Why AI is Both Great and Dangerous for Research
AI will answer any question you ask with the same confident tone, whether it’s completely accurate or completely wrong. It doesn’t pause to say “I’m not sure about this one.” It doesn’t hedge the way a careful person would. It just answers.
This is actually a well-documented limitation — AI systems are optimized to produce fluent, helpful-sounding responses, not to accurately flag their own uncertainty. The technical term is “hallucination.” The practical meaning is: you’ll get confidently stated wrong information sometimes, and it won’t look any different from the right information.
Here’s the thing though: knowing this doesn’t mean AI is useless for research. It means you use it for the right parts of research and handle the rest yourself. Used correctly, AI can compress hours of orientation into minutes and help you ask better questions than you would have walked in with. That’s genuinely valuable — if you know the rules.
You’re often researching things you don’t have a specialist for — tax questions, contract language, industry regulations, supplier decisions. AI is useful for getting oriented before you talk to someone who actually knows. It’s not a replacement for that conversation. It’s preparation for it.
What AI Gets Right and Where It Falls Apart
Not all topics are equal. AI is significantly more reliable in some areas than others. Here’s a rough guide — not a guarantee, but a useful calibration:
The pattern: AI is reliable when it’s working from broad, well-established knowledge. It’s unreliable when it needs specific, recent, or highly specialized facts. The good news is you can usually tell which category you’re in — and you can ask.
Three Activities. Real Research Situations.
This is the single most useful thing in this volume. Pick a topic you actually need to understand right now — something with real stakes. A medical question, a financial decision, a legal situation, a business move, a major purchase. Something where being wrong matters.
“Give me a solid overview of [topic]. I’m trying to [what decision or action you’re working toward]. Tell me: what I need to understand, what’s contested or unclear, what the common mistakes are, and what questions I should be asking before I act on any of this.”
“Now tell me: what parts of what you just said are most likely to be wrong, outdated, or vary significantly by my specific situation? What should I verify before acting on any of this — and who would be the right person to ask?”
Read both responses together. The first gets you oriented. The second tells you where to focus your verification. Together they put you further ahead than most people going into that doctor’s office, that financial advisor meeting, or that contract negotiation.
Think of a real decision you’re currently sitting on. Something you’ve been turning over without a clear path. Could be a career move, a business decision, a financial choice, a relationship situation, anything where you’re weighing options and not sure what to do.
After you get the response, follow up with one of these depending on what came back:
Career decisions — which job to take, whether to ask for a promotion, whether to leave — are worth running through this framework. AI won’t make the decision for you, but it will surface the questions you should be asking and the things you’re probably not weighting correctly. Use it as a sounding board before talking to a mentor or manager.
This one is different from the other two. Instead of asking AI to help you figure something out, you’re going to ask it to push back on something you’ve already decided or believe. This is where it’s genuinely useful in a way that’s hard to replicate.
Pick something you’re fairly confident about — a plan you’ve made, a conclusion you’ve reached, an approach you’re taking. Then:
Read it without being defensive. You’re not trying to be talked out of your position. You’re trying to find the holes before someone else does.
A Four-Step Research Framework That Works Every Time
Once the habit is formed, this takes five minutes and consistently produces better outcomes than either pure AI reliance or skipping AI entirely.
Give AI the full context and ask for the overview. What do you need to understand, what’s contested, and what questions should you be asking? This replaces two hours of reading with ten minutes of conversation.
Ask it directly: what parts of your answer are most likely to be wrong, outdated, or vary by situation? Then treat those parts as hypotheses, not facts. Don’t act on them without checking.
For decisions with real consequences — legal, medical, financial, contractual — verify the specific claims that will actually affect your outcome. You now know exactly which ones those are from step two.
You’re not replacing the doctor, lawyer, or financial advisor. You’re going into that conversation knowing the right questions to ask and what to listen for. That combination — AI for orientation plus professional judgment for the important parts — is genuinely better than either alone.
Before any significant vendor decision, contract signing, or regulatory question — run this framework. Five minutes of AI orientation before a legal call means you understand what you’re being told and can ask follow-up questions. That difference alone can be worth hundreds of dollars in billable time.
The Research Prompt Cheat Sheet
One Thing to Do Today
Pick one real decision or question you’re sitting on right now. Run the two-step from Activity 1. Don’t save it for a bigger moment — use the actual thing in front of you. The framework works for small decisions too, and small decisions are where the habit forms.
By this point in the series you have the tools, the prompting skill, the ability to use AI at work, and a framework for research and decisions. Vol. 5 is about putting all of that into a routine that sticks — building the habit so it stops requiring conscious effort.
AI Newsletter
The AI Digest
What is actually happening in AI and what it means for your job, your money, and your life. Plain English. No hype.
Leave a Reply