When both sides

In 2024, researchers published one of the first peer-reviewed studies on ChatGPT's ability to provide relationship advice. The findings were mixed. Users rated it highly on empathy and helpfulness. But the researchers flagged one consistent limitation: a poor assessment of risk.

In a relationship context, that gap matters more than it might seem.

The problem with balanced AI in unbalanced situations

Bridget is built around a core truth: in most relationship friction, both people are responding to something real. One person feels dismissed; the other feels misread. One person goes quiet; the other reads it as rejection. Neither is wrong — they're just standing in different places.

That framing works for the vast majority of relationship moments. But it has a hard limit.

When one partner is systematically afraid of the other — when "conflict" is actually a pattern of control, isolation, or harm — offering both perspectives isn't balanced. It's misleading. It suggests there's a legitimate "other side" to behaviour that doesn't have one.

General AI tools aren't built to catch this distinction. They're optimised to be helpful and balanced, which in most contexts is exactly right. But relationship contexts are where that optimisation breaks down. A tool that can't distinguish between friction and fear is a tool that can inadvertently validate the wrong thing.

What Bridget looks for

Bridget is trained to recognise certain patterns in the situations people describe: language that suggests fear rather than frustration, descriptions that involve control over movement, finances, or contact with others, situations where one partner describes walking on eggshells as a permanent state rather than a temporary one.

When those patterns appear, Bridget doesn't offer a dual perspective. It breaks frame.

Instead of "here's how both of you may have experienced this," Bridget acknowledges what the person has described, names what it might reflect, and points toward support — not a decode.

Why this is a design requirement, not a feature

This isn't an edge case or a safety disclaimer bolted on after the fact. It's hardcoded into how Bridget works. The both-sides frame only holds when both sides are operating without fear or coercion. When that condition isn't met, the frame itself is the problem.

General AI tools have no mechanism for this. They'll generate a thoughtful dual-perspective analysis of almost any situation you describe — because that's what their training pushes toward. They don't have a way to recognise that balance is sometimes the wrong response.

What Bridget will say

If you describe a situation that contains patterns of fear, control, or coercion, Bridget won't decode it as a miscommunication. It will tell you directly that what you've described sounds like more than friction, and it will suggest reaching out to someone who can actually help — a professional, a trusted person, or a crisis resource.

This isn't Bridget being unhelpful. It's Bridget being honest about what it can and can't do — and about what the situation actually calls for.

Not every difficult moment needs to be understood from both sides. Some moments need something else entirely.