Background to this guidance

A government AI assistant pilot in 2025 highlighted several aspects of content design that influence how well large language models (LLMs) process information — these insights form the basis of the guidance provided here.

LLMs — Glossary of AI terms

Use clear headings and subheadings

Clear headings and subheadings help organise content into easy-to-read sections, making it easier for LLMs to understand the structure of the content.

This allows them to quickly determine context and prioritise the most relevant information.

Headings and subheadings

Write simple sentences and keep paragraphs short

Short, simple sentences are easier for LLMs to process and less likely to be misinterpreted. They reduce ambiguity and help LLMs understand your meaning more accurately.

Short paragraphs also improve comprehension by breaking information into manageable chunks, making it easier for LLMs to extract key ideas when summarising or answering questions.

Plain language

Break down content with bullets and lists

Bulleted or numbered lists break down information into easily scannable chunks, making it quicker for LLMs to spot and understand key points.

LLMs process content in segments, and lists give a clear signal about the main points. If content is organised in lists, LLMs are better at identifying the most important details and summarising them efficiently.

Lists

Avoid jargon and clarify technical terms

When content avoids overly technical or complex language, LLMs are more likely to generate accurate responses. This is because the vocabulary is better aligned with what the LLM has been trained on.

Offering definitions for complex or technical terms helps LLMs understand specific concepts. The LLM can also use these definitions to better comprehend the context and generate more accurate responses related to those terms.

Words to avoid and specialist words

Incorporate keywords naturally

Integrating relevant keywords naturally into content helps LLMs understand the main topics.

Keywords can improve content relevance and visibility, with the Retrieval Augmented Generation (RAG) database helping LLMs prioritise results based on both matching themes and search terms entered by users.

• Writing for search engines
• Google Analytics for content design
• RAG — Artificial Intelligence glossary — Ministry of Business, Innovation and Employment

Use active voice over passive voice

Using the active voice is clearer, more direct, and is processed more efficiently than passive voice by LLMs.

Passive sentences can sometimes be harder for LLMs to interpret accurately, especially if the subject or actor is unclear.

Tone and voice

Focus on intent and purpose

LLMs perform better when content has a clear focus and purpose. It helps generate responses that are more aligned with the user’s query or the content’s intent.

To improve content clarity for LLMs, you can:

• add a summary sentence
• put the most important information at the top of the page
• limit each page to 1 topic.

Page structure