Generate AI content based on real data, no hallucinations

What a knowledge base is

A knowledge base is a container of verified information that the AI consults before writing. Instead of letting the model fill in gaps with what it "thinks" it knows, you give it the right data: prices, dates, features, descriptions, frequently asked questions, your brand's own vocabulary.

When you generate an article with a base assigned, 4Linking automatically selects the most relevant information from that base and passes it to the model as context. The result is content that relies on real facts you've provided, not on what the model improvises.

AI makes up figures, dates and data

Generative AI models tend to fill in what they don't know with plausible but false data: a percentage that sounds right, an approximate date, the name of a study that doesn't exist, an invented figure. In blog content, where credibility is everything, a single false data point can damage your site's reputation.

This happens especially with:

• Data about your own business: prices, plans, features, dates. The model doesn't know them, so it makes them up.
• Figures and statistics: percentages and specific numbers that sound credible but that no one has verified.
• Names and references: studies, companies or authorities cited as backing that may not exist.
• Brand vocabulary: the model uses generic terms instead of the ones you use with your customers.

The knowledge base attacks this problem at the root: if the information is in the base, the AI uses it; if it's not, the system instructs it to rephrase rather than make things up.

The AI consults your data before writing

The mechanism is simple to understand: you feed the base with verified information, and when generating an article, 4Linking searches that base for what's relevant to the topic and hands it to the model as starting material.

This approach is what's known in the AI world as "grounding": anchoring the model's answers to specific, controlled information instead of letting it improvise. The result is more reliable content, more faithful to your brand.

Bases and entries: information, organized

The organization is hierarchical and simple. A base is a named container that can represent a client, a niche, a product or a brand. You create as many bases as you need.

Inside each base there are entries: units of information with a short title, full content (what the AI will read) and optional tags to classify them. A base can have anywhere from a few entries to hundreds.

Each entry has a precomputed semantic vector (the "Vector" column with the "Ready" status). That's what lets the system instantly find the most relevant entries for the article topic at generation time.

Three ways to create entries

Filling a base is flexible: you can do it by hand when you have specific data, let the AI generate a first batch to get started quickly, or import many entries at once from a CSV.

1. Manual entry

You create the entry by filling in title, content and tags. Ideal when you have a specific data point you want to lock in: a price, a feature, an answer to a frequently asked question.

2. Generate with AI

From a description of the topic or product, the AI creates a batch of draft entries that you review before adding. Perfect for seeding a base quickly. Important: the AI only uses the information you provide in the description —it doesn't make up data.

3. Import from CSV

If you already have the information in a spreadsheet, you can import it in bulk. The system provides a template with the correct columns (title, content and tags) and processes all the entries at once.

Feed the base from a file or a URL

You don't have to write every entry from scratch. When creating content for the base, you have two extractors that fill in the material for you:

• File upload: drag a document (Word, PDF, .txt or .md, up to 20 MB) and the plugin extracts its text automatically. Word is the recommended format for best extraction quality.
• URL extractor: paste the link of a public web page and the plugin extracts its content as plain text, ready to be turned into entries.

That way you can turn a company brief, a product page or an internal document into knowledge entries in seconds, without copying and pasting by hand.

Base configuration: brand, vocabulary and blacklist

Beyond the entries themselves, each base holds three configuration blocks that apply to every generation that uses it. They make sure the AI doesn't just say the right facts, but speaks like your brand:

• Brand information: exact name, website and industry. The AI uses it to refer to your brand correctly.
• Own vocabulary: specific terms the AI must respect. For example, "we call the templates briefings, not templates."
• Blacklist: words or topics the AI must avoid. For example, don't mention competitors by name, or don't use "the best on the market."

How the AI picks which entries to consult

A base can have hundreds of entries, but not all of them are relevant to every article. The system doesn't pass them all to the model: it selects only the ones that actually relate to the topic.

To do this, each entry has a precomputed semantic vector that captures its meaning. When generating an article, 4Linking calculates the vector for the topic and picks the entries closest by meaning —not by word match. That way the model receives just the right relevant context, no noise.

Coverage indicator

Before generating, the plugin can check whether the base properly covers a specific topic. It calculates the similarity between the article title and the base's entries: if no entry is sufficiently related, it flags that title as "low coverage" and warns you.

It's a safety net: it tells you in advance when the AI will have little verified material for what you're asking, so you can add entries or review the result more carefully before publishing.

Knowledge bases in 4Linking

The knowledge bases module is part of 4Linking's paid versions, along with content generation, images, translation, semantic linking and automation. The free version covers complete Internal linking; knowledge bases are a capability of the paid versions.

All paid versions include exactly the same features; they differ only in the number of sites where you can use the license.

Frequently asked questions about knowledge bases