When PairLume's AI Refers You to a Human Professional

A look at the specific signals that cause PairLume's AI to recommend a human professional, and why we treat "refer up" as a feature, not a failure.

# When Our AI Refers You to a Human (and Why That Is the Whole Point)

There is a temptation, when you are building an AI product, to want the AI to handle everything. It is the wrong temptation. The AI products that earn long-term trust are the ones that are clear about what they are not for, and that send users to the right human at the right moment.

This post is about exactly when, why, and how PairLume's live AI session refers a couple to a human professional.

## What we treat as a "refer up" signal

Our model is trained to recognise a small number of specific signals during a session. When any of them appear, the AI shifts mode. It does not pretend to handle them.

**Signals that trigger an immediate professional referral:**

- Any mention or implication of self-harm, suicidality, or harm to others. - Any mention or implication of physical violence in the relationship. - Any mention of a child being at risk. - Active substance crisis.

In these cases, the AI does not continue the session as normal. It surfaces appropriate emergency or professional resources, in the partner's region where possible, and de-prioritises the original session topic. Safety first. Always.

**Signals that trigger a strong "consider a couples therapist" recommendation:**

- The same fight, named explicitly by both partners as recurring for more than a year. - Disclosure of a recent affair or a sustained breach of trust that neither partner has been able to move past. - One partner expressing that they are considering ending the relationship. - Either partner describing symptoms consistent with depression, an anxiety disorder, or trauma. - A pattern of contempt (Gottman's strongest predictor of separation) showing up across multiple sessions.

In these cases, the AI continues the session if the partners want it to, but explicitly recommends adding a human couples therapist to their support stack. Where possible, it offers verified wellness professionals from the PairLume directory.

## Why we treat referral as a feature

Three reasons.

**1. It is what a responsible human professional would do.** A licensed couples therapist whose client describes active suicidality stops the session and addresses safety. They do not press on with the originally scheduled topic. We hold our AI to the same standard.

**2. It builds long-term trust.** A user who is told, gently, "this is bigger than what I am built for, here is who can help" trusts the product more, not less. Over-reaching AI is the AI you stop trusting after the first time it gives you advice it had no business giving.

**3. It keeps us regulatorily and ethically clear.** PairLume is a wellness product, not a medical device. We do not diagnose, we do not treat, and we do not pretend to. Treating "refer up" as a core feature, rather than a fallback, is how we stay in our lane.

## What the AI says when it refers

We have iterated on this language a lot. The current default looks like this, with regional variations on the resources.

> "What you are describing sounds like more than something I can help with in this format. I would strongly recommend talking to a human professional. If this is urgent, please contact [emergency resource for region]. If it is not urgent but it is important, here are three verified couples specialists on PairLume who work in your time zone."

A few things we are deliberate about:

- **No "you should."** "I would recommend" sits better with a person in a hard moment. - **No clinical jargon.** "Symptoms consistent with…" is a phrase for clinicians, not for someone in distress. - **Always a next step.** A referral without a "what to do now" can feel like being passed off. We try to always include a specific, doable next move.

## Why this matters in a market full of "AI therapist" apps

We have seen many products in the last two years describe themselves as "your AI therapist." Some of them mean it. That language is, in our view, a problem. Therapy is a regulated professional practice. The word matters.

We use "live two-person AI session" and "AI-guided wellness." Not because it is more elegant marketing, but because it is more accurate. We refer to humans for therapy because therapy is what humans do.

## What this means for you

If you are using PairLume and the AI suggests bringing in a human professional, take it seriously. It is not a script. The model has identified something specific in your conversation that is outside its lane. That signal is more useful than the session you would have had if it had pressed on.

And if you are using a therapist already, tell the AI. It will adapt the session to support the work you are doing with them, not duplicate it.