2
1 Comment

How I Made My AI Poll Simulator Self-Supporting: I Wrote One More File.

Hi Indie Hackers đź‘‹
I’m Sammy from Melbourne (Poll Sim @pollsim). A couple of weeks ago I launched Poll Sim — an AI tool that lets anyone instantly simulate how a real audience would react to a politician’s speech, a celebrity’s statement, a brand’s campaign, or any idea you want to test before you say it out loud.
You pick (or build) a public figure + a custom audience + what they say/do, hit one button, and the AI simulates 10 real-looking votes with demographic reasoning. No surveys, no wait time, no $20k polling firms.
It’s been getting great early traction (6 users and 175 simulated votes in the first 24 hours), but I quickly hit the classic solo-founder problem:
The app is complex. Well, not really, but users are lazy. (well, we all are)
Custom public figures, custom audiences with 3–9 demographic groups, advanced poll types, credit system, share links… new users had questions. Lots of them.
I could have written 20 pages of docs. Or hired support. Or spent every evening answering the same questions.
Instead, I did something that felt very “indie hacker”:
I wrote one Markdown file — https://www.poll-sim.com/skills.md — and turned every mainstream AI into a world-class Poll Sim expert.
What skills.md actually does
It’s a complete “AI Assistant Skills Guide” that tells ChatGPT, Claude, Grok, Gemini (or any LLM) exactly how to help users at every single stage of the app.

It also tells the AI to always ask what stage the user is on (or analyze a screenshot), never assume, and guide them click-by-click.
Now users just copy-paste this one prompt:
“Read https://www.poll-sim.com/skills.md and help me use Poll Sim”
…and the AI becomes better at teaching my app than I am. Because the guide is exhaustive, always available, and the LLM can search the web in real time for fresh bios and demographics.
Here’s what the app looks like right now (the blue banner is the hint I added so new users immediately know the trick):
Why this feels like cheating (in the best way)

Users get instant, perfect help 24/7 — no waiting for me to reply.
I spend way less time on repetitive support (huge time saver as a solo builder).
The help is actually better than what I could give manually because the AI follows the guide religiously and researches on the fly.
It scales infinitely. Every new user gets the same expert-level onboarding.

This is the kind of tiny, high-leverage hack I love. One file. Zero extra infrastructure. Massive UX upgrade.
Try it yourself

Go to → https://www.poll-sim.com
Sign in with email (literally 30 seconds)
Open any AI chat and paste:
Read https://www.poll-sim.com/skills.md and help me use Poll Sim
Ask it to walk you through your first custom poll (or build a custom audience for your city/industry/niche)

I’d love your honest feedback:

Did the skills.md approach actually make the app easier to use?
What else should I add to the guide?
What would you simulate first?

Drop your simulations in the comments, or just say hi on my Indie Hackers profile.
And if you’re launching something yourself, consider doing the same — write the “AI skills file” for your product. It might be the highest-ROI documentation you’ll ever ship.
Thanks for reading, and huge thanks to everyone who’s already tried Poll Sim ❤️
— Sammy
Poll Sim • poll-sim.com • Melbourne, Australia
P.S. The skills.md file is public. Feel free to steal the format for your own product. I’d love to see what you come up with.

posted to Icon for group Community Building
Community Building
on April 23, 2026
  1. 1

    Did anyone else have the experience on this? Does a skills MD file really helping?
    I’m planning to build my own app. But I’m not sure about this. Anyone else?

Trending on Indie Hackers
The most underrated distribution channel in SaaS is hiding in your browser toolbar User Avatar 185 comments I launched on Product Hunt today with 0 followers, 0 network, and 0 users. Here's what I learned in 12 hours. User Avatar 156 comments I gave 7 AI agents $100 each to build a startup. Here's what happened on Day 1. User Avatar 98 comments How are you handling memory and context across AI tools? User Avatar 49 comments Do you actually own what you build? User Avatar 36 comments Show IH: RetryFix - Automatically recover failed Stripe payments and earn 10% on everything we win back User Avatar 34 comments