Content Marketing Pipeline
AI-powered content automation that takes keywords and publishes SEO-optimised articles at scale — across Shopify, WordPress as well as preparing marketing materials in Klaviyo and Omnisend.
The Problem
Consistent, SEO-rich blog content is one of the highest-ROI activities for ecommerce brands, but it’s time-consuming and expensive to do manually. Most store owners either neglect their blog entirely or pay agencies for content that’s generic and slow to produce.
The problem compounds when you factor in distribution. Even brands that produce decent content struggle to connect it to their email and SMS audiences in a systematic way — content, SEO, and retention marketing sit in separate tools with no shared workflow between them.
There was no lightweight tool that could take a list of keywords, turn them into publish-ready content automatically, and push it to where it needed to go — with real quality controls built in.
How It Works
Users upload a CSV of topics and keywords, or add them manually. The app queues each topic as a job and runs it through a multi-stage AI pipeline:
- Research — the AI investigates the topic for current trends and depth
- Draft — a full HTML post is generated based on the topic, keywords, and a user-configured system prompt covering tone, length, and persona
- Evaluate — the draft is scored for SEO quality
- Refine — the content is improved based on evaluation feedback
Once refined, the post is pushed directly to the target platform — Shopify via GraphQL Admin API (with SEO metafields), or WordPress via REST API — as a draft ready for review. For distribution, the pipeline connects to Klaviyo and Omnisend via API, enabling content to feed directly into email and SMS campaigns without manual copying.
A background job queue manages concurrency and tracks status across all jobs. The dashboard gives a live view of the queue — pending, in-progress, completed, and failed — with stats on total posts generated, published, and overall success rate.
Tech Stack
Frontend React 18, TypeScript, Vite, Tailwind CSS, shadcn/ui, Wouter, TanStack Query, React Hook Form, Zod
Backend Express.js (Node.js), PostgreSQL via Neon (serverless), Drizzle ORM, Passport.js
AI OpenAI GPT (configurable model) — research, drafting, evaluation, refinement
Integrations Shopify GraphQL Admin API, WordPress REST API, Klaviyo API, Omnisend API, PapaParse (CSV)
Background Processing Custom in-process job queue with DB polling
Outcomes
- Fully automated pipeline from keyword to published draft with no human intervention
- Multi-stage evaluation loop consistently improves output quality over single-pass generation
- Configurable system prompt lets non-technical users tune tone and persona without touching code
- Bulk CSV upload enables hundreds of posts to be queued in seconds
- Supports Shopify, WordPress, Klaviyo, and Omnisend from a single workflow
Lessons Learned
The evaluation and refinement loop matters more than prompt length. A shorter initial prompt followed by structured critique and revision produces better output than trying to front-load all instructions — and it’s easier to tune per client.
Job queue observability is underrated. The dashboard’s status tracking was added relatively late but became essential for users to trust the automation and debug failures confidently.
Shopify’s GraphQL API is powerful but brittle around metafields. Schema versioning and field naming conventions require careful handling — a recurring source of integration bugs that’s now well-documented internally.
GPT-4o-mini hits a ceiling on deeply technical or niche topics. For specialist industries, the research stage benefits from retrieval augmentation or a more capable model to avoid generic output. Model configurability was added partly for this reason.