AI Support & Content Agents on WordPress Docs

A small marketing product company had already invested in good documentation, tutorials and articles on their WordPress site, but their support team still spent a lot of time answering questions that were already covered there.

They wanted a lightweight way to turn this content into a helpful L2 support layer on their website, so common questions could be answered quickly while freeing the team to focus on real customer issues. At the same time, as a small team, they needed to produce new articles faster without adding more headcount.

The solution uses WordPress as the source of truth, n8n as the orchestration and agent runtime, and Supabase as both database and vector store for semantic search over the documentation.

Scheduled n8n workflows call the WordPress REST API to fetch posts and pages, normalize and chunk the content, generate embeddings and upsert them into Supabase. When a user interacts with an agent, another workflow embeds the query, runs a vector search and passes the top matching snippets into the model to generate either support answers or article drafts.

Key components:

• WordPress REST API for accessing existing documentation, tutorials and articles.
• n8n workflows for extraction, transformation, embeddings and agent orchestration.
• Supabase vector store for fast semantic retrieval across product and pricing content.

The main design choices focused on reusing one knowledge base for different agents and keeping the solution simple enough for a small team to own.

1. Use vector search on top of WordPress content: Instead of keyword search, I stored embeddings in Supabase so the agents can answer both documentation and pricing questions based on semantic similarity. This made the system more robust to how users phrase their questions.
2. One shared knowledge base, two agents: I used the same vectorized content to power a customer facing support chatbot and a separate internal agent that generates draft articles, which increased the value of the initial integration work.
3. n8n as the central orchestrator: n8n handles the full pipeline from data extraction to agent responses, which keeps the solution low‑code and easier to adjust without building a custom backend.
4. Automated refresh instead of manual updates: Recurring workflows keep the embeddings aligned with the live WordPress site, so both agents always work on current documentation without extra maintenance from the team.

The public support chatbot did not go live due to changing priorities, but the internal article draft agent became part of the team’s content workflow and reduced the time needed to produce first drafts.

For a small company, this meant more hours each week available for improving the product itself instead of starting every article from a blank page. The project also left them with a reusable AI knowledge layer on top of their documentation that they can plug into future agents when they are ready.

Additional benefits:

• Demonstrated a complete RAG style architecture on top of an existing WordPress site.
• Created a pattern the client can replicate for other products or content sets.
• Showed how small teams can get practical value from AI agents without a large engineering investment.