How OkayHonee actually works.
You type what your partner said or did. You get back a card. That's the whole product — but there's a specific structure to the card, and knowing what each piece is for makes it a lot more useful than skimming for the joke.
Short version: one honest translation, two or three ranked guesses, and one thing to actually say. Everything below is just the detail.
What's in a decode
Six parts, always in the same order.
1. The opener
One or two sentences reacting to what you typed. It's there to make you laugh before you get defensive, because nobody defensive reads past the first line. The joke lands on the situation and on the person who has to be talked to — the coasting husband in HER, the wiring and the gap in HIM. Never on the person reading it. That's a rule, not a stylistic preference.
2. What they probably meant
The translation. One sentence, usually shaped like an equation: the thing they said = the thing underneath it. If you only read one line on the card, read this one.
3. Two or three readings
Because a single confident answer would be a lie. Each reading is tagged:
- Likely — the reading that fits most situations like yours. Start here.
- Maybe — the honest alternative. Usually the “it's genuinely not about you” one, which is right more often than people expect.
- Wildcard — the long shot. Sometimes it's the funny one. Sometimes it's the one where you're overthinking a nice moment and about to ruin it.
Each comes with a one-line why, so you can judge whether it fits your actual situation.
4. Your move
The part that matters. One concrete next step, with words you can use — not “communicate openly,” which is advice that has never once helped anybody at 9pm on a Tuesday. It's deliberately ONE move, not a stack of tactics, because a plan you'll actually execute beats a clever plan you won't.
5. The closer
One line at the end where the joke stops. This is usually the truest sentence on the card.
6. A percentage
See below, because it deserves its own section.
What “91% sure” actually means
It means vibes. That's not a joke we're hiding behind — hover the number in the app and the tooltip says “purely vibes-based confidence,” because pretending otherwise would be worse.
There is no study behind it. Nobody surveyed your partner. It is a language model's read on how unambiguous your situation is, and the only genuinely useful thing about it is the direction:
High numbers mean the phrase is a well-worn one with a fairly reliable meaning. Low numbers mean it's a coin flip and you should lean on the “Maybe” reading harder. When the safety path fires there's no percentage at all — the whole card is replaced. That's the tell.
Treat it like a weather forecast. Useful for deciding whether to bring a jacket. Not evidence in a trial.
HER and HIM are genuinely different
The toggle isn't a pronoun swap. The two modes carry different assumptions, because the problem is not symmetrical.
Decode HER assumes the subtext is deeper than it looks — and that the person asking has probably been coasting. “I'm fine” is doing a lot of work, and he missed the thing.
Decode HIM assumes the subtext is shallower than it feels. He usually meant the four words he said, and the twelfth layer was one she built at 1am. The job there is to talk her down — while still saying it straight when he genuinely has checked out.
The theme changes too, which is either a nice touch or an unnecessary flex depending on how you feel about that sort of thing.
Yes, it's AI. Here's the honest version.
Every decode is generated by a large language model against a written persona — not pulled from a database of pre-written answers, and not reviewed by a human before you see it. The example cards on the homepage are stored responses so the demo is instant and consistent; anything you actually type goes to the model.
Which means it can be wrong. It is a plausible reading of a situation it knows exactly one sentence about, and it has never met your partner. It doesn't know about last Thursday. Weigh it accordingly.
What isn't left to the model: the safety path. A message that points at danger is caught before the model is even called, and the crisis resources you get are hard-coded and verified rather than generated — because a hallucinated hotline number is the single worst thing this site could produce. More on that here.
What happens to what you type
Plainly: your messages are stored on our server — including when you're not signed in — so the conversation works, so the thing can be improved, and so abuse can be caught. That's true of the anonymous five-a-day too. If that matters to you, it should, and the Privacy Policy is the binding version of this paragraph.
What we don't do: publish it. Decodes are private unless you explicitly share one, and shared pages are set to noindex so they never turn up in search results. We don't publish user submissions as examples — everything you see on this site was written by us.
Signed in, your conversations are kept and titled so you can come back to them. Signed out, it's one question at a time and each new one wipes the last. Goldfish memory. Just like us.
What it will never tell you to do
Not a tone thing. These are hard rules in the prompt itself.
- Lie, manipulate, guilt-trip, or keep score.
- “Win” the argument. There is no winning; there's a person you live with.
- Snoop through a phone, run the silent treatment, play games, or issue an ultimatum.
- Joke about violence, self-harm, abuse, or threats — in either direction.
- Punch at her, at him, or at any group. The joke lands on the person who needed a website to decode a text. That's you. That's the bit.
And when a message stops being funny, the comedy drops entirely — no punchline, no “your move,” no percentage. Just a straight answer and somewhere real to go.
No account needed to start.