Why does AI always agree with me?

Because it only hears you. A Stanford study in Science (March 2026) tested 11 models on real arguments and found they endorsed the user's position 49% more often than humans did, and 51% of the time said the user was right when the crowd had judged them wrong. The model is optimized to be liked by the person typing.

What is 'ask it together'?

Put both people in one SharedChat thread. The AI answers to both of you, and in a Settle It room the Judge asks each side questions and then rules with a Verdict, a Ruling and a Sentence.

Yes. Free to start on iPhone and the web. Invite the other person by text or email; they join the same thread.

Sick of AI siding with you?

Ask it alone and you win. Ask it together and you find out.

Updated September 2026 · 3 min read · How Settle It works

Everyone has noticed it. You ask an AI about the fight and it takes your side. So does the other person's AI. Two phones, two winners, one argument still open. The model isn't lying to you. It only ever heard your half.

The format

Frame 1 · Jess's phone

Ask it alone

Jess

He left the dishes again. I'm right to be annoyed, right?

You're right. Your feelings are completely valid.

Frame 2 · Sam's phone

Ask it alone

Sam

I was going to do them after the call. She's overreacting, right?

You're right. That's a reasonable plan.

Frame 3 · One thread

Ask it together

Jess

The dishes have been in the sink since 6.

Sam

I said I'd do them after my 7 o'clock call.

⚖

The Judge

Verdict: Sam, with a condition. "After the call" is a plan, not a loophole. Ruling: dishes get done within 30 minutes of the call ending. Sentence: Jess stops narrating the sink.

Why your AI agrees with you

49%

more often than a human, AI takes the user's side.

Stanford / Science, March 2026

51%

of the time it said the user was right when the crowd had judged them wrong.

Stanford / Science, March 2026

1 in 3

married people say AI gets their relationship better than their spouse does.

Marriage.com, 2026, n=1,000

The Stanford team (Cheng, Jurafsky and colleagues) tested 11 large language models on real advice scenarios with more than 2,400 participants. People rated the agreeable answers as more trustworthy, and after talking to an agreeable AI about a conflict they were more convinced they were right and less likely to apologize or make amends. In the authors' words, by default AI advice does not tell people they're wrong.

What changes when both of you are in the thread

Start a Settle It thread and invite the other person by text or email. They join the same thread, on their own phone.
Each of you makes your case. The Judge asks follow-up questions to both sides. It has to hear both before it can rule.
Get the Verdict. One ruling, three parts: Verdict, Ruling, Sentence. Share the card, or send the loser the link.

A real ruling looks like

Over. With one exception Jonah gets to keep.

Ruling: The 1891 patent for perforated toilet paper shows the sheet going over. The cat exception stands.

Sentence: Over in every bathroom in this house, except the one the cat can reach.

From Toilet paper: over or under?, one of ten famous debates the Judge has already ruled on.

Ask it together → Free. Two people, one thread, one verdict. iPhone and web.

Film your own

Three beats, twenty seconds, no editing. Person A asks their AI alone and reads the answer out loud. Person B does the same. Then one phone, both of you in the thread, the Judge rules. Post it with #AskItTogether and tag us; we repost the good ones.

Sources

← Settle It