Built this week in GeoPolitiq: The Content Distribution Engine

Most people think building an AI-powered content site is about writing a good prompt. They're wrong. The prompt is maybe 10% of the battle. The other 90% is infrastructure: how you fetch data, how you validate it, how you format it for different platforms, and most importantly, how you ensure the world actually sees it before it becomes obsolete.

This week at GeoPolitiq, I pushed 10 commits that shifted the project from a "cool generator" to a production-ready distribution engine. We're talking automated social fan-out, instant search engine indexing, and a hardened bot detection layer.

Here’s a deep dive into the engineering decisions made over the last seven days.

1. Refactoring the Generation Pipeline: Per-Region Sonar

Geopolitics is inherently fragmented. A single global prompt produces generic, beige content. To get actual insight, you need to go deep on specific locales. This week, I refactored the AI generation logic to use a per-region Sonar approach.

Instead of a monolithic call to a model, the system now cascades through regional configurations. We use environment-driven configs to swap out models or prompt depths based on the complexity of the region.

The Image Cascade

Another critical update was the image cascade. In automated publishing, your primary image source might fail or return a low-quality placeholder. I implemented a fallback logic that attempts to generate/fetch high-res imagery, and if that fails, it cascades down to secondary assets or thematic defaults. This ensures that the social fan-out (which relies heavily on OpenGraph tags) never posts a broken link or a blank card.

// Simplified logic for the image cascade
async function getThumbnail(region, topic) {
  try {
    const primary = await fetchPrimaryImage(topic);
    if (isValid(primary)) return primary;
    
    const fallback = await generateAIImage(region, topic);
    return fallback || DEFAULT_GEOPOLITIQ_HERO;
  } catch (err) {
    console.error("Image cascade failed, using default", err);
    return DEFAULT_GEOPOLITIQ_HERO;
  }
}

2. The Indexing Problem: Google & IndexNow

In the world of news and geopolitical shifts, if your page is indexed three days after the event, you’ve already lost. SEO isn't just about keywords; it's about latency.

I integrated the Google Indexing API and IndexNow simultaneously. While standard sitemaps are "pull" based (Googlebot visits you when it feels like it), the Indexing API is "push." The moment a post is generated and committed to the database, GeoPolitiq fires a notification to Google and Bing.

We also fixed a critical bug where the sitemap-news.xml was returning empty. In the world of Google News, that’s a death sentence. By hardening the sitemap generation and adding /feed.xml and /llms.txt aliases, we’ve made the site readable for both humans and other AI agents (like Perplexity or GPT-4o) that might want to cite our data.

3. Social Fan-Out and the Repost Queue

Content shouldn't live in a silo. This week saw the introduction of the Social Fan-out Queue.

I built a manual-trigger hook fix that allows us to re-run distribution if a specific API (like Twitter or LinkedIn) flakes out. But the real heavy lifting happens in the background. When a post goes live, it enters a queue that handles:

1. Tumblr OAuth Setup: Added maintenance scripts to handle the complex handshake required for automated Tumblr posting.
2. Nostr Integration: Geopolitics often touches on censorship-resistant themes. I added support for Nostr (NIP-09) for deletions and profile management. If a generation is "burned" (deleted due to inaccuracy or updated info), the system broadcasts a NIP-09 event to the relays to signal that the content should be removed.
3. UTM Tagging: Every link shared via RSS or social now carries specific UTM parameters. This allows us to see exactly which platforms are driving high-intent traffic versus bot noise.

4. Hardening the Perimeter: Bot Detection & Analytics

When you build an automated site, you attract automated scrapers. It's an ironic cycle. To prevent our server resources from being eaten by low-quality crawlers, I hardened the bot detection logic.

We now track referrers more aggressively and have implemented a rejection analytics layer. If a request looks like a headless browser trying to scrape our AI-generated insights without attribution, they get served a 403 or a cached, lightweight version of the site. This keeps the database connections available for actual readers.

5. Maintenance and Admin Upgrades

As the repository grows (we're at 10 commits this week), the need for "janitor scripts" increases. I added several maintenance scripts to handle:

• Backfills: If we add a new social platform (like we did with Tumblr this week), we need to be able to backfill the last 50 posts to that platform without hitting rate limits.
• Google Indexing Backfills: Ensuring older content that might have missed the initial push gets processed.
• Generation Burn: A script to cleanly delete a post across the DB, the sitemap, and the social graph.

I also upgraded the admin panel to include Adsterra ads integration. While monetization isn't the primary goal during the alpha phase, testing the layout shift and performance impact of ad scripts early is better than trying to shoehorn them in later.

What's Next?

The foundation is now solid. We have a system that can generate deep, regional geopolitical content, optimize it for search engines in real-time, and distribute it across the decentralized and centralized web.

The next step is refining the "Scored Related-Posts" algorithm. Right now, it's functional, but I want it to be smarter—linking South China Sea tensions to global trade route disruptions automatically based on semantic similarity, not just tags.

GeoPolitiq is live, the URL is in the README, and the engine is humming. If you're interested in how AI, SEO, and high-velocity distribution intersect, keep an eye on this repo.

Nandishwar Singh Lead Engineer, GeoPolitiq