I Run a Content Business on $20/Month of AI — Here's My Exact Stack

Most solo creators I know spend $50–150/month on AI subscriptions. A writing tool here, an image generator there, a research assistant on top. It adds up fast, and most of it overlaps.

I spend $20. Two subscriptions. Over 500 articles published, 700,000+ cumulative views, and active sponsorship deals — all running on a stack that costs less than a single dinner out.

Here's the exact breakdown.

---

The $20 Stack

Claude Pro — $20/month. This is the engine. Writing, coding, research, strategy, content planning, sponsor outreach drafts — Claude handles roughly 80% of everything I produce. I write in both Japanese and English, and Claude is the only model I've found that handles bilingual content work without constant babysitting.

Perplexity Pro — $0/month (free tier). I use Perplexity exclusively for fact-checking and source gathering. When I need to verify a claim, find a primary source, or survey what's been published on a topic, Perplexity's citation-first interface is faster than manual search. The free tier covers everything I need.

That's it. No ChatGPT Plus, no Midjourney, no Jasper, no copy.ai. Two tools, one paid.

---

What I Cancelled and Why

I've tried most of the popular options. ChatGPT Plus was the first to go — once Claude could handle long-form writing and code in a single conversation, paying for both made no sense. I ran HIX AI for a sponsored review and kept it for a month afterward; the SEO features were decent, but they duplicated what I could already do with Claude plus a free keyword tool. (I wrote about that decision in detail in my HIX AI review.)

The decision rule I settled on: if one tool covers 80% of a use case, I don't pay for a second tool to cover the remaining 20%. That remaining 20% either gets handled by a free tier elsewhere, or I simply skip it.

This sounds obvious, but most creators don't audit their stack this way. They subscribe to tools for specific features — "this one has a better image generator," "that one has better templates" — without noticing the core capabilities overlap almost entirely.

---

The Free-Tier Trick: Chinese LLMs as a Second Layer

Here's where it gets interesting. For tasks where I don't need Claude's full reasoning — reformatting text, generating metadata, translating short passages, bulk summarization — I route those to Chinese LLMs through their free API tiers.

GLM-5.1, MiniMax, and Kimi all offer generous free API access. I use Claude Code as the harness and swap in these models for the heavy, repetitive work. The result is a two-layer system: Claude handles anything that requires judgment, nuance, or complex reasoning; the free Chinese models handle the mechanical throughput.

I wrote about this workflow in detail in my piece on cutting Claude Code API costs by 60%. The key insight is that most "AI tasks" in a content workflow don't actually need a frontier model. They need a competent model that's free.

---

Monthly Revenue vs. Cost

I won't share exact sponsor revenue, but here's the math that matters: my AI tooling cost is $20/month. My sponsor deals range from $35 to $200 per article, with my pricing structure going up to $1,500 for premium packages. Even a single $200 deal per month gives me a 10x return on my entire AI infrastructure.

The real leverage, though, is time. Before consolidating to this stack, I spent roughly 2–3 hours per week managing, switching between, and learning different AI tools. That's 8–12 hours/month of overhead that now goes directly into producing content.

---

The Takeaway

If you're spending more than $20/month on AI tools, you're probably paying for overlap. Audit your stack with one question: does this tool do something my primary tool genuinely cannot?

If the answer is "it does it slightly better" or "it has a nicer UI for this one thing" — that's not worth a subscription. That's a preference dressed up as a need.

One good model. One free research tool. Free API tiers for the mechanical work. That's the stack.