I run a handful of products out of my bucket.
a couple of workflow and others.
Claude touches almost every part of that workflow, from writing fight-prediction emails to reviewing pricing pages.
For most of that time I’ve prompted the way everyone does:
describe the task, get an answer, fix what’s wrong, repeat.
Last week, on my friend Piyush’s suggestion, I changed one thing. Before any task that wasn’t trivial, I added this line to the end of my prompt:
Before you start, ask me the 5 most important questions that would help you do this well. After I answer, then begin.
That’s it.
No other change to how I work.
Same tasks, same week-to-week workload, just this one instruction bolted onto the front of everything.
I wanted to see if it was actually worth the extra back-and-forth or just a nice-sounding trick that looks good in a thread or tweet.
Okay, I’ll keep it short.
What I expected versus what happened
I expected the questions to be filler like the kind of generic clarifying questions a junior freelancer asks because they’re supposed to, not because they need the answer xD
What actually happened was closer to a junior PM doing their homework before a kickoff call :)
More importantly, when I asked Claude to help me restructure the GritGlean Sellers page, instead of starting with a layout, it asked me things like who the page is actually competing against for attention, whether someone landing there already knows what TrustMRR or Flippa listings are, and what action I actually wanted them to take versus just browse.
None of those were questions I’d written down anywhere.
They were questions I hadn’t consciously answered for myself yet, and writing the answers out forced me to commit to a decision I’d been quietly avoiding.
That pattern repeated all week.
The questions weren’t filling gaps in Claude’s knowledge.
They were filling gaps in mine.
More importantly,
The rework difference was the real number
I didn’t track this scientifically, so take it as a gut read rather than a spreadsheet, but the difference that mattered to me was rework.
In a normal week, a decent chunk of my Claude output needs a second pass, not because it’s wrong, but because it solved a slightly different problem than the one I actually had.
With the step added, that second pass mostly disappeared.
Not because the writing got better.
Because the first draft was finally aimed at the right target.
The cost was time upfront.
But Shashwat, answering five questions before getting an output takes longer than firing off a request and reading whatever comes back.
Yeah.
And for genuinely small tasks like fixing a typo, rewording a sentence: this was an overkill and I dropped it by day three myself.
But, for anything with more than one reasonable interpretation, it was worth every extra minute.
Now, I’ll be a little more honest with you.
Where it didn’t help
Two places this fell flat.
First, tasks where I already knew exactly what I wanted and had already done the thinking since I was answering questions I’d already silently answered in my own head.
Second, on tasks I was using to think out loud and explore an idea I hadn’t formed yet, the questions sometimes forced me to fake certainty I didn’t have, which produced answers that felt more confident than the situation actually warranted.
So the rule I landed on by the end of the week:
use it for anything with real ambiguity or stakes, maybe like content strategy, pricing decisions, anything customer-facing.
Skip it for anything like for genuinely open-ended brainstorming where premature specificity is the enemy.
So, I’m keeping the habit, but not as a blanket rule.
It’s like it’ll become a judgment call I make in about a second before I hit send.
This also is making me realize how often I send off a task without having actually decided what I want and how much faster everything moves once something, even an AI asking dumb-sounding questions, forces that decision out of me first.
Liked the above article? You might like the one below too: