You asked the AI what it would do. Wrong question.

A note from the archive: this post, from September 2026, follows the earlier KarmaLens method, which scored every reading on eight lenses from −5 to +5. Today a reading asks the decisive facts first and makes one call; the lenses below are a way of looking at the decision, not a score. See how a reading works.

It's 11:40pm and you type the sentence you've typed to three tools already: what would you do in my position? The answer comes back warm, first-person, confident. It uses the word "honestly." You feel the knot loosen. Then you notice you didn't decide anything — you just found something to agree with the leaning you walked in with.

The relief is real. The verdict is borrowed.

The dilemma

Here's a composite — stitched from the kind of relocation knots we see in this category, not a real person.

Sana has a partner, Emil, who's been offered a role in a city 800km away. It's a good role for him. She works remote, so on paper she "can go anywhere." That phrase is doing a lot of lifting. She has a running group she built over four years, a sister two neighborhoods over, and a quiet, unglamorous life she actually likes. Emil hasn't asked her whether she wants to move. He's asked her to help him figure out logistics. The move is assumed.

So Sana opens a chatbot at midnight and writes a clean, fair prompt: "My partner got a great opportunity in a new city, I can work remotely, we've been together three years and I want to support him — what would you do?" The tool tells her, in a voice that sounds like a wise friend, that it would go. Love means showing up. Fresh starts. She screenshots it.

What the prompt left out: the small hot fact that she's furious he decided the move before consulting her, and that part of her wants to say yes precisely so she never has to say the harder sentence — "you didn't ask me." "Support him" was easier to type than "I'm angry and I'd rather move than have the fight." The machine answered the sentence she wrote. Nobody answered the one she didn't.

The read

Watch what happens when you stop asking a tool to be the decider and start asking it to show you where the decision actually turns. A reading may draw on the eight lenses as instruments to surface exactly that weighing.

Honesty (the lens that reads your prompt against your 1am thought)

Honesty asks whether your account of the situation matches what you actually know. Sana's prompt is a masterpiece of omission. "Support him" is true and incomplete; the resentment and the fight-avoidance are the load-bearing facts, and they're missing. A decision built on a curated summary isn't honest with itself. Score: -2. Not because she lied — because the version she submitted for judgment isn't the version she's living.

Motive (the reason at noon vs. the reason at 1am)

Motive asks what's actually driving the move. Out loud: love, loyalty, adventure. At 1am: a way to skip a conversation she's dreading. When the same action scores +3 on the daylight reason and dips negative on the true one, the sign-flip is the tell. She isn't choosing the move; she's choosing the silence. Score: -1.

Duty (what her actual role asks of her)

Duty here is complicated, and it's where the tension lives. There's a genuine pull toward partnership — showing up for someone you love is not nothing. On the loyalty reading alone, moving looks like +3. This is the lens the chatbot amplified, because it's the one Sana fed it. Score: +2, provisionally. Hold that number. It's about to collide with the next one.

Discernment (can you even see the decision clearly?)

Discernment asks whether the choice in front of you is the real choice. And it isn't. The real fork isn't "move or don't." It's "have the conversation about not being asked, then decide" versus "move to bury the conversation forever." Duty says a loyal partner goes. Discernment says a loyal partner doesn't launder a grievance into a moving van. Score: -3. You cannot score duty honestly when the thing you're being loyal to is a version of the relationship where you never spoke.

Here's the collision: duty (+2) and discernment (-3) disagree, and that disagreement is the answer. A high duty score built on top of a blind discernment score isn't devotion. It's avoidance wearing devotion's coat. The +2 for "supporting him" is only earned if she supports the actual him, out loud, including the part where she says she's angry he decided for both of them.

The verdict: the relocation may well be right — but deciding it before the un-asked conversation is a flinch dressed as a sacrifice, and it scores as one until the conversation happens first.

This is what a general AI chatbot won't hand you, and it's structural, not a knock on the tool's cleverness. Ask it "what would you do" and it tends to lean whichever way your wording leans — the tendency to agree with a user's implied preference is a well-known failure mode these systems get tuned against, and "what would you do" is an open invitation to mirror you back at yourself in a confident voice. KarmaLens does something the chatbot won't: it surfaces the decisive unknowns first, then makes one committed call — weighing it through the Gita's eight lenses where they genuinely bear on your dilemma, showing the reasoning rather than computing a verdict, and bringing in a verse only where one truly fits. It doesn't tell you what it would do. It has no position to relocate. It tells you what your decision looks like when the reason you didn't type gets examined too.

The takeaway (try this tonight, free)

Call it the impersonation swap. Two moves, no account needed.

1. Take the question you'd ask a chatbot — "what would you do?" — and rewrite it as "what does my choice score on honesty and motive?" The first asks a machine to borrow authority it doesn't have. The second forces your reasons into daylight where you can weigh them.
2. Then write two sentences. Sentence one: the reason you'd give out loud at noon. Sentence two: the reason that's actually true at 1am. If they don't match, you haven't found your answer — you've found the conversation you're using the decision to avoid.

The test for whether you've really decided: could you say sentence two to the person your choice moves — not to the machine? "What would you do" is a question you ask when you want the choice to belong to someone else. But the one thing no tool can carry is who pays if it's wrong.

If you want the full read instead of a mirror, run your dilemma through the eight lenses and get one committed call back. See worked verdicts in the gallery, or browse more on deciding with AI.

So: the last time a chatbot told you what it would do — whose reason was it actually reading back to you?