on july 3, 2026, i shipped my first product: citeable.
the idea: generate llms.txt and schema.org for your site so ai assistants like chatgpt and perplexity can understand it, and maybe cite it.
the mvp took one day. crawl a site, score it, sell the files behind a stripe checkout.
this is the story of what happened next.
the first proof: my wife's website
my wife carmen is a photographer. i ran her site through citeable.
score before: 72.
score after: 100.
that before/after became the hero of my landing page. it was real, and it was the best thing on the page.
the launch
i prepared everything: hacker news, reddit, product hunt, tinylaunch. then i switched to x-first and wrote a 10-day launch calendar.results: ~100 visitors, 0 sales.
lesson: first time i had to face it. distribution is hard.
the pricing rollercoaster
in 4 weeks i changed pricing 5 times. a one-time tier, then a studio tier at 199€, then 3 plans with a monthly agency offer, then i killed the pro plan and dropped solo to 14,90€, then switched everything to usd.
none of it moved the needle. when nobody is buying, pricing is rarely the problem.
the cold email machine
no audience, so i built outreach. google maps to find local businesses (photographers, lawyers, plumbers...) in 12 french cities, a scan of each site, then a 3-email sequence.
490 prospects in the pipeline. when the sending tool's paid plan ran out, i rebuilt the sender myself with smtp and imap. cost: 0€.
the email that stopped everything
one agency replied. not to buy. to tell me my pitch was wrong.
my emails said llms.txt was "the file ai reads first". then i checked the research. a 2026 ahrefs study: 97% of llms.txt files got zero ai-bot traffic across 137k sites. se ranking: no significant link between llms.txt and ai citations
across ~300k domains. google said it doesn't support it.
my own faq already admitted most ai crawlers ignore it. my cold emails said the opposite, to the audience most likely to know.
i stopped all sending that day, and rewrote the copy to claim only what holds.
btw, i also got my first cold email reply. i'll let you translate it 🫠
why i shut it down
no sales, no replies, and i'd lost trust in the product. so i moved on to my next project: proofbench.
what i'm keeping
shipping in a day works. the hard part was never the code.
a real proof case beats any marketing copy.
don't build outreach before you have something people ask for.
never claim what you can't prove. the market will check.
distribution is hard.
next up: building proofbench in public. subscribe to follow along, the wins and the failures.