Qualitative Agentic AI empowered with context: start small, deliver big

⚡ TL;DR

You don’t need a vector database to give your AI agent context. In this guide, I show you how to build a minimal RAG setup using n8n and a Google Sheet to automate internal link suggestions in blog content. It’s fast to implement, SEO-friendly, and perfect for proving value before scaling.

New to AI content strategy? Content drives revenue! See how AI helps you scale it profitably lays the foundation for why tools like minimal RAG matter.

This guide is for you if you’ve built or are building a content-generating AI agent that creates articles for your business. It writes, formats, maybe even schedules.

But here’s one thing it probably doesn’t do yet: suggest relevant internal links.

Without internal linking, your SEO performance suffers, and users lose valuable context.

The good news? Even if your internal links aren’t stored in a vector database, and you have zero experience with semantic embeddings (yet), you can still fix it.

In this post, I’ll show you how to build a minimal RAG (Retrieval-Augmented Generation) system using n8n that powers smart internal link suggestions using just a spreadsheet.

This minimal RAG guide is ideal for:

MVPs, to help you demonstrate the value of RAG to stakeholders
Lean teams who want to introduce SEO-conscious automation into their content process without overengineering anything

Hi, I’m Oni,

I’m passionate about building agentic AI systems that work with people, not around them. Systems that enhance human creativity, reduce busywork, and actually make teams better at what they do. Not future hype — just practical, incremental progress that delivers real value now.

I believe in starting simple, building smart, and scaling collaboratively, because sustainable change doesn’t come from massive launches, it comes from useful tools people want to keep using.

This blog is part of my mission to demystify AI, make it accessible, and help organisations adopt it in ways that feel intuitive, empowering, and genuinely helpful — not just technically impressive.

🧠 What is Minimal RAG, and why is it a good way to start giving context to your agents?

Let’s start a little bit technical without being to technical. Retrieval-Augmented Generation enhances AI output by injecting information relevant to your business and relevant to the agent’s task before generation — so it doesn’t rely solely on the model’s training data.

Let’s look at a few examples of where AI falls short out of the box:

AI models know almost nothing about your actual products.
Ask them to promote your product range, and you’ll get vague, generic content — not tailored messaging that reflects your unique offering.
They only have surface-level knowledge of your industry.
That deep field expertise you’ve developed over years? The AI doesn’t have it, unless you teach it. So the insights it offers will be shallow, missing the nuance your audience actually cares about.
They have no awareness of your existing content library.
Coming back to the example of problem we are trying to solve here: unless you explicitly give the AI access to your articles, blog posts, or resources, it simply can’t suggest relevant internal links, which means missed SEO opportunities and disconnected user journeys.

Put simply, RAG (Retrieval-Augmented Generation) means giving your AI model a helping hand — by supplying it with actual knowledge about your organisation.

Full Rag

A “full” RAG setup typically involves:

Storing your content in a vector database like Pinecone or Weaviate
Embedding that content so the AI can understand it numerically
Using semantic search to retrieve the most relevant information during generation

It’s powerful stuff — but it comes with a cost: implementation time, infrastructure setup, and storage. That’s why it makes sense to start small, test the concept, and prove the value first before going all in.

Minimal RAG

The Minimal RAG I am going to teach you in this article is a lightweight version:

Uses structured metadata like titles, tags, summaries
Matches on keywords or categories
Requires no embeddings or vector DBs
Can be implemented with a spreadsheet and n8n

This makes it perfect if you:

Are early in your AI adoption journey
Have a manageable amount of content
Want to demonstrate value before scaling AI infrastructure

The problem we are trying to solve: Optimise SEO by adding relevant internal linking

If SEO matters to you, internal linking isn’t optional: it’s essential. And if you’ve already made it a priority, you probably know the struggle: finding the right pages to link to, figuring out how to naturally weave them into the content, and doing it consistently across every post.

Internal links:

Help Google crawl and index your site more effectively
Distribute authority across your pages (known as link equity)
Reinforce topical authority for your domain
Improve user experience by guiding visitors to related content

Most SEO best practices recommend including around 6 internal links for articles averaging 1200 words. But picking the right ones and phrasing them well can be surprisingly time-consuming and very often it becomes inconsistent. Especially when you’re working at scale or under tight deadlines.

By introducing a Minimal RAG system, you can « automate » this process. Your generative agent can reference existing content intelligently — boosting SEO and content quality in every post.

🛠️ What we’ll build

Today I want to show you how to solve this problem effectively using n8n and a minimal RAG.

In this example — one of the MVPs we set up for a client — we designed a multi-agent system, where one of the agents was responsible for generating a blog brief.

As part of the AI generated brief, the client wanted to receive recommendations for internal links to include in the eventual blog post. In fact, they didn’t want the agent to just automatically insert links during content generation later in the process. Instead, they wanted to review the suggested links first, understand why each one was chosen, and decide which to include.

This approach is known as a user feedback loop, and it’s something we actively encourage with all our clients. It keeps humans in control, builds trust in the AI, and allows teams to streamline their workflow without giving up decision-making power, which is key for successful AI adoption.

Your internal linking agent will:

Receive a blog brief or title
Reference a spreadsheet containing existing blog posts
Use keywords or tags to find related posts
Suggest 2–5 internal links to include in the new article

Reminder: This guide is for you if you’ve built or are building a content-generating AI agent that creates articles for your business on n8n.

Tools you’ll need

n8n (self-hosted or cloud)
Google Sheet
Optional: OpenAI or Claude (if connecting to a content agent)
Basic n8n familiarity (eg: ability to add credential to access a Google sheet in n8n)

Step-by-Step: Building the Minimal RAG Workflow in n8n

1. Starting with the data

Whether you’re setting up a minimal RAG system or going full scale, your data is the foundation. And don’t worry — if you don’t have this in place yet, I’ll walk you through exactly what you need.

The first step is to create a structured source of truth for your internal links. In this example, we’re using a Google Sheet — lightweight, accessible, and perfect for MVP setups.

To help your agent understand and use the internal links effectively, your spreadsheet should include the following columns:

Column Name	Description
Title	The title of the page or blog post
URL	The unique identifier — the full link to the page
Page Description	A clear summary of what the page is about; critical for link relevance
Tags	2–5 keywords or themes describing the content
SEO Description (optional)	The meta description used for SEO — adds context for the model

🧠 Remember: your AI is only as useful as the context you give it. If the description or tagging is vague, the agent will struggle to make helpful suggestions.

2. Use the spreadsheet as a tool in your n8n agent

Now that you have your data source, we can integrate it to yout n8n agent. You’ll be surprised how simple — and powerful — this setup really is.

In the example I will use here, we’re using on of our client’s blog brief generator agent.

Here’s how it works: the agent pulls a topic from a Google Sheet the marketing team regularly updates with article ideas. Each entry includes:

A working blog title
Context about what they want to cover
The intended purpose of the post (e.g. promotional, educational)

As an output, the agent generates a detailed blog brief that includes:

Suggested target audience
Key headlines
SEO elements like target keywords and a draft meta description

This one agent alone has already saved their marketing team a huge amount of time.

Minimal RAG Configuration

To increase its value, we simply added the internal link spreadsheet as a tool to the OpenAI node.

Step 1: In your OpenAI node, go to the Tools section.
Step 2: Search for and select Google Sheet as your tool.
Step 3: Configure the Google Sheet by linking it to the Internal Link Inventory spreadsheet you prepared earlier.
Step 4: Rename the node clearly to reflect its purpose. This will help you stay organised as your workflow grows.

Once you’ve connected the spreadsheet, it’s important to tell the model how and when to use it in your prompt. This step is critical. You can see in the picture below an example of how we configure our prompt.

💡 Best practice:
We recommend creating a TOOLS section in your prompt, where you give specific instructions on how to use each tool (especially useful as your agent starts accessing multiple tools).

Steps to add a Google Sheet as a Tool to enable minimal RAG on a OpenAI node.

Example of prompt to integrate the tool in the OpenAI model, enabling AI to access the tool when relevant - key step to enable minimal RAG.

And that’s it! No heavy lifting. Just a spreadsheet, a clear prompt, and suddenly your AI is a lot more useful.

📈 The problem we are trying to solve: Optimise SEO by adding relevant internal linking