Issue# 3: Research, AI Style
AI-generated cartoon of Paula McConnell and Jay McGrane sitting at a desk in front of a mind map board with smartphones and a laptop.
Written by Jay McGrane with an assist from Paula
We tried researching with AI. Here’s what stuck.
Today isn’t about telling you how fast Deep Research works or gushing about NotebookLM, or saying “perplex it” instead of “google it.”
No, it’s a behind-the-scenes look at how to research with AI.
Paula and I both hold graduate degrees so we’ve been around the research block a time or two. We’d like to give you a chance to see how we’re using these tools in real-time. We’re right in the messy middle of experimenting with AI research so we definitely don’t have it all figured out.
We hope it helps you become more effective whenever you toggle research mode
Jay’s Workflow
My attitude towards research remains the same. I follow the research to find my ideas.
Back in the day, this method looked like copious notes either by hand or an endless google doc full of quotes/sources.
Deep Research provided an efficient shortcut. It quickly pulled the research for me and I found interesting data to mine. However, I often had to prompt it to use more academic sources. In my opinion, Deep Research isn’t human-oversight free yet.
The real magic came when I began brainstorming a structure. I used to spend about a week carefully combing through the research to uncover the angle. This time I worked with the AI to quickly surface the contradictions in the data I wanted to explore. In one day, I had a white paper roughly outlined with sources.
Did I mention I did all of that on a mobile phone while I was teaching kids all day?
Paula's Workflow
Sure, now that I’m all about AI, research looks different to me.
But the habits run deep.
I’ve traded my orange vinyl grad school binders for AI apps, but the process is surprisingly familiar.
I use NotebookLM as my main workspace. Everything goes in: PDFs, transcripts, dense podcasts I want to unpack. I tag, summarize, and build connections between ideas using their new mind map tool. It’s part research assistant, part thinking partner. And when I need to talk out loud to sort through my thoughts, I use Notebook LM’s spoken podcast feature.
But AI moves fast. Sometimes too fast. So when I want to process deeply, I print the materials. I take my highlighter and a lawn chair to the dog park.
Yes, really. (See? I'm not all about the AI all of the time.)
That’s how I give my brain space to digest.
At the same time, I’m experimenting with custom GPTs that act like smart filing systems. I’ve built one that’s trained to tag content, surface patterns, and help me find links between ideas when I’m ready to create. What used to take hours or days now takes a few minutes.
I’ve found a rhythm that feels like thinking out loud with help.
Two Humans, One AI, and a Bunch of Guardrails
(aka how we stay intentional in the research rabbit-hole)
So what does staying human-in-the-loop actually look like for research? Like a lot of skepticism. We’re always questioning the AI’s views, ourselves, and the quality of the sources. Otherwise, we find AI does all the thinking and it creates a subpar product. Here’s how we do it:
Jay’s #1 Method — The Direct Quote
I often ask the AI to surface a direct quote for me during any research. This newsletter was built using this exact method from a conversation between me and Paula.
I like direct quotes because they are less likely to include any additional AI “guidance.” However, I get a big efficiency boost from not having to comb through for exact quotes in the research myself. I also ask AI to audit any summaries for logical leaps or bias.
Paula’s #1 Method — The AI “Self” Evaluation
If I suspect a chat is going off track (remember it’s all patterns and your AI can get sidetracked if it has too much to work with) I’ll ask it to stop and evaluate its behavior. Give itself a grade.
If you’ve given it strong context and clear directions, it will usually recognize when it’s drifted from your intent — and realign with the goal.
If it’s not realigning?
That’s your cue to pause and offer more context, or tighten the instructions.
This kind of reflective prompting feedback loop is a real power move when it comes to getting performance from your AI tools. If you like poking around in the process, go one step further:
Take a look under the hood and ask:
Why am I seeing these answers?
Why are the patterns off?
Yes, you may see some tech terms. But you’ll also get a glimpse of how these things really work.
Our Favourite Way to Stay the Human at the Center
Slow your AI down.
AIs are pushy pieces of code. Inveterate people pleasers. And, generally overachieving busy bodies.
As the human at the center, you have the right to:
Refuse the help.
Take your time.
Tell your chatbot to be quiet.