The Math Problem
When we started building out the network, the math was unforgiving. A traditional affiliate site operation requires:
- Research and keyword planning
- Content production
- Editing and quality control
- Image creation
- Technical SEO and link building
- Site maintenance and updates
- Monetization and conversion optimization
For one site, that's a team. For 40 sites, it's a department.
We have a small team. So we had to figure out how to multiply our output without multiplying our headcount.
Two years in, here's the workflow that lets us run 40+ sites with the team we have. None of this is novel in isolation. The integration is what makes it work.
The Stack
We use a layered approach where each tool does what it's best at:
- Claude (and similar models) for research, outlining, and first drafts
- Custom Python scripts for data processing and image manipulation
- Nuxt-based templates for site architecture
- Airtable for editorial pipeline tracking
- Ahrefs and Search Console for SEO intelligence
- Image generation (via MCP) for hero and inline imagery
The integration point is a content brief — a structured document that flows through the pipeline.
The Content Brief
Every article starts as a brief, not a blank document. The brief is generated by combining:
- The target keyword and search intent analysis from Ahrefs
- SERP analysis of the top 10 results (what's working, what's missing)
- Site context (this site's voice, internal link graph, related content)
- Editorial direction (which cluster this serves, which pillar it links to)
- Specific human instructions (this article must include X, must take a stance on Y)
The brief is detailed. A typical brief is 800-1,500 words covering structure, must-include points, internal links, tone guidance, and SEO targets.
Generating the brief takes about 20 minutes. Writing the article from the brief takes about 2-3 hours of human work (versus 6-8 hours from scratch). The savings compound across hundreds of articles.
Research, Outlining, and First Drafts
For research-heavy content, AI handles the initial pass:
- Summarizing the top 10 SERPs to identify what's covered
- Pulling primary source data from manufacturer sites
- Identifying common questions in forums and reviews
- Proposing article structures based on what's working
We don't accept the first draft as final. The first draft is a starting point. Humans then:
- Add first-person experience and specific examples
- Cut generic statements that don't add value
- Restructure sections where the AI logic didn't fit the topic
- Insert opinions, recommendations, and judgment calls
- Add internal links to relevant site content
- Adjust tone to match the specific site's brand voice
The revision is where the article becomes ours. Without it, the content is generic AI output. With it, the content is human-authored, AI-assisted.
The 80/20 of Quality
Not every article needs the same level of polish. We've tiered our content:
Tier 1: Pillar Content
The cornerstone articles — buying guides, comprehensive comparisons, definitive resources. These get full human attention: detailed brief, full revision, original research, custom imagery, multiple review passes.
Production: 3-5 articles per site per month. These are the articles that drive rankings and revenue.
Tier 2: Supporting Articles
The articles that complete the topic map. Long-tail queries, specific questions, niche subtopics. These get AI first drafts with substantial human revision but don't need the same depth as pillar content.
Production: 10-20 articles per site per month. These capture long-tail traffic and demonstrate topical coverage.
Tier 3: Programmatic Content
Template-driven content for high-volume query patterns. AI handles structure; humans handle edge cases and quality control. Programmatic refresh cycles keep content current.
Production: Scales to hundreds of pages per site where the data supports it.
Each tier uses different tools and workflows. The mistake is applying Tier 1 effort to Tier 3 work — it kills margins. The opposite mistake is applying Tier 3 effort to Tier 1 work — it kills quality.
Image and Asset Production
Every article needs images. Producing them used to be a bottleneck.
Now: each article's brief includes a description of 3-5 images. AI generates the hero image; humans select or refine the result. Inline images get the same treatment, with specific prompts for diagrams, product shots, or comparison visuals.
The image generation runs in parallel with the article draft. By the time the article is ready for review, the images are usually ready too.
The Editorial Review Process
AI content without human review is a liability. We have a multi-step review:
- Self-review by the author — first pass for voice, accuracy, and gaps
- Editorial review — second pair of eyes for quality, internal linking, brand alignment
- SEO check — meta tags, schema, image alt text, internal links verified
- Final fact-check — specific claims, product specs, regulatory statements
The review is where most sites fail. They generate content with AI and skip the human steps. We've found that the review is non-negotiable — it's what separates content that ranks from content that gets hit.
What We Don't Automate
Some things require human judgment, period:
- Site selection and niche strategy: Deciding which sites to build, which to sunset
- Brand voice definition: Establishing how each site sounds, what it stands for
- Final editorial judgment: Which articles deserve promotion, which need more work
- Partnership decisions: Affiliate programs, sponsorships, collaborations
- Crisis management: When something goes wrong (a penalty, a complaint, a security issue)
- Strategic pivots: When a site's economics change and decisions need to be made
These are the high-leverage moments where human attention compounds. Automating them produces a worse business.
The Numbers
After two years running this workflow:
- Content production: ~3x what we produced before AI integration, with similar or higher quality
- Cost per article: ~40% lower than fully human-produced content
- Editorial team size: Same as before (we didn't replace anyone with AI)
- Site count: Grew from ~15 to 40+
- Traffic and revenue: Both grew proportionally
The cost savings aren't from replacing humans. They're from amplifying humans. Each person produces 3x more without working 3x harder.
What I'd Do Differently
Some mistakes we made:
Over-rotating on AI for Tier 1 content early on. AI first drafts of pillar content required so much revision that it would've been faster to write from scratch. We now use AI more selectively for top-tier work.
Under-investing in image generation initially. Stock photography was a quality drag on early sites. Investing in AI-generated imagery was one of the highest-leverage moves we made.
Trying to standardize voice too early. Each site has its own voice. Forcing them all through the same AI template produced generic content. We now build site-specific voice guides that the AI prompts respect.
Not tracking enough editorial metrics. We tracked traffic and revenue but not editorial quality metrics like time-to-publish, revision cycles, or topic coverage gaps. Adding these helped us optimize the workflow.
The Bigger Lesson
AI doesn't replace affiliate site operators. It replaces the tedious parts of the work — research, data synthesis, first drafts, image production — so humans can focus on the high-leverage parts: strategy, voice, judgment, relationships.
Sites that figure out this split thrive. Sites that try to automate everything produce low-quality content farms. Sites that refuse to use AI produce too slowly to compete.
The middle path is the only one that scales.
