Ask ChatGPT the same 'should I do X or Y' question twice, in two fresh chats, with the exact same facts. The only thing you change is the order you list the two options. The verdict flips too, and it flips more often than anyone would guess.
The original poster ran this exact test inside a popular prompting community and watched the recommendation swing on nothing but sequence. Nothing about the situation moved, no new detail, no shift in the stakes. The model never weighed both choices to begin with, so whatever loaded first became the frame for everything after.
This isn't a stray glitch: AI evaluation research documents the same lean, where models judging two answers side by side favor whatever sits in a certain slot. The pattern bleeds straight into everyday calls too, from which job to take to which apartment to sign to which laptop lands in your cart. Here's what you get below: four short prompts that turn a manufactured verdict back into the honest close call it always was.
Unlock Your Next Income Stream: 100 Side Hustles for Ambitious Founders
In a world where financial freedom feels like a distant dream, smart women are building wealth on their own terms.
• Finally, a curated database of 100 proven side hustles (that actually work)
• Each idea comes with required startup costs, time investment, and potential earnings
• Exclusive insights from founders who've turned side gigs into 6-figure empires
• Detailed skill requirements so you can match your talents to the right opportunity
• Bonus: Priority scoring system to identify which hustles align with your lifestyle
Don't let another month slip by watching others build their empire. Your next income stream is hiding in our database, waiting to be discovered.
👉 Download the Side Hustle Database Now
*Ad
Why the first option keeps winning
The old way is asking once and taking the recommendation at face value. You type your question, ChatGPT hands you something confident-sounding, and you move on with your day. The trouble: that confidence is manufactured.
The mechanism is simpler than it looks. The model builds its reply one word at a time from whatever already sits in the conversation, so the first option you name becomes an anchor. Every sentence after it leans on that anchor, which means the 'winner' can be locked in before the model reasons about a single tradeoff.
It isn't only order, either. The option you describe with the warmest language gets the same head start, because glowing words prime a glowing conclusion. What struck me here: the model isn't lying to you, it's pattern-completing from a frame you handed it without noticing.
The practical cost is real, and I think it's what people underrate. A manufactured verdict feels like a decision, so you stop looking. You act on a coin flip believing it was a clear call.
The swap test that catches it
Run the swap test on anything that matters. Open two fresh chats, ask the identical question, and reverse the order of the two options in the second one. If the recommendation holds both times, trust it more.
If it flips, you've found a close call the answer was dressing up as obvious. That gap is the real signal. The deciding factor is now yours to name, not the model's to hide.
Picture two apartments at near-identical rent, or two laptops a hair apart on specs. Ask once, and the model sounds certain about whichever you named first. Swap them and watch: a sudden reversal tells you the choice was tied all along.
And the fix costs almost nothing. Two minutes, two chats, one reversed question. That's cheap insurance against a decision you'll live with for years.
*Ad
Stay ahead of inflation and market volatility with a rate that doesn't move once you're in it. Groundfloor Notes now offer 4 fixed-rate terms — 12-month at 8.5%, our new 6-month Preferred Note (accredited only) at 7%, 3-month at 6%, and 1-month at 5% — so there's a fit for almost any timeline. Every Note is backed by a diversified pool of real estate loans, with no fees and no strings attached — what you see is what you get. Groundfloor has never missed a payment since launching Notes in 2018, paying out over $38 million in interest to date. Rates are subject to market conditions and can change, so today's rate is only guaranteed if you lock it in now. Join over 300,000 investors already parking their cash somewhere reliable — and earn your share of the next $38 million.
Three prompts for when you only get one shot
No time for two chats? Use neutral framing instead. Tell the model the options are listed in no particular order, then make it build the strongest honest case for each one alone before it recommends anything.
Here's why I lean on this one: it blocks the model from leaning on whatever loaded first. It has to argue both sides at full strength, then name the single factor that decides. If that factor is weak, you've told it to say so plainly.
Comparing three or more? Ask for a ranked list with a one-line reason each, then ask whether that ranking would survive a reversed order. That second question is the whole trick: it makes the model flag its own instability instead of burying it in a clean, confident-looking list.
The last prompt is about you, not the machine. Strip out 'I'm leaning toward,' 'I really like,' and 'or maybe' before you hit send. The creator found your own lean steers the answer harder than listing order ever does, so a neutral question is the only path to a neutral answer.
I ran the swap test on my own last three either-or questions and two of them flipped, which is how I ended up rehearsing neutral framing instead of hoping I'd remember it under pressure. The chatgpt track in 3 Minute AI is built for that kind of rep: a short lesson, then a lab chat where you run the prompt against a real question before it counts. Two full courses and the daily lessons are free.
The caveat that keeps you honest
Two things worth holding onto. Sometimes the first option is the better one, and the swap test is how you confirm that instead of assuming bias where there's none. A flipped verdict doesn't mean both answers are wrong.
It usually means the call hinges on a factor you haven't shared yet. That's your cue to add it and ask again. The extra minute buys you a real answer instead of a coin flip wearing a suit.
One voice in the thread was blunt: decide the deciding factor yourself, never hand the whole call to an LLM. That's the right instinct. These prompts aren't about letting the model choose for you, they stop it from choosing for you without you noticing.
Product teams aren’t short on ideas. They’re missing a system.
Jira Product Discovery gives teams one place to capture customer feedback, prioritize ideas with consistent frameworks, and build living roadmaps everyone can align on. And when it’s time to build, those decisions connect directly to delivery in Jira, so everyone can see how the roadmap turns into real work.
*Ad
Run it on your last decision
Pick the last decision you handed to ChatGPT, ask it again with the two options flipped, and watch whether the verdict holds.
The swap test and the three neutral-framing prompts walk you through catching a fake-obvious answer before you act on it.
Worth 10 minutes if you lean on ChatGPT for real either-or calls and want to stop the order of your options from quietly deciding for you.
Credits to the original creator.
Some teams never seem to stop moving. They're on Attio, the agentic CRM.
It’s your always-on revenue engine: agents and workflows build pipeline, chase every buying signal, and move deals forward alongside your team.
Teams like Parallel, Turbopuffer, and Wordsmith build on Attio. Are you one of them?
*Ad
Poll: You swap the options on your last big ChatGPT decision and the verdict flips. What do you actually do?
Hit reply and tell us why.
