AI Content Automation Workflow

Using AI for content used to mean: find an article, ask AI to rewrite, organize manually, format manually, publish manually. It helped individual tasks, but much repetitive work stayed manual.

To see whether the whole process could become a repeatable workflow, I designed and ran this AI content production workflow from retrieval to WeChat draft creation.

Using AI to rewrite text only speeds up isolated steps; much coordination and formatting stayed manual.

The real difficulty was not “finding a better model” but chaining retrieval, restructuring, cover generation, formatting, and publish-prep into a repeatable, clearly bounded workflow.

The first workflow starts from an article URL: extract title and key points, analyze and restructure, generate a new title, AI-generated cover, upload assets, convert to WeChat format, create a draft via API, and finally human review and human publishing; cover generation is automated with prompt-controlled style (see the full flow below).

The second adds an automated front end: scheduled news collection, filtering, and storage in Feishu, then keyword search retrieves a URL and runs the production workflow — evolving from manual discovery into a complete loop of automated discovery, filtering, production, drafting, and human review.

The project mainly used Coze Workflow and actually integrated Feishu, APIs, OpenClaw, and the Qwen cloud LLM; the OpenClaw + Qwen cloud LLM approach was also used in the formal workflow and completed production tasks.

I was responsible for workflow design, business decomposition, AI tool selection, prompt design, API integration, content logic, automation testing, retrieval experiments, and end-to-end optimization.

The key was not any single AI tool but combining multiple AI capabilities, data sources, and tools into a workflow that could actually run.

This was not a concept demo: over 50 materials were filtered, around 10 complete workflows were run, and all 10 runs successfully reached the WeChat draft box — a 100% success rate — with final publishing always done manually.

The project validated an end-to-end AI content workflow and produced real content, reinforcing that AI automation’s value is connecting fragmented tasks into one workflow. The core model became AI Automation + API Integration + Human-in-the-loop.