Short answer
When you are stuck between two options, do not ask the AI to pick one. Ask it to compare them, and keep the choice for yourself. Five small steps get you there: write down what you are actually deciding and what "better" means to you, give the AI both options and those criteria, ask it to argue each side instead of crowning a winner, add the things only you know, then write a short decision note and make the call. The note takes about ten minutes and becomes a record you can reuse the next time a similar choice comes up, instead of starting from a blank chat and a gut feeling.
Key takeaways
- A chatbot's confident pick is not a decision. It is one opinion with no stake in the outcome, and treating it as the answer hides why you chose.
- The useful move is to make the AI compare, not choose. A good comparison lays out trade-offs so you can weigh them; a verdict just hands you a conclusion you cannot check.
- Decide what "better" means before you ask. The same two options flip depending on whether you care most about speed, cost, or trust.
- The thing only you know, your budget, your customers, your appetite for risk, is usually what settles a close call. The AI does not have it unless you tell it.
- A short decision note turns a one-off choice into something reusable. Next time you start from your own reasoning, not a fresh blank prompt.
The trap of "just tell me which one"
A solo founder has two messages for the top of her landing page. Option A is "Bookkeeping software that saves you ten hours a month." Option B is "Bookkeeping that runs itself, so you never think about it." She cannot decide, so she opens a chatbot, pastes both, and types "which one is better?" Back comes a clean answer: "Option B is stronger because it speaks to an emotional outcome rather than a feature." Done. She pastes B onto the page and moves on.
The problem is not that the AI is wrong. The problem is that she has no idea whether it is right, and now she cannot tell. A landing-page message lives or dies on whether her specific visitors click and sign up. The model has never met her visitors. It picked B because, across the huge pile of text it learned from, emotional phrasing tends to get praised. That is a pattern, not a fact about her business. She has swapped her own judgment for a confident-sounding average, and she has nothing written down to learn from when she checks her signup numbers next month.
There is a quieter cost too. When you let the AI choose, you stop thinking. The whole value of a hard choice is that wrestling with it teaches you what you actually care about. Outsource the verdict and you skip the part that makes you better at the next decision.
The fix: make it compare, then you decide
The fix is small. Instead of asking for a winner, ask for a comparison, and keep the verdict for yourself. An AI chatbot is genuinely good at the first half of a decision: laying out options, naming trade-offs, surfacing angles you had not considered. It is bad at the second half, the part where someone with skin in the game weighs those trade-offs against what they know and what they can live with. Split the job along that line and you get the best of the tool without handing it the wheel.
The five steps below are how you run that split. The output is a short decision note, a few lines you write and keep, so the choice leaves a trail.
Step 1: Name the decision and what "better" means
Before you open the chat, finish two sentences. First: "I am choosing between option A and option B." Second, the one people skip: "For me, the better option is the one that reaches my real goal." For the founder, that second sentence might be "the one that gets more of the right visitors to start a free trial." Not the cleverest line, not the one she likes reading. The one that drives trials.
This matters because "better" is empty until you fill it. Option A and Option B can each win depending on the test. If better means "clear to someone in a hurry," the plain feature line probably wins. If better means "memorable," the runs-itself line might. By naming your real measure first, you stop the AI from quietly choosing the measure for you, which is what happened when it decided, on its own, that emotional resonance was the thing that mattered.
Step 2: Give it both options and your criteria
Now go to the chat and set the job up clearly. A prompt that works: "I am deciding between these two landing-page headlines. My goal is more of the right visitors starting a free trial. Compare them on clarity, believability, and how well each fits a busy small-business owner. For each headline, give me the strongest case for it and the strongest case against it. Do not tell me which to pick." You are handing it your options and your definition of better, then pointing it at comparison rather than judgment. That is the difference between a vague request and a useful prompt: you specified the criteria and the shape of the answer.
What comes back is now something you can use. Instead of "B is stronger," you get that A is unmistakable in two seconds but sounds like every other tool, while B is distinctive and promises a feeling but might read as too good to be true to a skeptical buyer. Those are trade-offs you can actually weigh, because each one connects to your goal of starting trials.
Step 3: Make it argue both sides
Even a balanced comparison can lean. Models tend to agree with the framing they are handed, so if your prompt hinted you liked B, the comparison may quietly tilt toward B. Counter it by forcing both cases hard: "Now make the most convincing argument that A is the right choice. Then make the most convincing argument that B is. Assume each one is correct and sell it."
This costs you one more message and it is worth it. Reading a genuine defense of the option you were about to reject is the fastest way to find the hole in your thinking. Maybe the case for A reminds you that most of your traffic comes from a comparison site where buyers are scanning fast and skeptical, which makes A's plainness an advantage you had undervalued. The AI did not know that. The exercise of arguing both sides is what surfaced it.
Step 4: Add what only you know
Here is the step that actually decides most close calls, and the AI cannot do it for you. Write down the things about your situation the model does not have. The founder might note: her current visitors mostly arrive from a price-comparison directory, she has been burned before by a clever headline that got clicks but no signups, and she can only afford to test one message this month. None of that was in the chat. All of it bears on the choice.
This is also the honest answer to why you should not just trust the verdict. A chatbot writes the most likely words given what you typed, not the true answer for your business, and it will state a guess about your customers with the same confidence as a fact. Your private knowledge is the correction. When the model's tidy comparison meets the messy specifics only you hold, the right option usually becomes obvious, and it is sometimes the one the AI would never have crowned.
Step 5: Write the decision note and call it
Now make the choice and write it down in a few lines. A decision note does not need a template app, just these fields in any notes file:
- Date and the decision: "2026-06-20, landing-page headline."
- The options: A and B, in full.
- What "better" meant here: "more of the right visitors starting a trial."
- What each option won on: "A: instantly clear, fits skeptical comparison-site traffic. B: distinctive, but risks sounding too good."
- What I knew that the AI did not: "Most traffic is skeptical and fast. Burned before by clever-but-empty. One test only."
- The call and why: "Going with A. My traffic rewards clarity over cleverness, and I cannot afford a swing."
- When to revisit: "Check trial signups in four weeks."
That last field is what makes the note pay off. In four weeks she reopens it, sees what she predicted, and compares it to what actually happened. If A underperformed, she does not start from zero, she starts from a written record of her own reasoning and adjusts one piece. The ten minutes she spent is the difference between learning from the choice and just having made it.
A second example, so you can see the pattern
The method is not about headlines. Say a two-person team is deciding which repetitive task to automate first: sending invoice reminders, or sorting incoming support email. They are tempted to ask an AI assistant "which should we automate first?" Instead they define better as "saves the most hours with the least chance of an embarrassing mistake." They ask the AI to compare both on time saved, setup effort, and blast radius if it goes wrong. They make it argue each side. Then they add what only they know: invoice reminders touch money and a wrong one would rattle a client, while a misfiled support email is low-stakes. The note writes itself, support email wins, and the reasoning is on record for the next automation argument. Same five steps, completely different decision.
That is the point. Once you have run it twice, comparing two options with the AI as a sharpener rather than a judge becomes a habit you reach for whenever you are stuck between two roads.
When the two options look equally good
Sometimes you run the whole thing and the options come out genuinely tied. That is useful information, not a failure. A real tie usually means the choice matters less than the time you are spending on it, so pick the one that is cheaper to reverse and move. Write that in the note too: "Tied on the criteria, chose A because it is easier to undo." A reversible choice made quickly beats a perfect choice made slowly, and your note will remind you that the stakes were low if you are ever tempted to relitigate it.
Common mistakes to avoid
- Asking "which is better?" and taking the answer. That hands the AI a job it cannot do well and leaves you nothing to learn from.
- Skipping the definition of better. Without it, the model picks the measure for you, usually whichever one its training data tends to praise.
- Letting the comparison lean. If you signaled a favorite, force the AI to argue the other side as hard as it can.
- Leaving out what only you know. Your budget, your customers, your risk tolerance are usually the deciding facts, and the AI does not have them.
- Not writing the note. A choice with no record is a choice you cannot learn from, so the next similar decision starts from scratch.
FAQ
Isn't asking the AI to decide just faster?
It feels faster and is usually slower in the end. A pick you cannot explain is a pick you cannot defend, fix, or learn from, so you end up re-deciding the same kind of thing again and again. Ten minutes spent comparing and noting saves the repeat.
What if I genuinely have no opinion and want the AI's pick?
Then ask for its pick, but ask for the reasoning and the strongest case against it in the same breath. Read the case against before you accept the case for. The goal is never to ignore the AI, it is to stay the person who weighs it.
Does this work for choices with more than two options?
Yes, but it is sharpest with two. With three or more, first use the AI to cut the list to the two real contenders, then run these steps on those two. Comparing everything at once tends to produce mush.
Can I just keep all this in the chat instead of a separate note?
You can, but you probably will not find it again. A decision note lives where you already look and carries a date and a revisit reminder, so it actually comes back to you when the next similar choice lands.
Sources
- https://www.nist.gov/itl/ai-risk-management-framework, NIST, AI Risk Management Framework: a public standard built on keeping a human accountable for decisions that AI informs, rather than letting the system decide.
- https://platform.openai.com/docs/guides/safety-best-practices, OpenAI, Safety best practices: official guidance recommending human review and human judgment over AI output in consequential decisions.
- https://docs.anthropic.com/en/docs/test-and-evaluate/strengthen-guardrails/reduce-hallucinations, Anthropic, Reduce hallucinations: why a model can state a confident pick or claim that is not true for your situation, and why outside judgment is the check.