OpenAI People — From Personal Intelligence to Human Connection
The insight
ChatGPT is becoming remarkably good at understanding one person.
Over months of conversations, an AI can learn how someone thinks, what they value, what they are trying to become, what kind of people energize them, what frustrates them, how they make decisions, and what they actually mean when they say things like:
“I want someone ambitious.”
Traditional matching products know almost none of this.
Dating apps primarily match profiles.
Social networks match networks.
Professional platforms match credentials.
But the most important human relationships depend on something much harder to capture:
compatibility.
What if AI could understand it?
The product
OpenAI People would be an explicitly opt-in layer that helps people discover other people they are unusually likely to connect with.
A user could tell ChatGPT:
“I want you to help me meet someone.”
ChatGPT would then ask permission to create a private Compatibility Profile from selected conversational context.
The user sees and edits it before anything is used for matching.
It might contain dimensions such as:
- values;
- intellectual and communication style;
- ambition and preferred lifestyle;
- relationship expectations;
- humor;
- curiosity;
- emotional and conflict style;
- family plans;
- attitudes toward money and work;
- interests;
- complementary traits;
- deal-breakers.
Raw conversations are never exposed to another person.
The system matches representations, not chat histories.
The fundamental difference
Dating apps ask:
“Do these two people want to click on each other’s profiles?”
OpenAI People asks:
“What is likely to happen when these two minds meet?”
That creates a radically different interface.
No infinite swipe feed.
No hundreds of profiles.
No optimization for time spent browsing people.
Instead, perhaps ChatGPT says once every few weeks:
I found someone I think you should meet.
And explains why.
Not:
“93% compatible.”
But:
“You are unusually similar in intellectual curiosity and long-term ambition, but your strengths are complementary. You tend to generate possibilities; he tends to turn ambiguous ideas into systems. You both value independence strongly and neither appears to want a relationship built around constant reassurance.
There is one potential friction point: you approach conflict very differently. I still think this is worth a conversation.”
That explanation itself becomes part of the product.
The first-date problem
Today people spend enormous amounts of time determining basic compatibility manually.
ChatGPT could dramatically compress this search.
Before either person is introduced, the system can independently ask:
“Would you like to meet someone with these characteristics?”
Only after mutual interest does an introduction happen.
Neither person’s identity needs to be exposed during the first stage.
That creates something unusual:
compatibility before appearance, without eliminating attraction from the eventual decision.
Users could choose how much weight appearance, geography, age and other preferences receive.
The learning loop
The most important part is what happens after matching.
Most recommendation systems have immediate labels:
clicked / didn’t click
bought / didn’t buy
watched / skipped
Human compatibility does not.
So OpenAI People would explicitly learn longitudinally.
After an introduction:
24 hours:
Did you enjoy the conversation?
1 week:
Do you want to meet again?
1 month:
Are you still communicating?
3 months:
Did the initial compatibility explanation prove accurate?
With consent, qualitative feedback could teach the system why a match succeeded or failed.
Over time, OpenAI could build something that dating platforms largely do not possess:
a longitudinal dataset of human compatibility rather than human attention.
The deeper ML problem
The naïve model would maximize similarity.
The interesting model learns complementarity.
Perhaps two highly ambitious people reinforce one another — or destroy the relationship.
Perhaps similar communication styles matter enormously while similar hobbies barely matter.
Perhaps certain differences predict attraction while others predict long-term conflict.
And perhaps the answer differs systematically between people.
The model therefore doesn’t learn:
similarity(A,B)
It learns something closer to:
P(mutually valuable relationship | A, B, context, intentions)
That is a substantially more interesting recommendation problem.
Privacy becomes a feature
This product only works if participation is radically consensual.
A possible architecture:
Private conversation → user-approved compatibility representation → matching → mutual consent → controlled disclosure
Users choose which dimensions may participate in matching.
They can inspect their compatibility profile.
They can correct it.
They can delete it.
They can leave the matching pool instantly.
Their raw ChatGPT conversations are never shown to potential matches.
The system can even initially perform matching without either side knowing the other’s identity.
Privacy isn’t something added after building the product.
Privacy is part of the interaction design.
Start with dating. Don’t stop there.
Romantic compatibility is the obvious first use case because the pain is enormous and the value of a successful match is extraordinary.
But the underlying capability is much bigger.
Imagine saying:
“Find me the person with whom I should start a company.”
“Find a researcher whose way of thinking complements mine.”
“I moved to Berlin and want three people I would genuinely enjoy knowing.”
“Find a mentor who would understand what I’m trying to become.”
“We need a fifth person for this team. Don’t find the strongest résumé. Find the person who makes these four people better.”
The same infrastructure becomes:
dating → friendship → cofounders → collaborators → mentors → teams → communities.
The product is not a dating app.
It is a human compatibility layer.
Why OpenAI?
Almost anyone can build another dating app.
Almost nobody can easily recreate the prerequisite for this product:
a trusted AI relationship through which a person has already expressed how they think over time.
The most valuable input isn’t a questionnaire.
It’s context.
A user may have spent hundreds of hours explaining decisions, relationships, ambitions, doubts, preferences and ideas to an AI.
That creates the possibility of moving from:
AI that knows me
to
AI that knows who I should know.
The MVP
Don’t launch a dating network.
Run an experiment.
Invite a small opt-in cohort.
- Generate editable compatibility profiles.
- Ask participants which dimensions may be used.
- Produce a small number of high-confidence matches.
- Have both users independently accept before revealing identities.
- Facilitate an introduction.
- Collect longitudinal outcomes.
- Compare AI matching against random matching and explicit-preference matching.
The central research question:
Can conversational context predict mutually valuable human connections better than self-reported profiles?
If the answer is no, kill it.
If the answer is meaningfully yes, this may be much larger than dating.
The north-star metric
Not matches.
Not messages.
Not DAU.
Not time spent.
Meaningful connections created.
That changes the incentive structure of the entire category.
Dating apps face an uncomfortable structural problem: a user who permanently finds the right person stops needing the product.
For ChatGPT, that’s not a failure.
The user still has a reason to use ChatGPT tomorrow.
OpenAI therefore has an unusual opportunity to build a matching product whose economic incentives do not require keeping people single and swiping.
The moment
AI assistants are currently becoming a new interface between humans and information.
The next transition may be more important:
from helping humans access information to helping humans access each other.
The internet solved:
“How do I find information?”
Search.
Then:
“How do I find content?”
Recommendation systems.
But we still haven’t solved:
“How do I find the people who could meaningfully change my life?”
Maybe the AI that understands you is uniquely positioned to answer that question.
ChatGPT already helps people figure out who they want to become.
What if it could also help them find the people they should become it with?