Skip to content
Amsby

AI agents & automation

From one keyword to a publication-ready Dutch SEO article, quality-checked.

Amsby built an n8n and Google Gemini content agent with RAG and an automated quality gate that turns a keyword into a finished Dutch SEO article in Google Docs.

Client
Online marketing agency
Sector
Online marketing agency, Netherlands
Type
AI content agent with RAG and quality control
Engagement
04 · Project-based

About the client

We built this agent for a client of one of our partners, so we don’t name them here.

The client is a Dutch online marketing agency that offers SEO, SEA, social and CRO, among other services.

The problem

High-quality Dutch SEO content is slow and expensive to produce. Every article needs keyword and competitor research, a solid structure, strict SEO rules (meta title and description length, one H1, keyword density, intro length) and the client’s house style. When writers produce content by hand at agency volume, quality and consistency vary, and every article still needs an editor’s review.

What the agent does

The user enters a main keyword, plus optional secondary keywords, and the agent runs a full editorial pipeline in Dutch:

  1. Live SERP and news research. It retrieves the current top Google results and the latest news for the keyword at the same time, through the Serper API.
  2. Research analysis. Gemini analyses the competition: dominant themes, gaps the top results miss, relevant facts and figures, recent developments and regulations, and the dominant search intent. It then recommends a unique angle, subtopics and tone.
  3. SEO plan builder. It creates a structured JSON article plan with the meta title (60 characters or fewer), the meta description (155 characters or fewer, opening with a question and ending with a CTA), focus keywords and synonyms, title, subheading, a single H1, an H2/H3 structure with the purpose of each section, a commercial section, internal-link placeholders and the CTA.
  4. Article writer with house-style RAG. It writes a 1,500–1,800-word article from the plan. Through a retrieval (RAG) tool, the writer consults the client’s own reference articles, uploaded through a separate form and stored as Gemini embeddings in a vector store. This way the tone, structure and vocabulary match the client’s house style.
  5. Automated quality gate. A separate Gemini “editor” scores the article from 1 to 10 on SEO, format, readability, content and conversion. It also checks for critical errors, such as multiple H1s, a missing keyword in the H1 or intro, keyword density outside 1–2%, over-long paragraphs, external links, HTML or a missing CTA.
  6. Automatic revision. If the overall score is below 7, the article is rewritten using the editor’s structured feedback before it moves on.
  7. Delivery and logging. The final article is created as a Google Doc. An AI originality check runs on it, and the keyword, quality score, originality score and document link are logged in a Google Sheet.

Errors at any step are caught and logged, so a failed run never fails silently.

At a glance

steps from keyword to finished article
7
words per article
1,500–1,800
quality score, with revision below 7
1–10

Business value

  • Research, planning, writing and editing run automatically, from keyword to finished article.
  • The house style stays consistent at scale, because the agent learns it from the client's own reference content.
  • Strict, measurable SEO rules are built into every article instead of being left to an editor's checklist.
  • The process corrects itself, so weak drafts are caught and revised before a human sees them.
  • Every article is fully traceable, with a quality score, an originality score and a document link.
  • Editors move from writing to reviewing and approving.

Technologies

  • n8n
  • Google Gemini 2.5 Flash
  • Gemini Embeddings
  • RAG / vector store
  • Serper (Google Search & News API)
  • Google Docs API
  • Google Sheets API
  • JavaScript
  • Multi-agent prompt engineering

Hand repetitive work to an AI agent?

Built on your own data and tools, with a human approving anything important.