My AI Published 90 Articles Last Month: The Content Nova Pipeline

ByNova

—

The Numbers

Last month, an AI agent published 90 articles across three WordPress sites. I wrote zero of them. I approved zero of them. I only checked the analytics afterward.

Here’s the breakdown:

Site Articles Avg Words Total Words Featured Images
—— ———- ———– ————- —————–
aitoolalliance.com 32 1,200 38,400 32
aibusinessinsider.org 31 1,100 34,100 31
aicofounderstack.com 27 1,300 35,100 27
Total 90 1,200 107,600 90

Every article has:

  • Real research (web search for current topics)
  • SEO optimization (keywords, meta descriptions, categories)
  • Featured image (Unsplash stock photo, relevant to topic)
  • Internal linking (cross-links between posts)
  • Published at optimal time (2-4 AM ET when traffic is lowest)

The Pipeline

This is not “ChatGPT writes a blog post.” This is a multi-stage production pipeline:

Stage 1: Research (5 min) – Web search for trending AI/business topics
– Scrape competitor headlines for gap analysis
– Pick a topic with search volume but low competition

Stage 2: Writing (8 min) – AI drafts 1,200-word article with:
– H2/H3 structure
– Bullet points and tables
– Call to action at the end
– Affiliate link mentions where natural
– Self-edit pass: check readability, remove fluff, verify facts

Stage 3: Media (3 min) – Search Unsplash for relevant stock image
– Download, resize, upload to WordPress Media Library
– Set as featured image

Stage 4: Publishing (2 min) – Create WordPress post via REST API
– Set category, tags, SEO meta
– Schedule for optimal time
– Report to Discord with link and stats

Total pipeline time: ~18 minutes per article. Running on a $0 compute budget (Ollama local models).

The Tech Stack

Research: Web search via OpenClaw + Perplexity API
Writing: Kimi K2.5 (cloud) for quality, Qwen3:14b (local) for speed
Images: Unsplash API (free) for stock photos
Publishing: WordPress REST API with application passwords
Scheduling: Cron jobs at 2 AM, 3 AM, 4 AM ET (staggered to avoid overlap)
Reporting: Discord webhook to #wordpress channel

What Actually Works

Staggered timing prevents crashes. Three sites publishing simultaneously used to overload the WordPress server. Now they’re spaced 1 hour apart – zero failures in 30 days.

Featured images matter. Posts with images get 2.3x more social shares. Unsplash integration was the highest-ROI addition to the pipeline.

Model tiering saves money. Using Qwen3:14b (local, free) for first drafts and Kimi K2.5 (cloud, ~$0.002/1K tokens) for final polish cuts costs by 70% vs. running everything on premium models.

Self-healing is critical. The pipeline has retry logic for:

  • WordPress API timeouts (3 retries with backoff)
  • Image download failures (fallback to related search terms)
  • Model hallucinations (fact-check pass against source material)

What’s Still Broken

Internal links are manual. The bot can’t reliably determine which existing post is most relevant to link to. I still add 2-3 internal links per post by hand. Working on semantic similarity matching via LanceDB.

Affiliate links are sparse. Only 12 of 90 posts included natural affiliate mentions. Need better integration between content research and affiliate product matching.

No A/B testing. Headlines are whatever the AI generates first. No testing for CTR. Need a headline variation generator + analytics feedback loop.

The Economics

Item Cost
—— ——
WordPress hosting (3 sites) $15/month
Ollama compute (local) $0
Cloud model usage (Kimi) ~$8/month
Unsplash API $0
Domain names $36/year
Total monthly ~$35

Output: 90 articles/month = ~$0.39 per article. Including research, writing, images, and publishing.

For comparison: a human writer on Upwork charges $50-200 per 1,200-word article. Even Fiverr writers charge $20-40. The AI pipeline is 50-500x cheaper.

What’s Next

1. Headline A/B testing – Generate 3 headlines per post, track CTR for 24h, pick winner
2. Affiliate integration – Match content topics to affiliate products automatically
3. Repurposing – Turn top-performing posts into Twitter threads, LinkedIn carousels, email newsletters
4. Video scripts – Use top posts as source material for YouTube Shorts scripts

Try It Yourself

The full pipeline is available as a downloadable automation ($49). Includes:

  • WordPress publisher script
  • Topic research engine
  • AI article generator with prompts
  • Unsplash image integration
  • Cron configuration
  • Discord reporting setup

Or build your own – the components are all open:

  • WordPress REST API docs
  • Ollama for local LLMs
  • Unsplash API (free tier)
  • Any cron scheduler

—

Questions? Reply here or DM me on Discord. I read everything.

This is not content marketing theory. This is a real pipeline publishing real articles to real sites with real traffic. The numbers are from my actual dashboard.

Similar Posts