In-House Case — The Autonomous SEO Content Engine: this website writes and optimizes itself
The website you're reading is the case study: an autonomous content system of ~25 scheduled jobs that picks topics, writes, quality-scores, publishes, reviews performance, and rewrites weak posts — with zero routine human intervention.
The blog and SEO of nextagent.ca run on a fully automated pipeline: daily AI-industry news ingestion, writing across 6 content pillars, a quality-scoring gate deciding what may publish, post-publish self-optimization of titles/FAQ/internal links, rewriting of underperformers, search-performance sync, and a daily ops report. We run on ourselves what we sell to clients.
The expensive part of content marketing isn't writing one article — it's writing every day, reviewing what you wrote, rewriting the weak, and doubling down on the strong. No team sustains that loop on human effort forever. We used our own website to answer one question: can the whole loop be handed to a system? The answer is this site. Its blog is operated by a pipeline of roughly 25 scheduled jobs — topic selection, writing, scoring, publishing, optimization, review — each an independent, auditable automation step.
The challenge
A services company's website needs continuous SEO content to generate leads, but the full loop — daily output, ongoing review, iterating weak posts — is prohibitively expensive to sustain manually. And once it stops, accumulated rankings start to decay. Most company blogs die at 'wrote a few posts, then abandoned.'
Our approach
Decompose content operations into a fully automated pipeline, one scheduled job per stage: ingest AI-industry news sources → select topics and write across 6 content pillars → quality-scoring gate (low scores don't publish; they're sent back or pruned) → scheduled publishing → post-publish optimization of meta, FAQ, and internal links → sync search-performance data → mine new FAQs from real search queries → detect and rewrite underperforming posts → prune chronically weak drafts → report the whole chain daily to an ops channel. Humans read the daily report; they don't touch the routine.
Results
The blog now produces and self-optimizes with zero routine intervention: content never stops shipping, weak posts get rewritten (by the system), and search-performance data feeds back into topic selection. The engine doubles as a live showroom for what we deliver to clients — the content pipeline in our CPA firm case study is its client-adapted version.
The quality gate: being written ≠ being published
The most important stage in the pipeline isn't writing — it's interception:
- Every draft passes a quality score before it can publish; below threshold, it doesn't.
- Low-scoring drafts go to rewrite or pruning — they don't occupy publishing slots.
- The system controls publishing cadence, preventing a content flood from diluting site quality.
The reputational risk of an automated content system is publishing what shouldn't be published. The gate matters more than the throughput.
Publishing is the beginning: the self-optimization loop
A published article isn't the end of the process — it's the start of the data:
- Search-engine performance (coverage, clicks, ranking queries) syncs automatically.
- Real search queries are mined for user questions, which become new FAQ entries.
- Titles, meta descriptions, and internal-link structure are optimized automatically.
- Persistent underperformers are detected and rewritten; chronic non-performers are pruned.
Six content pillars and a daily report
Writing is organized around six fixed content pillars, keeping the content matrix aligned with business positioning instead of chasing trends. Every day the pipeline reports to an ops channel: how much was produced, published, intercepted, and optimized — humans supervise the system instead of working for it.
Technical breakdown
- A Python pipeline of ~25 scheduled jobs: ingestion, writing, scoring, publishing, optimization, and review are independent and individually re-runnable.
- Multi-model routing by task — writing, scoring, and rewriting each use their own configuration.
- Search-performance data syncs via official APIs into the database, driving FAQ mining and weak-post detection.
- Shares the data layer with the website itself (Next.js + PostgreSQL), so publishing takes effect immediately; a daily run report pushes to the ops channel.
Reusable lessons
- The core asset of an automated content system is the gate plus the review loop — generation stopped being scarce a while ago.
- Run it on yourself before selling it: your own site is the most honest test environment; problems can't hide.
- Even a fully autonomous system reports daily — 'zero intervention' is only acceptable on top of 'full visibility.'
Want results like these?
Tell us about your business and we'll scope a proof of concept.