180+ articles per month at under 1 cent each: the SYDDK automation pipeline

A news site that runs itself

SYDDK covers four cities in Southern Denmark: Sonderborg, Aabenraa, Haderslev, and Tonder. It publishes over 180 articles per month. Local news, events, business updates, community stories.

It has no human writers.

The entire editorial pipeline, from topic discovery to research to writing to image sourcing to SEO optimization to publishing, runs automatically. Built on n8n, powered by AI, monitored by humans but operated by machines.

We built it in one week. The cost per article is under 1 cent. And it has been running at 90% autonomy since launch.

This is not a thought experiment about what AI content could do. This is a production system that publishes real content for real readers in real cities every single day.

The problem: local news is dying

Local journalism has an economics problem. A reporter costs money. An editor costs money. A photographer costs money. Multiply that across four cities and you have a payroll that advertising revenue cannot support, especially in a small Danish market.

The result is predictable. Local papers close. Coverage shrinks. Communities lose their information infrastructure. The stories that matter, council meetings, business openings, road closures, event announcements, go uncovered.

We did not set out to replace journalists. We set out to cover the stories that were not being covered at all. The gap between what local communities need and what traditional media economics can deliver is massive. AI does not close that gap entirely, but it fills in a lot of the blanks.

The tech stack

WordPress with a custom Full Site Editing block theme

We built a theme specifically designed for automated publishing. Clean layouts, fast load times, proper schema markup, and an admin interface that makes it easy to review and edit content when needed. No page builders. No bloated plugins. Just a clean WordPress installation that does exactly what it needs to do.

n8n as the automation backbone

Every step in the editorial pipeline is an n8n workflow. Topic discovery, research, writing, image sourcing, SEO optimization, and publishing are all separate workflows that chain together. This modular approach means we can update any step without touching the others. If we want to change the writing model, we update one workflow. If we want to add a new image source, we add one node.

AI for research and writing

The system uses large language models to research topics and write articles. But it is not "ask ChatGPT to write a blog post." It is a multi-step process: discover a topic, research it from multiple sources, generate an outline, write the article section by section, fact-check key claims against source material, and optimize for SEO. The quality comes from the process, not the model.

Automated image sourcing

Every article needs an image. The system sources relevant images automatically, processes them for web delivery, and attaches them to the article with proper alt text and attribution. No stock photo subscriptions. No manual image hunting.

How the pipeline works

The pipeline runs in stages, each one an independent n8n workflow that passes output to the next.

Stage 1: Topic discovery

The system monitors local news sources, event calendars, municipal websites, and social media for topics relevant to each of the four cities. It prioritizes based on relevance, timeliness, and whether the topic has already been covered. This runs continuously and produces a ranked list of potential stories.

Stage 2: Research

For each approved topic, the system gathers information from multiple sources. It pulls facts, quotes, context, and background. It cross-references to avoid single-source dependency. The output is a research brief that contains everything needed to write the article.

Stage 3: Writing

The AI writes the article based on the research brief, following editorial guidelines specific to SYDDK. Tone, structure, length, attribution style, everything is defined in the prompt architecture. The system generates the article in sections, reviews it for coherence, and produces a final draft.

Stage 4: SEO optimization

The draft gets optimized automatically. Meta title and description, heading structure, keyword placement, internal linking to related articles, schema markup. This is not "sprinkle in some keywords." It is structured SEO work that follows the same process a human SEO specialist would use.

Stage 5: Image and media

The system finds or generates a relevant featured image, compresses it for web, adds alt text, and attaches it to the article. For certain article types, it also pulls embedded maps or relevant social media posts.

Stage 6: Publishing

The finished article is published to WordPress via API. It goes live with the correct category, tags, author attribution, and publication timestamp. The system also submits the URL to search engines for indexing.

Stage 7: Monitoring

After publication, the system tracks article performance, indexing status, and reader engagement. Articles that underperform get flagged for human review.

The numbers

The 90% autonomy number is important to understand correctly. It means 90% of articles go from topic discovery to publication without any human touching them. The remaining 10% get flagged for human review, usually because the topic is sensitive, the sources are ambiguous, or the article requires local knowledge that the system does not have.

This is intentional. Full automation without a human safety net would be irresponsible for a news platform. The system is designed to handle the volume work automatically and surface the edge cases for human judgment.

The under-1-cent cost per article factors in AI API calls, server hosting, image processing, and n8n cloud infrastructure. The marginal cost of adding one more article is effectively zero. This is what makes the economics work where traditional journalism cannot.

What we learned building it

Process matters more than model. The quality of the output has almost nothing to do with which AI model you use and almost everything to do with the process wrapped around it. A mediocre model with excellent research, clear guidelines, and multi-step review produces better content than a top-tier model with a single "write me an article" prompt.

Modular beats monolithic. Building the pipeline as seven separate workflows instead of one giant process was the best architectural decision we made. When something breaks, we know exactly which stage failed. When we want to improve quality, we update one workflow. When we want to add a new city, we duplicate the configuration.

Monitoring is not optional. Any automated content system without monitoring is a liability. We built alerting for quality drops, publishing failures, SEO regressions, and source availability issues. The system runs itself, but we always know how it is running.

Start with the workflow, not the tech. Before we wrote a single line of automation, we documented exactly how a human editor would produce each article. Then we automated each step. The technology serves the editorial process, not the other way around.

What this means for content at scale

Most businesses do not need 180 articles a month. But every business needs content that is consistent, optimized, and produced at a cost that makes sense.

The same pipeline architecture works at any scale. A B2B company that needs four blog posts a month. A SaaS company that needs documentation updates. A local business that needs weekly news about its industry. The principles are identical: automate the repeatable work, keep humans in the loop for judgment calls, and build the process before you build the technology.

We wrote about the broader philosophy of automating everything that should be automated in a separate post. SYDDK is the most extreme example, but the approach applies everywhere.

If you are producing content manually and the volume or cost is not sustainable, we can build a pipeline that changes the math. Check out our automation services to see how we approach it.

And if you want to start with understanding where your current content stands, an SEO audit report will tell you what is working, what is not, and where the biggest opportunities are.