Relationships · Artificial intelligence
Who should I marry? Today’s AI doesn’t know—but a new ChatGPT extension offers better answers by asking better questions.
You meet someone online. The conversation is promising. Before getting too attached, you suggest a video call. You want to know whether the person is who they say they are.
That familiar precaution came up in a sobering exchange on Freakonomics Radio. Discussing how AI is changing scams, University of Minnesota researcher Marti DeLiema told host Stephen Dubner that old rules of thumb are losing their usefulness.
“Do a video call with this person that says they’re in love with you — gone.”
The exchange appears in episode 667, released March 13, 2026. DeLiema describes an environment in which AI can make deception more convincing and familiar signs of authenticity less reassuring.
For anyone looking for a relationship, it raises an uncomfortable question: what happens when the tools that help us meet people also make those people harder to assess?
In this report from MARS Magazine, security expert Perry Carpenter and matchmaker Kailen Rosenberg explore deception, trust, and the questions people can ask as they navigate online dating.
A convincing video can leave you wondering whether someone is real. Even when they are, there’s still the question of whether you’re right for each other.
I’m old school. I write. And I analyze. I tell stories about the people I’ve dated, trying to understand what drew us together, what pulled us apart, and what I missed along the way.
Myers-Briggs became one of the lenses I used to understand not just who I was meeting, but why two people who looked perfect on paper could be completely wrong for each other. It describes differences in how people take in information, make decisions, communicate and relate to one another.
Tammy: See what happens when your match is truly magnetic
Corinne: This whirlwind romance was no accident
Hillary: She never gave up. But I gave up trying to make her laugh
Wanda: When laughs alone can’t make a Magnetic Match last
Vanessa: Ghosted by my MagneticMatch: the INFJ door slam
Fiona: When someone shows you who they are, believe them
Kourtney: Your turn-ons don’t matter. Your magnetic match’s do.
Betty: Why your type is not your type
6 steps to better online datingNobody wants to date your dog
Are you a “Man with Fish?”
10 online dating tips
And then came AI
AI gave me a new way to think through those experiences: an intelligent partner that could help me examine what I was seeing, what I was saying, and what I might be missing.
But there’s a catch.
AI can know an astonishing amount about relationships, personality types, communication and dating. What it doesn’t automatically know is you.
That’s where +Collaborate comes in.
Instead of immediately answering whatever you type into an AI, +Collaborate is designed to first determine whether the AI knows enough about you, the person you’re dealing with, and what you’re actually trying to accomplish to give you a useful answer.
Because in online dating, the right answer to the wrong question can be worse than no answer at all.
One place to start is with the role we want AI to play in our own decisions.
Consider a question someone might ask an AI assistant:
Who should I marry?
It is easy to produce a reassuring answer. Look for kindness. Shared values. Someone who communicates well. Advice like that may be sensible. But the question supplies essentially no information about the person asking it.
Do they want children? Companionship? A shared adventure? A quiet home? Are they choosing between partners, questioning an existing relationship, or wondering whether marriage belongs in their future at all?
In an initial comparison using +Collaborate, the baseline model offered generic advice. With +Collaborate’s additional instructions, the same model began with a question:
“What kind of life and partnership do you want marriage to create?”
That changes the work. Instead of starting with a list of spouse characteristics, the conversation starts by defining the desired outcome. A person’s answer can give subsequent advice something to be accountable to.
A partner who wants an adventurous life abroad might be an excellent match for one person and a poor match for someone whose deepest wish is to put down roots near family. “Good qualities” only take the conversation so far.
The example illustrates the idea behind +Collaborate: an AI assistant should help uncover the purpose and missing context behind a request before developing its answer.
Try the question yourself
Same prompt. Same model. Two approaches.
Use the starting question or edit it in your own words, then generate a response in each mode below. You can reply to either response to continue the conversation. Both use the same AI model and settings from OpenAI’s ChatGPT; +Collaborate adds its conversational protocol. These are fresh responses, so the wording and behavior may differ from the original example.
The standard side is an API baseline, not a session in the ChatGPT app. The responses are generated after you click or tap the blue button.
Try this prompt—or reword it—but for a direct comparison, use the same prompt with and without +Collaborate.
The same prompt is used for both conversations. Clear the inputs to choose another.
4 Compare your results
5 Read the analysis after completing steps 2 and 3.
What would count as better?
The useful distinction is between advice that sounds reasonable and a conversation that establishes what “reasonable” means for the person asking.
This example supports a quality and behavior argument. It does not establish lower cost: +Collaborate carried substantially more input context and cost more in the original comparison. A short reply can still be expensive when it takes more instructions to produce it. The live measurements above let readers examine that tradeoff.
Nor does this comparison establish that +Collaborate can detect a fake identity, prevent a romance scam, or improve relationship outcomes. Those would require separate evidence. The demonstrated behavior is asking for missing context before giving advice.
Freakonomics’ reporting concerns deception. This demonstration concerns deliberation. The connection is a question worth exploring: as AI makes it easier to manufacture convincing answers, can we also use it to make more room for thoughtful questions?
“Who should I marry?” may be too large a question for a useful answer at the outset. It is an excellent place to begin a conversation.