+971 56 974 0358
Heymann Institute
Articles/AI
AI

Improve Your RAG System

Many organizations think the answer is to index more documents.

GH
Gustav Heymann
Managing Partner · Feb 6, 2026 · 3 min read

That is usually the wrong instinct.

A retrieval system does not improve simply because it has more content. It improves when the right content is clean, current, well-structured, and permissioned correctly.

The issue is signal.

The Noise Problem

Humans are good at ignoring document noise.

We skip the repeated footer. We overlook the draft watermark. We understand that a revision table is not the policy. We know that a page header is structure, not substance.

A retrieval system does not treat these elements the same way.

It sees text. It sees repeated tokens. It sees proximity. It may retrieve a chunk because the wrong words appear frequently, not because the chunk contains the best answer.

That is how noise becomes risk.

The system retrieves an outdated clause. It blends a draft with a final version. It reads a disclaimer instead of the rule. It connects broken text across page breaks. The user sees an answer and assumes the knowledge system worked.

The Container Is Not the Knowledge

Organizations often confuse document storage with knowledge readiness.

A shared drive, SharePoint site, or document management system may hold important material. That does not mean the material is ready for retrieval.

RAG systems need source design.

The content must be current, authoritative, structured, and accessible to the right users. Otherwise the model is being asked to compensate for poor knowledge governance.

It should not be expected to do that.

The Green Zone Strategy

Create a green zone for retrieval-ready content.

Only approved, current, high-signal documents belong there. Drafts, obsolete versions, redlines, duplicate copies, and administrative clutter should stay outside the indexed area.

This requires a tradeoff.

Dumping everything is easier. Designing a trusted source set is harder. But the quality of retrieval depends on the quality of the source.

The green zone should have entry criteria.

Is this the current version?

Is the owner named?

Is the document structured with clear headings?

Are definitions consistent?

Has administrative noise been removed?

Are permissions correct?

Is the review cycle clear?

Permissions Hygiene

Permissions are part of retrieval quality.

If permissions are too broad, sensitive content may surface to the wrong audience. If permissions are too narrow or inconsistent, users may not retrieve the right answer. If old folders carry inherited access no one understands, governance is weak.

RAG readiness therefore includes access review.

Content owners and information security teams need to know who can retrieve what and why.

Source Design Checklist

Before indexing, apply a practical checklist.

Remove repeated headers and footers where they add no meaning.

Remove revision histories unless they are needed.

Archive old versions outside the retrieval set.

Use consistent headings.

Define key terms.

Use meaningful filenames.

Avoid scanned images without OCR.

Keep one authoritative source.

Assign an owner.

Review permissions.

The Closing Test

Before adding content to a RAG system, ask whether a knowledgeable employee would trust that document as the current answer.

If not, the retrieval system should not either.

Better RAG starts before retrieval.

It starts with governed source material.

Related insights