AI Readability

We believe AI is rapidly becoming omnipresent, offering a level of assistance unimaginable just a few years ago. At Mynt, we aim to be the first early childhood company to place AI at the core of our many endeavours. To make this possible, we must first organise our information so these models can process it with ease.

How AI Consumes Text

• Chunking: AI models often break text into chunks or “tokens.” Clear headings, subheadings, and bullet points help the model parse information more accurately.

• Context Windows: Large language models (LLMs) typically have a limit on how many tokens they can process at once. Shorter, well-defined sections keep topics within a manageable scope.

Structuring Information

• Top-Level Categories: Group similar policies under broad headings (e.g., “HR Policies,” “Finance & Procurement,” “Data Security,” etc.). This allows both humans and AI to locate relevant sections quickly.

• Nestled Pages: Within each category, divide topics into separate pages. For instance, under “HR Policies,” you might include pages for “Leave Policies,” “Attendance,” and “Conflict Resolution.”

• Consistent Page Layout: Begin each page with a concise summary or “purpose” statement. Follow with details, subheadings, bullet points, and any relevant procedures.

Write for AI (and Humans)

• Short, Clear Paragraphs: Limit each paragraph to a single main idea. This helps the AI pinpoint key concepts without wading through overly long text.

• Use Bullet Points and Numbered Lists: Whenever enumerating steps in a process, opt for bullet points or numbered lists. This improves clarity for both AI and human readers.

• Meaningful Headings: Make headings descriptive; for example, use “Leave Application Procedure” instead of “Procedure.” These headings act like signposts for the AI model.

• Define Terms and Acronyms: Provide clear definitions. For instance, “PDO (Paid Day Off) – a full day of paid leave available to full-time employees.” If possible, maintain a glossary page for commonly used acronyms so the AI can build context accurately.

• Be Consistent with Language: Use the same term consistently (e.g., “Annual Leave” vs. “Vacation Leave”). AI models handle uniform terminology better, reducing confusion.

Provide Context in the Right Places

• Add Summaries: At the start (or end) of each policy page, offer a brief summary. Summaries can serve as quick references for staff and give the AI a distilled version of the key points.

• Include Examples: If you have policies that require interpretation (e.g., “Work-From-Home Policy”), show practical examples or scenarios. Contextual examples help AI understand nuance and how rules are applied.

• Reference Dates/Versions: If policies change over time, note “Last Updated” dates or version numbers. This helps the AI keep track of current vs. outdated policies if you plan to feed multiple versions into it later.

Optimise Metadata and Linking

• Use Tags or Labels: Some GitBook setups allow tagging or metadata fields. Meaningful tags help AI and search algorithms locate content quickly.

• Cross-Link Relevant Pages: Insert links to related topics. For example, the “Data Handling Policy” page might link to “Data Security” and “Privacy Policy.” Cross-references help AI build a knowledge graph, providing deeper context.

Usage Guidelines

• User Guidelines for Staff: Encourage staff to keep questions specific when asking the AI about policies: e.g., “What is the maternity leave policy?” Provide examples of good prompts so the model can surface relevant policy text effectively.

• Feedback Loop: If staff find unclear or outdated content, prompt them to highlight it so you can revise GitBook entries. Continual improvements keep the AI’s source material accurate.