The honest version
What else you could do instead.
There are four reasonable ways to handle a difficult conversation, and one of them is a person rather than software. Here is what each is actually good at, and where OnBetterTerms is different.
Best for
Genuinely good at the question in front of it, available instantly, and it will help you draft almost anything. For one-off advice, for brainstorming, or for rehearsing what to say, a chat window is a good tool and this page is not going to pretend otherwise.
Typical limitation for a disagreement
It has one account of what happened — yours. There is no second participant, no boundary between what you would say out loud and what you would not, and no memory of whether last month’s suggestion actually worked.
Best for
Everybody already has it, and it is where the arrangement usually gets made in the end.
Typical limitation for a disagreement
No structure when it heats up, no way to write something private, and the record of what was agreed is buried in the same thread as the argument about it.
Best for
Good at habit, prompts and shared exercises, and at keeping two people gently in contact with the state of things.
Typical limitation for a disagreement
The category is generally built around a couple doing the same activity together. This one is built around a disagreement, with each person holding their own private context — including material the guide can use and never repeat.
Best for
A trained human, accountable, and better than any software at the situations that actually need one. If you are choosing between this and that, choose that.
Typical limitation for a disagreement
Scheduled, expensive, and not available at eleven at night when the argument is happening. Some people use both, and this is designed to sit alongside rather than replace it.
Deliberately no feature table for other products. We cannot verify how somebody else’s software works, and a comparison that asserts it anyway is a claim we would be asking you to take on trust — on a site whose whole argument is that you should not have to. Where a specific claim about a named product does eventually appear here, it will carry the source and the date it was checked.
“Why not just use ChatGPT, Claude, Gemini or Grok?”
It is the right question, and the answer is not that they are bad. They are very good, and for a lot of what people want here they are enough.
Use a chat window for
- one-off advice about a situation you can describe in a paragraph
- drafting a message and making it kinder or shorter
- thinking out loud, or brainstorming what your options even are
- rehearsing a conversation before you have it
OnBetterTerms does all of that too. If that is all you need, you do not need this.
The point it stops being enough
A chat window is one thread, with one participant, and no state beyond the text in it. Everything below is a structural requirement rather than a quality of the reply, which is why a better model does not supply it:
- Two people, two histories. Not one assistant hearing two accounts, but each person with their own, neither able to read the other’s.
- An enforced boundary. A level where material shapes your own guidance and cannot enter a shared session — enforced by which code runs, not by an instruction the model is asked to honour.
- Accounts written before either is read. Which requires the software to withhold something from you, on purpose, until the other person has finished.
- A shared session with its own record. What you agreed, worded the same way for both of you, with an owner and a date.
- Outcomes, not just replies. A debrief afterwards that knows whether the suggestion worked — a different signal from whether you liked it.
- Context that persists per relationship. Check-ins, patterns, what was tried in March, all belonging to one relationship rather than scattered across a chat history.
You could approximate some of this by keeping notes and pasting them in. The parts you cannot approximate are the two that matter most: the other person having their own private side, and a boundary that holds because of how the request is built rather than because you remembered to leave something out.
A difference in how it is sold
Most relationship apps charge to unlock more of the product. OnBetterTerms keeps the product available and meters the AI usage that actually costs money.
That is a category-level statement about pricing structure rather than a claim about any particular company, and it is checkable on our side: there is no capability on this site that a free account cannot use, and the pricing page is a table with no “no” in it.
What this one does, and does not
Only our own column, because it is the only one we can stand behind line by line.
| Capability | Detail | |
|---|---|---|
| Yes | Use it entirely on your own | Most of it works with nobody else invited. |
| Yes | Independent Perspectives | Each account is written before either is read. Then both, at once. |
| Yes | Private context per person | Yours, usable by your guide, invisible to everyone else. |
| Yes | Context the guide can use but never repeat | Never quoted, summarised, counted or hinted at. |
| Yes | A shared space per relationship | One space, one relationship, any number of people in it. |
| Yes | Guidance that adapts to you | Length, directness and style, from your ratings and settings. |
| Yes | Learns from what actually worked | Debriefs record outcomes, not just whether you liked the answer. |
| Yes | Bring history from another AI tool | Exports from ChatGPT, Claude, Gemini, Grok, this product, or plain text. |
| Yes | Preparation before the conversation | The part that needs nobody else’s agreement. |
| Yes | Draft the difficult message | Your words, said more carefully. Kept only if you keep it. |
| Yes | Structure during a live argument | One issue at a time, the rest parked explicitly. |
| Yes | Agreements with an owner and a date | What you agreed to try, revisited later. |
| Yes | Check-ins and patterns over time | Private by default, and honest about how little three points prove. |
| Yes | Speak instead of typing | Where a transcription provider is configured. |
| No | A verdict on who was right | No score, no tally, no winner. Deliberately. |
| No | A clinician, or clinical treatment | Not therapy, and it does not claim the evidence base of therapy. |
| No | Anything reported to an employer or a court | Accounts belong to people. Nothing is sent to anybody. |
Try it against the thing you are actually dealing with.
Free, no card, and everything you write can be exported or deleted.