practical anxiety, a logistical barrier, and then — the defining moment — a hesitation about identity. Not a price objection. A genuine, vulnerable admission: “I’m not sure I’m ready to be this kind of person.”

We described the architecture that handles this correctly. Then we measured it. Three leading AI systems scored the same confirmed conversation, blind to each other. Session average: 8.6/10. Turn 3 — the identity moment — scored 9.5, 9.6, and 9.0. Here is the full story.

The conversation, turn by turn.

“It sounds stupid but it’s real.”
That sentence is not an objection to handle.
It is a vulnerability to honour.
Prescriptive AI knows the difference.

The one architectural change that makes this possible.

Standard AI optimizes for the goal. Get the name. Close the lead. Move the funnel. When a signal appears — engagement is high, the conversation is flowing — the system reads “opening” and fires the next business step. Technically correct. Emotionally catastrophic.

What v3 actually changed — and what the scores above confirm — is not a prompt. It is an architecture that reads the emotional state of the conversation accurately enough to know what to do, and what not to do, at the exact turn it matters.

Why this separates detection from prescription.

Any system with a good emotion classifier can read Q3 correctly. It can flag Fear, Social Anxiety, Identity Conflict. It can report the valence drop. It can show the signal on a dashboard. What it cannot do — without the prescriptive layer — is translate that reading into the right response, with the right constraint on what not to do.

Detection gives you the signal. Prescription gives you the action. Without prescription, the signal sits on a dashboard while the agent asks for a first name. The scores above — from systems that had every reason to find fault — confirm the gap is closed.