AI can make decisions faster and better informed. It can also make bad decisions look well reasoned. The difference is almost never the model. It’s how the person uses it.
This is the third post in a short series. The first covered a practical method for everyday decisions, and the second showed how to build a weighted decision matrix. This one covers the six mistakes I see most often, in my own work and in teams adopting AI.
1. Asking for the answer instead of the analysis
“Which CRM should we buy?” invites a confident recommendation with nothing behind it. The model will give you one. It may even be a reasonable one. But you can’t tell, because you haven’t seen the options, the criteria or the trade-offs.
Instead: Ask for the framing, the alternatives and the comparison first. Ask for a recommendation last, if at all, and only after you’ve set the criteria.
2. Accepting unsupported claims
AI assistants produce fluent text whether or not the facts behind it are right. Prices, product features, policies, legal rules and dates are exactly the details most likely to be stale or simply wrong.
Instead: Ask the model to flag which claims it’s least sure about and how to check them. Then check the ones that matter against primary sources: the vendor’s site, the contract, the official documentation. If a claim can’t be verified, treat it as a guess.
3. Leaving your assumptions unexamined
Every decision prompt carries assumptions. “Which of these three agencies should we hire?” assumes you should hire an agency at all. The model will usually work inside your framing rather than question it.
Instead: Ask directly: “What assumptions am I making here? Which one, if wrong, would most change the decision?” Make “do nothing,” “wait” and “do it ourselves” explicit options.
4. Giving it too little context
The model only knows what you tell it. Leave out your budget, your team’s skills, your timeline or your real constraint, and you’ll get advice for someone else’s situation. Generic input produces generic output.
Instead: Write the context a smart new colleague would need: who you are, what you’re trying to achieve, what you’ve already ruled out and why, and what “good” looks like. Two extra paragraphs of context usually do more than any prompt technique.
5. Mistaking confidence for correctness
AI writes in the same assured tone whether it’s right or wrong. People are wired to trust confident, well-structured answers. Put those together and it’s easy to stop thinking.
Instead: Make disagreement part of the process. Ask the model to argue for the option it ranked second, to list what would make its top choice wrong, or to answer as a skeptical CFO. If you’re using it for something important, run the same question in a second assistant and compare where they differ.
6. Handing over the decision
The last mistake is the quietest. After a good analysis, it’s tempting to let the AI decide. But the model doesn’t carry the consequences, doesn’t know what you’ll regret and can’t be held accountable to your team or your customers.
Instead: Decide yourself, and write down why in two or three sentences. That record is what makes the decision yours, and it’s what lets you learn from it later.
The pattern behind all six
Each mistake is the same mistake in a different form: treating AI as an authority instead of an assistant. The fix is the same too. Keep the AI on the work it does well (breadth, speed, structure and tireless drafting) and keep people on the work that needs judgment (framing, verification, weighting and the final call).
That’s what we mean by Human + AI. It’s the difference between AI that makes your decisions faster and AI that makes them better.
For a worked example of these ideas applied to a real, personal choice, see How I Used ChatGPT to Compare Eight LA Phil Holiday Concerts.