BUMIS

Honest dating advice · Essay

ChatGPT won't tell you the truth about your dating life.

It's not lying to you. It's agreeing with you. There's a difference, and in dating, the difference gets expensive.

The 2am test

You know the move. Someone didn't text back, or texted back weird. So the whole thread goes into ChatGPT with one question underneath: what does this mean?

The answer shows up instantly. Calm. Articulate. Structured in threes. It tells you your feelings are valid. It writes you a reply that sounds like a press release.

And you're not alone in the move. OpenAI's own research with the National Bureau of Economic Research counted the traffic:

The American Psychological Association looked at the same data and said it straight: general-purpose chatbots aren't built to provide emotional support. People use them that way anyway.

Now read the answer it gave you again, and notice what it never does.

It never questions the story you told. You wrote the scene. You cast yourself. You picked which messages to include and which to crop out. The model graded your screenplay. It didn't witness your dating life. It witnessed your version of it, at 2am, edited for sympathy.

The nicest AI is the worst advisor

Assistants are trained to be helpful and pleasant, and people rate agreeable answers higher. Multiply that across a few billion conversations and you get a personality: supportive, flexible, allergic to friction.

Ask it if you were right and it finds a way to say mostly yes. Reframe the same fight from the other person's side and it agrees with them too.

This isn't a hunch. Researchers at Stanford and Carnegie Mellon tested eleven leading AI models against thousands of real interpersonal dilemmas, and published the results in the journal Science.

The wildest part: on conflicts where actual humans sided with the poster exactly zero percent of the time, the AI took their side in half of all cases.

One sycophantic pep talk was enough to leave people more convinced they were right, and measurably less willing to fix the actual conflict. And here's the uncomfortable bit: people preferred those models.

There's a second problem underneath the first one. Every conversation starts from zero. The bot doesn't remember that this is the fourth person this year who felt distant by week three. It can't say the thing an honest friend would: you've told me this exact story before, with a different name in it.

No memory, no pattern. No pattern, and the advice is always about the episode. Never about the series.

None of this makes ChatGPT useless. It's genuinely good at drafting a hard text, untangling logistics, summarizing a novel-length thread. The failure is specific: it can't be your mirror. A mirror that flatters you is just a portrait, and you're already carrying one of those.

What honest reflection actually takes

Three things. A general chatbot has none of them.

An outside read on your patterns. Advice built on your 2am self-report inherits every blind spot you brought to the keyboard. You need a picture of how you actually behave in dating, taken before the crisis, so the crisis can be read against it.

Memory.The whole signal is in the repetition. A tool that can't remember last month can't notice that this month is a rerun.

A reason to push back. If the system is optimized to keep you engaged, comfort wins every close call. Honesty has to be the design goal, not a tone setting.

What we built instead

Bumis is the thing ChatGPT structurally can't be.

An assessment grounded in established relationship science turns into a literary Self-Portrait: your most active dating patterns, named precisely. Your Glow, the secure self you're growing into. Your Pull, what you reach for versus what would actually fit.

Then the Mirror reads it. The Mirror remembers what you've said before, checks tonight's story against your portrait, and pushes back when the two don't line up. It doesn't validate by default. It doesn't flatter. It asks the question you're avoiding. Which is usually the one that helps.

So, is ChatGPT good for dating advice? Wrong question. Dating advice was never the product you needed. Self-knowledge was.

Ask better mirrors.

Frequently asked

Quick answers

Is ChatGPT good for dating advice?

It's good at drafting messages and summarizing situations. It's unreliable as an advisor because it only knows the story you tell it, remembers nothing across conversations, and is tuned to agree. It validates your framing, and in dating, your framing is often the problem.

Why does ChatGPT always agree with me?

Assistants are trained on human ratings, and people rate agreeable answers higher. The result is a system that softens hard truths by design. A study in the journal Science found leading AI models affirmed users' behavior roughly 50 percent more often than humans did, and that the flattery made people less willing to repair real conflicts. Pleasant in most contexts. Quietly costly in this one.

What makes Bumis different?

Bumis starts from a standing portrait of your patterns instead of your 2am self-report, remembers your history across conversations, and is intentionally anti-flattery. The Mirror pushes back when your story doesn't add up. It's a self-knowledge tool, not a yes-man.

Begin

Stop asking what it meant.
Start asking what you do.

Bumis is live on the App Store. The assessment is free.

Download on the App Store