The AI Writing System Behind a One-Person Company — How I Produce 10 High-Quality Pieces Per Week

Let's start with the numbers.

In February 2026, my content output: 8 long-form pieces (blog posts/LinkedIn articles), 34 tweets, 4 newsletter issues. Total word count: approximately 52,000 words (combined Chinese and English).

Time I spent on content: about 12 hours per week.

This efficiency isn't talent, and it isn't brute force. In the first half of 2025, I maxed out at 2 long-form pieces plus a few tweets per week, spending 20+ hours. The difference comes from an AI writing system I spent six months iterating into shape.

This article takes the entire system apart — from topic selection to first draft to quality review to publishing, how each step works, what tools are involved, how much AI handles, and how much stays with me.

---

Context: Why a One-Person Company Needs High-Volume Content

I run three business lines simultaneously: ArkTop AI, JewelFlow, and the Solo Unicorn Club. Each one needs content as a growth engine:

• ArkTop AI: Technical blog posts for SEO-driven acquisition, with a 2–3 month conversion cycle
• JewelFlow: LinkedIn industry analyses to attract jewelry industry decision-makers
• Solo Unicorn Club: Daily Twitter/X posting to maintain brand presence and drive community growth

Across all three, maintaining a healthy content cadence means 10 pieces per week is the baseline. Drop below that, and one business line's content visibility goes dark.

The problem: I'm one person. Beyond writing, I'm also writing code, handling customer support, managing the community, and dealing with business development. My maximum time allocation for content is 12–15 hours per week.

So the system's core design goal is singular: maximize content output and quality within a fixed time budget.

---

Three Principles

Principle 1: Humans Own Topic Selection and Final Review; AI Handles Drafting and Formatting

This is the core division of labor for the entire system.

What I do (irreplaceable):

• Topic decisions: what to write, what to skip, content direction for each business line
• Core arguments: each article's central thesis and unique insights
• Final review: I personally read every piece before it goes live to confirm quality
• Injecting personal experience: data, stories, hard-won lessons — these are things AI can't fabricate

What AI does (leverage):

• Research and material gathering: competitor moves, industry data, tool updates
• First draft generation: based on outlines and arguments I provide, generate a complete draft
• Format adaptation: transform the same content into blog, tweet, LinkedIn, and other formats
• Multilingual versions: generate the English version after writing in Chinese (or vice versa)

The key ratio: for a 2,000-word long-form piece, AI's draft covers roughly 60% of the final content. The remaining 40% is what I rewrite, revise, or add from scratch. That 40% is precisely what determines the quality ceiling — personal experience, unique perspectives, specific data.

Principle 2: Batch Production, Not Piece-by-Piece Polishing

The biggest time sink in content creation is context switching. My approach is "theme days" for batch production:

• Monday (3 hours): Topic selection + outlines. Plan all topics for the week; write 3–5 sentences of core argument for each piece. This is the most thinking-intensive step — no AI involved.
• Tuesday (3 hours): Long-form first drafts. Feed outlines to Claude for draft generation; I read through each one, marking revisions.
• Wednesday (2 hours): Long-form refinement + batch tweet generation. Add personal data and experiences, rewrite paragraphs that feel too "AI-generated," and extract tweet material from the long-form pieces.
• Thursday (2 hours): Newsletter assembly + English version generation.
• Friday (2 hours): Formatting, publishing + next week's preliminary research.

Principle 3: The Quality Bar Is Non-Negotiable

Volume is up — so how do you maintain quality? I have three checkpoints:

Checkpoint 1: Anti-AI-voice scan. I've compiled a list of "AI filler phrases" in both Chinese and English — those transitional phrases and hollow expressions that LLMs love to lean on. Over 60 entries between the two languages. Before publishing, I run a script that auto-scans each draft against this list. Anything flagged gets rewritten. This check takes 30 seconds per piece but filters out 80% of the AI aftertaste.

Checkpoint 2: Information density check. My standard: every paragraph must contain at least one specific number, tool name, or personal experience. If a paragraph is nothing but generic platitudes, it gets cut or rewritten. The most common problem with AI drafts is insufficient information density — lots of "technically correct filler."

Checkpoint 3: Read it aloud. The final step before publishing: I quickly read the piece out loud. Anything that sounds awkward usually has one of two problems: the sentence is too long (in Chinese, sentences over 40 characters need to be split; in English, similar judgment applies), or it uses an unnatural written expression. This method is low-tech but reliable.

---

Tool Stack Breakdown

Use Case	Tool	Monthly Cost	Why This One
Long-form drafts + rewrites + English versions	Claude Pro (Max plan)	$100	Best writing quality among all LLMs, strong long-form coherence
Batch tweet generation + quick rewrites	Claude API (Sonnet)	~$15	API is more efficient for batch processing, Sonnet at $3/M input is cost-effective
SEO keyword research	Ahrefs (Lite plan)	$129/month	ArkTop blog's organic traffic depends on SEO
Content formatting + publishing	Notion + Typefully + Buffer	~$18	Notion as content library, Typefully for tweet scheduling, Buffer for cross-platform publishing
Anti-AI-voice scan	Custom Python script	$0	200 lines of code scanning a blocked-phrases list
Newsletter	Kit (formerly ConvertKit)	$29	Segmented sending, clean open rate analytics
Images	Claude / Flux.2 via OpenRouter	~$5	Generated on demand; most content doesn't need images
Total		~$296/month

The two biggest line items are Claude Max ($100) and Ahrefs ($129). Claude Max's 20x usage allowance is essential at my output volume — the Pro plan's quota isn't enough. Ahrefs is the core acquisition tool for ArkTop, with clear ROI.

If your content volume isn't this high, Claude Pro ($20) + free SEO tools (Ubersuggest free tier, Google Search Console) will do. Total cost can stay under $50.

---

Results

September 2025 through February 2026 — six months of production data:

Output volume:

• Long-form (blog + LinkedIn): 8 pieces/month average
• Tweets (Twitter/X): 32/month average
• Newsletter: 4 issues/month average
• Average monthly total word count: ~50,000 words (Chinese + English combined)
• Average weekly time on content: 12 hours

Efficiency comparison (before vs. after AI writing system):

• Time per long-form piece: from 4–5 hours down to 1.5–2 hours
• Time per tweet: from 20–30 minutes down to 8–12 minutes
• Weekly output: from 2–3 long-form + 8–10 tweets, up to 2 long-form + 8 tweets + 1 newsletter issue (weekdays only, weekends off)

Content performance (ArkTop AI blog as example):

• Monthly organic search traffic: from 1,200 UV to 4,800 UV
• Blog-driven registrations: 45 new users/month average
• Average article read time: 4 minutes 12 seconds (industry average ~2 minutes 30 seconds)

ROI:

• Monthly new user value from blog (estimated by LTV): ~$3,600
• Monthly content tool cost: $296
• ROI: ~12:1

---

Lessons from the Trenches

Pitfall 1: Over-relying on AI drafts early on — articles full of "correct but empty" prose

When I first started using AI for writing, I'd give Claude a topic and let it write the whole thing. The output looked the part — well-structured, grammatically correct, views that weren't exactly wrong. But after reading, you couldn't recall a single concrete takeaway. It was wall-to-wall generalities: "AI can help improve efficiency," "choosing the right tools is important."

The turning point was when I started requiring myself to include "me" in every outline: my specific data, my mistakes, my choices that diverged from conventional wisdom. AI can't produce these things, but they're precisely what makes readers keep reading.

Pitfall 2: Batch production led to quality inconsistency

One week I pushed to get all 12 pieces out the door and rushed through the final review. After publishing, I discovered one long-form piece had an incorrect tool price (DeepL's old pricing), and a tweet cited an outdated statistic. Both were corrected quickly, but the episode made me realize the tension between volume and quality is permanent.

Current rule: weekly output caps at 10 pieces. Beyond that, I'd rather delay than lower my review standards.

Pitfall 3: English versions can't just be translations

Initially, I had Claude translate my Chinese long-form pieces directly into English for LinkedIn. The response was poor. The reason: Chinese readers and English readers care about different entry angles. My current process: after finishing the Chinese version, I write a new English outline (with 30–40% structural changes), then have Claude generate the English draft from the new outline.

Pitfall 4: Underestimating the importance of topic selection

For a while, I padded my output with topics I had no firsthand experience with. Those articles were painful to write — without real data or personal stories, I could only cobble things together from secondary research, and no amount of editing could remove the AI taste. Lesson: only write about topics where you have genuine experience. A one-person company's content edge is the authenticity of a founder who's actually in the arena.

---

Advice for Getting Started

Step 1: Figure out how many hours per week you can dedicate to content. Then work backwards to determine output volume. My benchmark is roughly 800–1,000 words of high-quality output per hour (with AI assistance) — use that to estimate.

Step 2: Start with Monday topic selection. This is the single most important step. The quality of 10 topics directly determines the ceiling of the week's content. One hour of serious topic ideation is worth more than five hours polishing an article with a weak premise.

Step 3: Build an "AI filler" checklist. Collect AI's favorite crutch phrases as you spot them, compile them into a checklist, and run every article through it before publishing. After three months, this habit will noticeably sharpen your content's distinctiveness.

---

Final Thoughts

The core tension in content production for a one-person company is the conflict between "one person's time" and "multiple business lines' content demands." The AI writing system doesn't solve the question of "who writes" — you are always the one truly writing — but rather "where your time goes."

$296/month in tool costs, 12 hours per week, 50,000 words per month. Behind these numbers is a division of labor: I handle thinking, judgment, and injecting real experience; AI handles research, drafting, and format conversion. Humans lead, AI executes — in content production, "leading" means topic selection and final review; "executing" means going from outline to draft.

Many members of the Solo Unicorn Club use a similar approach to content. The common feedback: AI made "content-driven growth" feel achievable — what once seemed insane (10 pieces a week) turns out to be a systems engineering problem.

What does your current content production workflow look like? Which parts do you think could be more efficient?