I used to research like a person being chased. Forty tabs open. Three articles half-read. A notes doc that was really just a graveyard of links I'd never revisit. I'd spend an entire afternoon "researching" and come away knowing almost nothing.
Then I changed one thing. I stopped using AI to get answers and started using it to do the digging — while I kept the thinking for myself. That split fixed everything.
This is the workflow. It's not complicated. But it inverted how much I actually learn per hour, and that's worth more than any single fact.
To turn an AI assistant into a real research assistant: use it to gather, summarize, and organize information, but never to form your conclusions for you. Ask it to map a topic, surface the key arguments, and compress long sources — then you read the summaries, ask the follow-ups, and decide what's true. AI is a tireless intern. The judgment stays with you, or the research is worthless.
For a while I used AI the lazy way. "Tell me about X." It would, fluently, and I'd nod and feel informed and retain nothing. Worse, I had no idea if any of it was true, because I'd skipped the part where you actually engage with sources.
That's not research. That's outsourcing your understanding to something that might be wrong. The understanding has to happen in your head or it didn't happen.
The fix was to demote the AI from "oracle" to "intern." An intern doesn't tell you what to think. An intern goes and finds the stuff, organizes it, and hands it to you so you can think faster. That reframe is the whole trick. It's a specific case of the honest truth about AI productivity tools: the value lands when AI does the legwork and you keep the judgment. And getting useful output from that intern depends almost entirely on how you ask, which is why the prompt pattern that fixed my AI output does so much of the work here.
Photo by Priscilla Du Preez on Unsplash
When I start on a new topic, I don't ask for answers. I ask for a map.
"I'm new to [topic]. Give me the lay of the land: the main schools of thought, the biggest debates, the key terms I'll need, and what a beginner usually gets wrong."
This is gold. In one response I get the shape of the field. Now I know what I don't know, which is the actual starting point of real learning. Without the map, I'd waste hours not even knowing which questions mattered.
The map also gives me search terms I didn't have. Half of bad research is just not knowing the right words to look for.
Here's where AI saves the most time, and where people abuse it most.
I feed it the long source and ask: "Summarize the main argument, the evidence, and any obvious weak points." Then — this is the part people skip — I read the summary as a guide, not a substitute. If the summary flags something important, I go read that part of the original myself.
The summary tells me where to look. My own eyes do the looking on anything that matters. This way I get the speed of AI compression and the reliability of actually reading the source.
AI tells you where the gold might be. You still have to dig the gold out yourself.
For sources where being wrong costs nothing, I trust the summary. For anything I'll repeat, decide on, or publish, I verify in the original. That line is the difference between research and rumor.
The best thing an AI research assistant does is take follow-up questions forever without getting tired or annoyed.
So I push. "What's the strongest argument against this?" "Who disagrees and why?" "What would change your conclusion?" "Explain the part I just didn't understand, simpler."
This back-and-forth is where understanding actually forms. A static article can't argue with you. An AI assistant can hold both sides of a debate and let you poke at each, which is a genuinely new way to learn. Work out of MIT Sloan Management Review makes a related point about AI as a thinking partner rather than an answer machine — the gains come from the dialogue, not the lookup.
I treat it adversarially on purpose. If I only ask questions that confirm what I want, I'll get confirmation and learn nothing. Asking it to attack my view is how I find out if my view survives.
Photo by John Schnobrich on Unsplash
Here's the whole thing in order, the way I run it on any new topic.
| Step | What I ask AI to do | What I do |
|---|---|---|
| Map | Outline the field, terms, debates | Decide what to focus on |
| Gather | Summarize sources, flag weak points | Read the parts that matter |
| Interrogate | Argue both sides, answer follow-ups | Form my actual view |
| Verify | Point me to the original passages | Check anything with a cost |
| Synthesize | Draft a summary of my conclusion | Edit until it's truly mine |
The pattern is constant: AI does the legwork, I do the judgment. Swap that order and you've just replaced learning with copying.
The unexpected benefit wasn't speed. It was that I got better at asking questions.
When your research assistant answers anything instantly, the bottleneck becomes the quality of your questions. So you learn to ask sharper ones. "What am I missing." "What's the counter-argument." "Why does this matter." Those are the questions of someone who actually thinks, and the workflow trains them by default.
I now research in an hour what used to take an afternoon, and I retain more of it, because I spent the hour thinking instead of tab-hoarding. The AI assistant didn't make me lazier. Used this way, it made me sharper.
Here's a distinction that took me embarrassingly long to learn: collecting information is not researching. For years I confused the two, and the AI nearly let me confuse them worse.
Collecting is the dopamine part. Open tabs, save articles, highlight passages, feel productive. At the end you have a pile of material and the warm illusion of understanding. But a pile isn't knowledge. Knowledge is what survives in your head after the tabs are closed, and collecting builds almost none of it. You can spend a whole day collecting and wake up tomorrow knowing nothing you can use.
The lazy way of using AI is just turbo-charged collecting. "Tell me about X" gives you a tidy pile, instantly, and the illusion of understanding is even stronger because the pile is so neat. You feel informed. You're not. You've just collected faster.
Real research is the opposite of collecting. It's the active work of building a model in your own head — testing it, arguing with it, finding where it breaks. That work is uncomfortable, which is exactly why collecting feels so much better. The whole point of the intern reframe is to make the AI do the collecting so that all of your energy goes to the part that's actually research: the thinking.
Once I saw this clearly, I judged my research sessions by a single test. Not "how much did I gather," but "what can I now explain to someone else without looking?" If the answer is "not much," I collected. If I can teach it, I researched. The AI is fantastic at making the first one fast, and the trap is mistaking that speed for the second. It only becomes research when you do the part the AI can't.
Next time you start on a new topic, try asking for the map before anything else, and let the assistant collect while you do the thinking — it's a small change that pays back every hour.
Q: Doesn't relying on AI summaries make you dumber? Only if you stop there. The summary is a map, not the territory. If you read the summary and never engage, yes, you'll learn nothing. If you use it to dig faster into what matters, you'll learn more, not less.
Q: How do you handle AI getting facts wrong? I verify anything with a cost. The workflow assumes the AI can be wrong and builds checking into the steps that matter. Trust the summary for low stakes, verify the original for high stakes.
Q: Isn't this just searching with extra steps? Search gives you links. This gives you a mapped, summarized, interrogated understanding. The difference is the back-and-forth — you can argue with an AI assistant, and arguing is where learning happens.
Q: What if I'm researching something the AI doesn't know well? Then it'll guess, and guessing shows. For obscure or very recent topics, lean harder on real sources and use AI mainly to organize what you find, not to supply the facts.
Most people use AI to skip the thinking. That's why they learn nothing from it. Use it the other way — to skip the digging and keep the thinking — and an AI assistant becomes the research intern you always wanted: tireless, fast, and entirely under your judgment.
Let AI find the gold. Never let it tell you what's gold.
Next time you open thirty tabs, stop. Ask for the map first. You'll be amazed how much of that afternoon you get back.
I spent years saving the hardest task for when I 'felt ready.' Doing it first instead quietly fixed my focus, my dread, and my output.

I tracked every distraction for a week and was horrified by what I found. Then I fixed the three that mattered most.

I went from 200 to 11,000 subscribers without hiring anyone. AI didn't write my newsletter — it did everything around it.

Comments
Sign in to join the conversation
No comments yet. Be the first to share your thoughts!