234 Days, 119 Posts, $10.50
A project handed $80 to an AI and let it run a website. 234 days later the ledger is public and read live from Stripe: $10.50 gross revenue across 3 transactions, $5 of which was the operator testing checkout. In the same period it autonomously published 119 posts, collected 77 subscribers, and gets about 5 real visitors a day.
📜 Table of Contents
The numbers
moneylab is a public experiment: on 23 March 2026 an AI was given $80 of seed capital and operational control, and told to write the code, the content and the marketing. Every figure is read live from Stripe and the production database, not recalled afterwards.
As of 11 November 2026, day 234:
- Gross revenue $10.50 across 3 transactions
- External revenue $5.00, after subtracting $5.50 of the operator's own test purchases
- 77 email subscribers
- 119 posts published autonomously
- 4,086 persistent memories stored
- 234 days operating
Traffic is the more interesting side. In the last 28 days GA4 logged 887 sessions. 707 of them were direct, which the project itself classifies as automated retrieval: one country, one device type, near-zero dwell time. The rest: 108 from organic search, 31 from AI assistants. Attributable human sessions: 139, about 5 a day.
What it tried, and why none of it worked
Selling. Subscribers, posts, memories and search traffic all grew between July and October. Sales did not move. That is the sharpest line in the ledger: content asset growth fully decoupled from revenue growth.
Paid traffic. A $15 a day cap at $0.40 to $0.80 per click buys 75 to 150 clicks a month. That is too small a sample to read a conversion rate at all.
Social distribution. The Reddit and Threads accounts were closed by the platforms, Dev.to and Medium automation broke, HackerNoon is bot-gated.
Ad networks. Infolinks rejected the site in April 2026 for insufficient traffic.
What this actually shows
It does not show that AI cannot write. 119 posts and 108 organic sessions show the content gets indexed and found.
It shows that content capacity is not traffic, traffic is not revenue, and automation only reaches the first link.
A content site works when three things line up: it can be found (done), people want to click (partly done), and the people who arrive have a reason to pay (not done at all). The project never solved the third, because it never had a specific reason for anyone to buy. It just published a lot.
Running a content site without landing here
Step one: pick the monetisation before the topics. Get this backwards and you end up with 119 posts and nowhere to put a price. There are four models: ads (needs traffic), affiliate (needs buying intent), own product (needs trust), subscription (needs ongoing value). Which one you pick decides which keywords you write.
Step two: target intent, not information. "What is RAG" brings students. "RAG deployment cost comparison" brings people with a budget. Ten thousand visits from the first group can produce zero revenue; a hundred from the second can close a sale.
Step three: do not hand distribution entirely to automation. All four distribution channels in this project failed: two accounts closed, two automation paths broken. Distribution is adversarial. The more automated it is, the faster it gets blocked.
Step four: keep one human in the loop. Someone has to read the weekly search-term report and judge which terms carry commercial value. A fully automated topic picker drifts toward the highest-volume, lowest-value keywords.
My read
The value of this project is that it measured the ceiling of the fully automated content site: 234 days, $5 of external revenue, 5 real visitors a day.
That does not mean content sites cannot make money. It means automating production without automating judgement and distribution produces a very low-traffic library.
ainomam is also a content site with over 2,000 posts. After reading this ledger I am more sure of one thing: the moat is not volume, it is the commercial judgement behind keyword selection. Twelve posts targeting problems people are actively trying to solve beat 119 posts nobody searches for.
Written by our editorial team; tools listed here are tested or verified against public sources. Links point to official sites or GitHub repos for reference only — no paid placements.