GEO Content Pipeline

Brief

Oxford Online Pharmacy wanted to grow visibility in AI-generated search answers across high-value medical categories. We built a multi-stage agentic pipeline that researches, briefs, writes, and quality-checks each article, with a human copywriter and client review built into the final stage.

To measure what was working, two approaches were run in parallel: AI-briefed, human-written content versus AI-briefed, AI-written, human-reviewed content, a structured A/B test to find the right balance of speed and quality for this client.

AI Implementation

• Research
  The agent researches the topic first as a dedicated step before anything else happens. Running research and briefing separately significantly improved accuracy and reduced errors in the output.

• Brief generation
  Using the research output, the agent builds a structured content brief aligned to target GEO prompts, brand voice, and medical accuracy requirements.

• Writing
  A separate writing agent produces the first draft from the brief. This is the output the copywriter receives, taking 1hr 15 to edit versus four hours to write from scratch.

• QA agents
  Multiple QA sub-agents run in sequence, writing quality (which also fixes, not just flags), factual accuracy against authoritative medical sources, regulatory compliance, internal linking across category and product pages, external links, and citation checks against target GEO prompts.

• Correction log
  All feedback from the copywriter and client feeds into a correction log that improves future writing runs and connects back into the content planning agent over time.

• Interface
  Built in n8n with Airtable as the interface for managing briefs, drafts, and QA outputs.

Video overview

Watch a walkthrough of the pipeline in action, from research and briefing through to QA, human review, and final output.

Human in the loop

A copywriter reviews and edits every AI-produced article before it goes to the client. The client then gives final approval. Both rounds of feedback feed directly into the correction log, continuously improving the pipeline's output quality over time.

Impact

• First draft production time reduced from four hours to 1hr 15 of copywriter editing

• QA agents fix content as well as flag it, removing a separate correction round.

• Internal linking, external links, and citation checks run automatically on every article.

• Structured A/B test built in from the start to measure real content performance, not just output volume.

• Correction log creates a compounding quality improvement loop with every piece published.