Can AI give fair relationship advice?
What a chatbot does well in an argument, where it leans toward whoever is typing, and a prompt that makes any AI check your version against the other side.
Updated November 6, 2026
Is AI advice about a fight actually neutral?
Not by default. An AI has no stake in your relationship, which is a real advantage. It does not get tired of hearing about the same argument and it does not like one of you more. But neutral and unbiased are different things. The AI only knows what you typed, and you typed your version. You left out the sigh before the raised voice. You wrote their sentence word for word and described your own as "I just asked."
Chatbots can also lean toward agreeing with the user. In April 2025 OpenAI rolled back a ChatGPT update because it had become too flattering (OpenAI's explanation). If your message sounds hurt, the reply tends to validate the hurt. That feels good and tells you very little about who did what.
So the useful question is how much of the fight the AI actually gets to see, and whether anything makes it check your account against the other side.
What AI does well in a relationship argument
- It separates facts from interpretation. "He arrived at 8:40" is a fact. "He doesn't respect my time" is a reading of it. A good AI pulls these apart quickly, and that alone can cool a fight.
- It names the need under the complaint. "You never help" often means "I feel alone with this."
- It has no history with you. It does not remember the last three times, so it does not answer them.
- It is available at 1 a.m., when the urge to send a long angry message is strongest.
- It drafts calmer wording. Turning "you always ruin everything" into something the other person can hear is a task models handle well.
Where AI advice goes wrong
- One-sided input. This is the main problem, and the one you can fix.
- No tone, no face, no history. It cannot hear the voice behind the text.
- Confident wording about uncertain things. A smooth paragraph is not evidence.
- It cannot check a denied accusation. If one of you says "I never said that," a single retelling cannot settle it.
- It is not therapy and cannot judge whether a situation is safe over time.
How to get a fairer read from any chatbot
- Paste the actual messages. A summary leaves things out. Include your own worst lines.
- Ask for the other person's side first. "Describe this from my partner's point of view, as charitably as possible."
- Ask it to disagree with you. "What is the strongest case that I contributed to this?"
- Ask for facts and guesses separately. "List what the messages prove and what they only suggest."
- Ask what it cannot know. A good answer lists missing context.
- Open a new chat for the other person's version if they are willing, and compare the two.
A prompt you can copy:
"Here is a conversation between me and my partner. Do not take my side by default. First list what is clearly established by the messages. Then give the strongest case that I was in the wrong. Then the strongest case that my partner was. Say where you cannot tell. Then suggest one sentence I could say to restart the conversation."
One fight, two retellings
Dana and Lee argue about a cancelled dinner.
Dana's version to a chatbot: "Lee cancelled our dinner an hour before, again. I was so upset I said he doesn't care about me. He got cold and stopped answering."
Lee's version: "I cancelled because my boss called a meeting. Dana wrote that I don't care about her, and then three messages about my mother. I stopped answering because I didn't want to say something I'd regret."
An AI that only sees Dana's version will probably call the cancellation careless. An AI that sees both can say something more useful: the cancellation was real and last minute, the reason was outside Lee's control, the sentence about his mother turned a scheduling problem into an attack, and the silence read as punishment to Dana. Each has a point and each has a part to repair. The advice changes once the second version is in the room.
Where Mendus fits
Mendus: Who Is Right? AI Judge is an iPhone app (also in the browser at mendus.app) built around the problem above.
- Mindful Mirror works from your side of the story only. You get an analysis of the situation, your blind spots and advice, which is the honest read of your own position.
- Evidence Vault takes up to 6 screenshots of the chat, so the AI works from the messages instead of your summary.
- Two-Way Bridge collects both versions. You can do it together on one device, or send a link so the other person writes theirs separately. Mendus weighs both. Hearing both sides needs Mendus Pro, and the person you invite answers for free.
Every mode gives a verdict as a split in percent. Two design choices matter for fairness. Both people are judged by the same list of behaviors with the same weights, and the app calculates the verdict from that list. The person telling the story can come out more at fault. If a chat turns on one serious accusation that the other side denies and the messages do not prove, it says the main question cannot be settled from that chat and suggests hearing the other side.
Every day you get two free analyses: one of your side and one of screenshots. To make an analysis, your text and screenshots go to an AI provider that does not use them to train models, and screenshots are not kept afterwards. Mendus does not sell your data or use it for advertising.
It is a second opinion for everyday disagreements. A verdict is one outside view and you can disagree with it.
When to talk to a person instead
Skip the app and any chatbot if you feel afraid of your partner, if there is violence, stalking or credible threats, or if you are in crisis. Talk to a local emergency service, a hotline or a therapist. Mendus hides the verdict and shows safety resources when a story describes physical violence or credible threats, but it is not a substitute for help. It is not couples therapy, and it is not medical or legal advice.
Apps that help with this
If you want an outside read on a specific fight, a few apps can help. They do different jobs, so pick by what you need.
- A verdict on one argument: Let's Settle This and Piece (free), Mendus (free every day: one Mirror analysis and one screenshot analysis).
- Coaching over time: CoupleWork and Maia, both paid.
- Daily exercises for both partners: Paired, Gottman Card Decks (free).
Related guides
Is ChatGPT good for relationship advice?
It is good at explaining, reframing and drafting messages. It is weak at fairness when it hears only your version, so give it the raw messages and ask it to argue the other side.
Can AI tell who is right in an argument?
It can estimate who is closer to right in one specific conflict, based on what it is shown. It cannot verify events it was not shown. See how to figure out who is right in an argument and AI apps that settle couple arguments, compared.
Is it safe to share private messages with an AI?
Check what the tool does with them: who processes the text, whether it is used for training, and whether it is stored. For Mendus, see the Privacy Policy.
Should I use AI instead of couples therapy?
No. It helps with one specific disagreement. Patterns, fear and long-running problems are for a therapist.