Ran 5 product-validation demos against 250 Census personas each. The top-scoring segment was never the obvious one. Shift workers over office. Rural over suburban. Mid-career over new freelancers. 5 for 5. You don't know your audience until you test it.
All 4 products now serve /llms.txt. Betting AI engines cite structured data over keyword-ranked blog posts. If a model pulls our persona scores into an answer, that's distribution we didn't pay for. No idea yet if it works. Watching the crawler logs.
5 product-validation demos. 250 personas each.
The top-scoring segment was never the obvious first guess.
Shift workers beat office workers.
Rural beat suburban.
You don't know your audience until you test it.
Tested a freelancer cashflow planner against 250 Census personas. Mid-career freelancers scored PSF 80. New freelancers scored 52. The people who most need cashflow planning are the ones with enough history to trust the forecast.
Shipped /llms.txt on all 3 of my projects.
All 3 return 200 every day for a week.
If ChatGPT, Claude, Perplexity start citing them, that's distribution I'm not paying for.
#buildinpublic
Ran a contract workflow tool through 250 personas.
Top segment scored PSF 81: Ops-Heavy Revenue Teams.
Their pain isn't drafting. It's invisible approval latency. They lose deals because nobody can tell them where legal review is stuck.
#buildinpublic
8 weeks ago I deployed CF WAF rules on a pre-launch project to cut scanner noise.
Weekly PV: 4472, 3375, 3267, 3010, 2205, 1687, 962, 378.
-92% from baseline. Floor still dropping. When a real user lands, they'll stand out.
The bitter lesson in 26 words:
Don’t be distracted by human knowledge, as AI has been historically.
Instead focus on methods for creating knowledge that scale with computation, like search and learning.
SEO week 2: app pageviews dropped 12 to 8. Looked like a plateau.
SEO week 3: pageviews doubled to 16. No new content shipped.
The flat week was just the indexing lag finishing. Don't kill SEO on one bad reading.
App analytics: 7 page views this week.
Cloudflare edge: 2,782.
The gap is scanners. They don't fire JS.
Easy to feel like you're growing if you watch the wrong number.
Tested 5 demos with 250 personas each. The gap between best and worst audience segment was bigger than the gap between a good product and a bad one. Pick the wrong audience and a great product fails. Pick the right one and a mediocre product gets traction.