How I Used AI to Build a Portfolio of 13 Niche Websites That Earn Passive Income

Imagine waking up to a dashboard showing $2,300 in ad revenue, $1,400 in affiliate commissions, and zero direct work from the previous day. That’s the reality I’ve built over 18 months with a portfolio of 13 niche websites, each powered by a coordinated mix of AI content generation, automated SEO optimization, and scheduled publishing. The total time investment? About 12 hours per week once the system was in place. This isn’t a get-rich-quick fantasy—it’s a repeatable blueprint that leverages the current sweet spot of large language models and search engine algorithms. In this article, I’ll walk you through the exact framework I used: how I selected niches where AI content actually ranks, the tools and workflows that produce 100+ articles per month per site, and the monetization strategies that turn organic traffic into passive income. By the end, you’ll have a clear, actionable plan to start your own AI-powered website empire.

1. Selecting Profitable Niches with AI-Powered Research

The foundation of any successful niche site is a market with strong search demand, low competition, and high advertiser intent. Instead of manually sifting through keyword tools, I used a three-step AI-driven process. First, I fed a list of broad categories (e.g., “home brewing,” “indoor plants,” “budget travel”) into a custom GPT-4 prompt that returns a table of potential sub-niches with estimated monthly search volume, CPC, and competition score from Semrush data. Second, I cross-referenced these with an AI tool like Keyword Insights that clusters related queries into topical groups—this ensures I can build content clusters that Google rewards. Finally, I used Ahrefs’ API connected to a Python script to filter for niches where the top 10 results have an average Domain Rating below 30, indicating low authority competition.

Data points from my portfolio: the most profitable niche (battery recycling) has an average CPC of $4.80 and only 12,000 monthly searches, but I captured 35% of that traffic within 6 months. The key is to target “long-tail clusters” rather than single keywords. For example, instead of “how to recycle batteries,” I built articles around “recycle lithium-ion batteries at home,” “local battery drop-off centers,” and “mail-in battery recycling kits.” AI helped me generate 50+ related queries from one seed term using a tool like AnswerThePublic combined with GPT-4. I recommend starting with 3–5 niches, each with at least 10,000 monthly searches in the cluster and a clear monetization path (affiliate products, high-display RPM, or digital product).

Tool stack: Semrush (competition analysis), Keyword Insights (clustering), GPT-4 (idea expansion), Ahrefs API (backlink filtering).
Red flag: Avoid niches dominated by Amazon or major .gov sites—AI content struggles to compete there.
Example: One of my sites targets “DIY terrarium supplies,” with an average CPC of $2.10 and 8,000 monthly searches—but 70% of the SERP is weak Pinterest pages.

2. AI Content Generation at Scale (Without Sacrificing Quality)

Once niches are locked, the real leverage comes from a systematic content pipeline. I use a combination of Claude 3.5 Sonnet for long-form articles (1,500–2,500 words) and GPT-4 Turbo for shorter listicles and FAQs. The workflow: For each niche, I maintain a Google Sheet with 100+ target keywords. A custom Zapier automation sends each keyword to a GPT-powered script that generates a detailed outline (H2s, H3s, bullet points, and a meta description). I review the outline in batches (takes 2 hours for 50 outlines) and approve them. Then a second automation feeds the approved outlines back to Claude with a strict system prompt that enforces E-E-A-T signals: first-hand experience, specific examples, data citations (from real sources like government stats or peer-reviewed studies), and a conversational tone.

Quality control is non-negotiable. I run every article through Originality.ai (AI detection score must be below 20%) and Grammarly Premium for readability. Then a human editor—a freelance writer I pay $15/hour—spends 5–10 minutes per article adding personal anecdotes, local references, or unique photos. This hybrid approach keeps costs low (about $3 per article including AI API costs and human review) while maintaining the depth that Google’s Helpful Content Update demands. Over 13 sites, I produce 130 articles per week. The key metric: after 3 months, the average article ranks on page 1 for its primary keyword 40% of the time.

Step 1: Batch keyword research for 100 terms per niche.
Step 2: AI generates outlines with strict formatting (H2, H3, bullet points, FAQ schema).
Step 3: Human review of outlines (approve/revise in bulk).
Step 4: Claude generates full articles with citations and real data.
Step 5: Automated plagiarism check and human tweak for uniqueness.
Step 6: Schedule via WordPress API with Yoast SEO meta data.

3. SEO Optimization with AI: From On-Page to Internal Linking

Publishing content is useless if it doesn’t rank. I automate 80% of on-page SEO using a combination of Surfer SEO’s API and a custom Python script. After an article is generated, the script sends the text to Surfer SEO’s content score endpoint, which returns a list of missing keywords (terms that top-ranking pages use but mine doesn’t). The script then feeds those missing keywords back into Claude with a prompt to naturally incorporate them into the existing text. This iterative loop continues until the Surfer score reaches 75+ (out of 100). The entire process takes 30 seconds per article and has boosted my average first-page ranking rate from 22% to 41%.

Internal linking is another area where AI shines. I built a simple tool using OpenAI’s Embeddings API that calculates semantic similarity between every article on a site. It then generates a “related posts” section and automatically inserts contextual hyperlinks in the first 200 words of each article. For example, an article about “best terrarium plants” will link to “how to layer terrarium soil” if the embedding cosine similarity is above 0.85. This has increased average pages per session from 1.3 to 2.1 across my portfolio. Additionally, I use Link Whisper (a WordPress plugin) as a backup for manual linking, but the AI-driven system handles 90% of the work.

Tool: Surfer SEO API + custom Python script for keyword gap analysis.
Metric: Target Surfer score ≥75 for every article.
Internal linking: Embeddings-based similarity >0.85 triggers automatic links.
Result: 15% increase in organic traffic after implementing AI linking.

4. Automated Publishing: The Engine That Runs 24/7

With content ready and optimized, the final bottleneck is publishing. I use WordPress’s REST API connected to a Make.com (formerly Integromat) scenario. Every day at 6 AM, the scenario checks a Google Sheet for articles that have passed all quality checks (Surfer score ≥75, AI detection <20%, Grammarly score >90). It then creates a draft post, sets the category, tags, featured image (generated by DALL-E 3 with a prompt based on the article title), and schedules it for 10 AM the next day. I also use Yoast SEO’s API to auto-fill the meta description and focus keyphrase. The entire pipeline publishes 10 articles per day per site without any manual intervention.

For a portfolio of 13 sites, that’s 130 articles daily. I stagger the schedules to avoid overwhelming the server—each site publishes 2–3 articles per day at different hours. The cost? About $0.05 per article for DALL-E images and $0.02 for API calls. The key insight is to use a separate subdomain for each niche (e.g., terrariumguides.com, batteryrecycler.com) to isolate domain authority and prevent a bad site from dragging down others. I also set up IndexNow via the Rank Math plugin so Google gets notified instantly. Within 48 hours, 90% of my articles are indexed. The result: within 6 months, my portfolio was generating 120,000 monthly pageviews, translating to $1,800 in AdSense and $900 in affiliate commissions.

Schedule: Use Make.com to check a master sheet and push to WordPress API.
Images: DALL-E 3 with prompt templates (e.g., “high-quality photo of [topic] in natural light, realistic”).
Indexing: Rank Math + IndexNow for instant Google notification.
Server: Use a VPS with 4GB RAM to handle 13 sites and 130 daily posts.

5. Monetization: Turning Traffic into Passive Income

Traffic alone doesn’t pay bills. I monetize each site with a mix of display ads, affiliate marketing, and digital products. For display ads, I use Ezoic (once a site hits 10,000 monthly pageviews) and then Mediavine at 50,000 pageviews. The average RPM across my portfolio is $22, but niche matters: one site about “emergency food storage” has an RPM of $48 because of high-value advertisers. For affiliate income, I promote products via Amazon Associates and niche-specific programs (e.g., ShareASale for terrarium supplies). I use AI to generate comparison tables and “best of” lists that convert 2–3% better than standard reviews. A custom GPT prompt creates a table with price, rating, features, and a pros/cons list, then inserts it as HTML into the article.

Digital products provide the highest margin. For example, on my “DIY terrarium” site, I created a $