I Used Agentic AI to Build a Smart Ring Affiliate Site. Here’s What Actually Happened.
This post contains affiliate links. If you buy through a link, I may earn a commission at no extra cost to you.
I spent roughly $300 and the better part of two months building a smart ring review site almost entirely with an AI agent. It earned $4.08.
That’s not a typo, and it isn’t a cautionary tale about AI being useless — it’s closer to the opposite. The agent did nearly everything I asked, quickly. The catch is that “build the website” and “build a business” turned out to be two different jobs, and only one of them is automatable right now. Here’s the whole thing, numbers and all.
The experiment in one paragraph
The site you’re reading — Smart Rings Reviews — was the second project in a personal experiment: could a non-programmer point an agentic AI at a real, monetizable niche and stand up a working affiliate site without hand-coding it? I picked smart rings because the category is small, fast-moving, and full of genuine buying decisions (Oura, Ultrahuman, RingConn, Samsung Galaxy Ring, and the rest). Claude Cowork handled research, drafting, SEO formatting, images, and WordPress publishing. I directed, verified, and paid the bills. This very post — and the archive notice now sitting at the top of every page — were built the same way. The full project log lives on GitHub.
Why I bet on smart rings
The niche math looked good on paper:
- Real price points. $200–$500 rings mean affiliate commissions actually worth chasing.
- High buyer intent. People researching smart rings are usually close to purchasing, not just browsing.
- A fast news cycle. New models, subscription changes, and FDA clearances reward fresh content.
- Narrow enough to look authoritative. A focused site could, in theory, build topical authority quickly.
That last assumption is the one that didn’t survive contact with reality.
How the agentic workflow actually worked
This wasn’t “press a button, get a business.” It was human-directed and AI-executed — closer to managing a very fast, very literal junior employee than to passive automation.
What the AI did well
Drafting research-backed posts, formatting them for WordPress, applying basic SEO, generating featured images, and scheduling everything through the WordPress REST API — end to end, hands-off, as long as I batched the work into chunks of three or four. I also built a custom “skill” that pre-loads the site’s voice, structure, and rules so the agent doesn’t relearn the basics every session. That part genuinely works.
Where it still needed me
Constant verification. Early on, when I let it write product links without supplying the URLs myself, it produced confident links that 404’d or pointed at the wrong page — the same overconfidence that, in an earlier project, talked me into a system command that broke my PC. Confident phrasing is not evidence. Every link, price, and claim still needed a human check before it went live.
What the traffic actually looked like
Over a typical 28-day window the site pulled about 252 visitors. Here’s where they came from:
| Traffic source | Share of visitors |
|---|---|
| Direct | 40% |
| Organic social | 28.6% |
| Organic search | 20.8% |
| AI assistants (ChatGPT, etc.) | 5.5% |
| Other | 5.1% |
Search Console tells the real story: roughly 2,400 impressions, but only 10 clicks — about a 0.4% click-through rate — and 53 unique visitors from search over the period. Translation: Google is indexing the content and starting to surface it, but a brand-new site has no authority, so it sits on page five for anything competitive. Impressions were climbing; clicks stayed basically flat. That gap is the single most important number on the project, and AI can’t close it for you.
Cost vs. return — the part most “AI income” posts skip
| Metric | Result |
|---|---|
| Approx. project spend | ~$300 |
| Affiliate earnings | $4.08 |
| Net result | −$295.92 |
| Revenue recovered | ~1.36% of spend |
| Break-even gap | ~73.5× more revenue needed just to recover costs |
Recovering about 1.36% of what I put in isn’t a rounding error you grow out of next month — earnings would have needed to be roughly 73.5 times higher just to break even. No amount of additional AI-generated posts changes that equation on its own, because volume was never the bottleneck. Demand, trust, and ranking were.
So did the AI actually work?
Yes and no, and the distinction is the whole lesson. The machine worked: an AI-assisted publishing pipeline, run by someone who doesn’t code, is completely real and frankly a little remarkable. The business didn’t: “AI passive income” is a misleading frame. This was not passive — I spent real hours reviewing outputs, debugging, setting up accounts and hosting, checking links, and fixing mistakes. AI can build the machine. It does not create demand.
The bottom line
I’d call this a successful experiment and a failed business, and I’m fine with that — the learning was the point. If you’re tempted by the “spin up an AI cash-machine site” pitch, treat it as an experiment first and a business a distant second. The hard parts of affiliate marketing — product selection, niche research, building trust, earning SEO authority, decent affiliate programs, compliance, analytics — are still hard, and still yours. The AI just gets you to the starting line faster. If you want the unvarnished build log, including the project that broke my PC, it’s all on GitHub.
Frequently Asked Questions
Did the AI build the entire site by itself?
No. The agent did the heavy lifting — research, drafting, formatting, images, and publishing — but every batch was human-directed and human-verified. Calling it fully autonomous would be inaccurate.
How much did the site actually earn?
$4.08 in affiliate earnings against roughly $300 in spend, for a net of about −$295.92 — around 1.36% of costs recovered.
Is “AI passive income” real?
Not in the way it’s usually sold. AI removes production friction, but it doesn’t generate traffic, trust, or rankings. The income side was neither passive nor, in this case, profitable.
What did the AI do best?
Speed and consistency on production tasks: researching, drafting, SEO formatting, generating images, and scheduling posts through the WordPress REST API in repeatable batches.
Would publishing more posts have fixed the revenue?
Unlikely. Volume wasn’t the limiting factor — domain authority, buyer trust, and time were. More posts without those is just more pages no one ranks for yet.
Net, I’m down $295.92 — but call it tuition. For what I learned about directing an agentic AI workflow, it was worth every dollar I didn’t make back.