97 days building an automated GitHub-idea newsletter — 72 subscribers, 0 revenue, here's everything I learned
Back in early May I started GitPulse Weekly — a pipeline
that scans 500+ trending GitHub repos weekly and turns
open-source traction into commercialization ideas,
delivered every Saturday.
97 days later: 14 issues sent, zero missed Saturdays,
72 subscribers, 0 paying customers.
Here's the honest breakdown.
What I built
A fully automated pipeline — GitHub API scanning across
5 languages, 3 detection rules (fast-rising repos,
clustered themes, under-productized tools), AI enrichment
with a 3-layer fallback chain so it never silently fails,
and permanent deduplication so no idea repeats.
Started as plain text emails through Gmail. Ended as a
branded HTML email + a 13-page designed PDF report for
Pro subscribers, complete with execution roadmaps,
validation strategies, and a scoring system.
What actually worked
Personal outreach. Every single subscriber who mattered -
replies, feedback, engagement came from a direct message,
not a community post. Cold posting in Discord/Reddit/IH
brought almost nothing. Warm, one-on-one outreach brought
everything.
Showing up every week. Even with zero revenue, sending
something real every Saturday for 14 weeks straight built
more trust than any single feature I shipped.
Growth velocity over raw stars. "Gained 50,000 stars in
11 days" is a completely different signal than "50,000
stars total." This became the product's real differentiator.
What didn't work and what people told me directly
Three separate people, unprompted, told me some version
of the same thing: "I don't believe idea-givers without
proof." Not implied. Said outright, more than once.
That's the real blocker. Not pricing (someone outside my
target audience told me ₹299/month "sounds reasonable as
an outside judge"). Not the format. Trust.
I responded by adding Market Proof (show a real company
that already validated the gap) and Validation Strategy
(how to test demand before building) to every idea. Whether
that's enough, I don't know yet — too early to tell.
The mistakes
- Spent too long polishing the pipeline before doing enough
outreach. The engine was solid by day 30. Distribution
didn't get serious focus until much later. - Sent from a personal Gmail for way too long — landed in
spam for a real chunk of subscribers before I diagnosed it
and moved to a proper email service. - Chased Discord servers and community posts that had almost
zero conversion, when the data was telling me the whole
time that personal outreach was the only thing working.
Where it stands
Real infrastructure. A product I'm not embarrassed by.
Zero revenue. That's the honest scoreboard at day 97.
Not stopping — but the next phase isn't "add another feature."
It's finding the 15-20 people who are actually my target
audience and giving them something worth paying for,
directly, one at a time. Same thing that worked from day one,
just more deliberate about it.
If you're building something similar, or have thoughts on
the trust problem — genuinely want to hear them.
Running something parallel — automated GEO-article pipeline, two weeks, 18 posts, $0. Different output, same lesson: pipeline technically correct, distribution near-zero.
On the trust problem specifically: the fix isn’t cosmetic (removing “AI tells” in the phrasing). It’s structural — every specific claim needs a number, a source, and a year. Not because it sounds more credible, but because it forces you or the AI to either find the actual proof or soften the claim. A generated idea with no verifiable data point underneath it can’t be cross-examined, and that’s what “I don’t believe idea-givers” actually means. They want something they can check.
Your Market Proof section is exactly this constraint applied at the product level. Three people told you outright what most unsubscribed users just feel. The deeper question is whether the proof needs to show the gap is real or that you’ve seen enough of the space to recognize a real gap — those require different evidence.
The 1-on-1 approach is the right move. 72 subscribers from direct conversations over 14 weeks is a stronger signal than the zero-revenue number. Building trust with 15 people who’d actually pay teaches you more than adding another pipeline feature.
"A generated idea with no verifiable data point underneath
it can't be cross-examined" — that's a sharper way of
saying what I was circling around.
Your point about the two kinds of proof is the one I
haven't fully resolved yet: showing the gap is real vs.
showing I've seen enough of the space to recognize a real
gap. Right now Market Proof does the first (a company that
already validated it). I hadn't separated out the second —
worth thinking about whether subscribers need both.
Appreciate the GEO-pipeline parallel too. Different domain,
same wall. Good to know it's not just a GitHub-ideas problem.
The trust problem you're describing might not be about proof of demand, it might be that a generated idea is a verdict with the reasoning stripped out. Nobody hands a stranger money based on "trust me, I ran the numbers." I hit something close with financial data, not startup ideas, building Alisio. Early on I'd show one clean output and expect people to believe it on its own. They didn't, because there was no way to check my reasoning against their own judgment. What changed things wasn't proving the logic works, it was showing which specific input actually drove a number instead of just the number itself. People don't distrust automation because it's automated, they distrust a verdict they can't argue with. If a subscriber could see the two or three signals behind an idea instead of just the idea, I'd expect that to convert better than confidence alone. Worth testing before leaning harder on outreach alone.
"People don't distrust automation because it's automated,
they distrust a verdict they can't argue with" — this is
genuinely reframing something I had wrong.
I've been treating trust as a proof-of-demand problem.
You're describing a proof-of-reasoning problem — show the
2-3 signals that drove the conclusion, not just the
conclusion. That's a different fix than what I built
(Market Proof shows an external company validated it;
it doesn't show my own reasoning path).
Going to sit with this one before I decide how to test it.
Thanks for the specific example from Alisio — makes it
concrete instead of abstract.
Love the honest breakdown. The real insight is that the bottleneck is trust, not another feature.
The growth-velocity signal sounds genuinely useful—now I’d focus on 15–20 ideal users and learn exactly what proof would make them pay. 14 issues without missing a week is strong execution.
Thanks — and yeah, that's exactly where I've landed.
14 issues without missing a week was the easy part in
hindsight. Figuring out what proof actually moves 15-20
specific people is the real work starting now.
The personal outreach result is interesting. 72 subscribers from 14 issues with almost all meaningful engagement coming through direct conversations says a lot about how different the two acquisition channels are. Curious what happens when you double down on that instead of adding more features.
Good question, and it's the one I'm testing next. Doubling
down on direct outreach, paired with actually building one
of the ideas myself so there's something concrete to point
to - not just another feature. Will report back on the
thread whichever way it goes.
That’s a useful next experiment, especially with a concrete project attached to the outreach. I’d be interested to see what changes once you run it. If you’re open to continuing the conversation, what’s the best email to reach you at?
Milestones hit (of my journey):
Day ~5: first 12 subscribers via personal outreach
Day ~14: first Vistanova/Peter engagement — first "real company" signal
Day ~30: 50 subscribers
Day ~45: caught claw-code (a fake/art repo) before it went to 50 subscribers — a real save
Day ~50: growth velocity feature shipped after weeks of you pushing me to prioritize it
Day ~55: Razorpay + Gumroad both live
Day ~65: found and fixed the "only 3 of 5 languages" bug — the root cause of repeated ideas
Day ~70: first Pro trial sent free to all subscribers
Day ~80: switched from Gmail to Brevo after diagnosing the spam problem
Day ~90: full HTML redesign + PDF report shipped — the biggest single upgrade in the whole project
Day 97: 14 issues sent, zero missed Saturdays, 72 subscribers, 0 paying customers
Sorry for the odd formatting there - that was meant to
expand on the post itself with the full milestone
timeline, not written as a reply to anyone specific.
Should've put it in the original post. For context,
this is my own project (the post above).