How I get expert-level output and learn from it at the same time

On a call last month, a client mentioned wanting to do a member survey. The idea came up in conversation, and like a lot of good ideas on calls, it didn't go anywhere that day.

This week, they emailed asking for the survey questions. I remembered the discussion but couldn't bring the specifics to mind.

Luckily, I didn't have to rack my brain or invent something.

We record every call, and I have an automation set up to download every transcript onto my computer, where Claude Cowork sorts them by client.

All I had to do was ask Claude to search the most recent calls, find where the survey discussion happened, and surface exactly what had been said: what they were trying to figure out, what kind of clarity they were hoping to get.

Claude found the call, summarized the context, and drafted 10 survey questions.

They were okay. But this survey might go out to several thousand members, and 'okay' wasn't good enough, especially since we wanted to ask about how they paid for the memberships and that could be a bit sensitive.

So I asked Claude:

It explained:

"Mostly just the transcript. I reverse-engineered the questions from what you said you wanted to know. I layered in basic survey hygiene (short, clear options, no double-barreled questions) and some retention logic (frequency of use, cancellation risk, referral behavior). That's it. It was serviceable but not rigorous."

Then it gave me three lenses it could use to improve: jobs to be done, our internal community framework, and revealed preferences vs self-report.

Except that wasn't what I wanted. I wanted to know what professional and proven frameworks existed for creating excellent surveys.

So I added a specific direction:

I'm not a professional survey designer. I've created plenty of surveys over the years for different projects and clients. But I've never been trained in it, and I didn't know what I didn't know.

Claude came back with six angles:

I told it that skip logic isn't something I want in this one, ignore that. Take everything else and rebuild the survey from scratch.

What it came back with was a far more client-ready output.

Question ordering by sensitivity. Answer options that were mutually exclusive with no overlap. A framing on the payment question that didn't feel like screening. An anchored open-text question at the end that would pull useful positioning language instead of a wish list. Seven questions instead of ten.

Ready to send to the client.

Sure, you could give it a bunch of feedback. That's better than just telling it to make something better or more concise or more professional or whatever other vague words you might typically try.

But I've found that in addition to giving your own feedback, it is useful (and eye-opening) to ask AI to name the lens it used, and then ask what other lenses it could use to create the output.

When AI gives you a draft, it's working from what it thinks you asked for and drawing on something it was trained on.

The problem is that you don't know which training data it's pulling from for any given task.

Unless you tell it to make the frameworks and inputs explicit.

Asking "what other lenses could you draw on?" surfaces the expertise it may not have considered...and that you may not be familiar with.

In this case, I didn't know the Dillman questionnaire ordering principles existed. I didn't know "MECE answer options" was a thing with a name. I didn't know that where you place a sensitive question in the sequence affects how people answer it.

I know those things now.

That's the other cool thing about this technique: I got a better output, and I learned new stuff.

Both things happened from the right follow-up questions.

Anywhere you're asking AI to build something, analyze something, or write something, ask:

Then pick what applies, tell it what to skip, and have it build (or rebuild).

Hit reply and tell me one thing you've built with AI that felt fine but you weren't sure why it didn't feel great.

And if this was useful, forward it to someone who's been settling for good enough.

Q: Do I need to know the frameworks going in? No. That's the whole point. I'm asking AI to surface them. Bring the question, not the expertise.

Q: What's the exact prompt? The general version for any task: "What lens, framework, or expertise did you use to create this? What other professional frameworks could you draw on to make it stronger?" Then pick what applies, skip what doesn't, and have it rebuild.

Q: What if AI gives me frameworks I've never heard of? Good problem. I ask it to explain each one in a sentence before I decide what to keep. That's part of where the learning happens.

Q: What else does this work for besides surveys? Anything with a professional discipline behind it: reports, positioning documents, research frameworks, scoring rubrics, copy audits. If there's a field of practice that exists, AI has been trained on it.

Thanks for reading!

Nathan Rodgers

👋 Say hello on Substack and LinkedIn

Were you forwarded this email? Click here to subscribe.