CEO, Owner, and Co‑Founder
Todd is responsible for driving Four Kitchens’ vision and long-term strategy.
February 10, 2025
Building an effective AI chatbot is a technical challenge, but one that presents real possibilities in terms of delivering information to website users. Instead of paging through search results or bouncing around your website, users can ask for what they need and get a response in an instant.
But to deliver correct, meaningful answers to users, you need to take a close look at your content. AI has made great strides, but a chatbot’s effectiveness hinges on the quality of your source material. Not only should your content be accurate and up-to-date — it should be properly organized, too.
Interpreting dense paragraphs of information is hard for people and machines alike. Forcing a chatbot to produce answers from poorly structured content leads to responses that are wrong — or even inappropriate. Along with frustrating users, these unpredictable answers erode trust in your organization.
But! If you’re paying attention to how your content is organized, a chatbot will deliver more accurate answers and more consistently positive experiences. When properly designed, a chatbot provides a cost-effective way to quickly deliver information while also improving your content’s readability, usability, and accessibility — for everyone.
Why AI chatbots are only as good as their content
Many universities and nonprofits have been exploring AI chatbots to streamline how website users access information. Implementing an architecture called retrieval-augmented generation (RAG) enables your organization to apply a set of guardrails to a chatbot as it pulls relevant answers to queries from a defined knowledge base.
However, even with RAG, chatbots can still “hallucinate” — that is, return plausible but incorrect answers. The root of this problem often lies in its source material — the content the AI uses to generate answers — rather than the technology that powers it.
Refining chatbot responses in creating Yale’s AI chatbot
Yale University asked Four Kitchens to help them implement AI tools across their web platforms (like YaleSites). We started by working with the school’s dining information, which was a relatively low-stakes dataset where inaccuracies would be easy to identify.
In early iterations of Yale’s AI agent, we asked it for vegan dining options. It suggested chicken nuggets, which is obviously wrong. But our issue wasn’t due to a problem with the AI tool. It came down to how the source data was organized.
The chatbot was pulling information from a Drupal website that listed food options in dense paragraphs. In much the same way people have a hard time interpreting large blocks of text, the AI tool was unable to correctly categorize and retrieve the right information.
To resolve the issue, we reworked the content to make the food groupings more obvious: adding headings, separating menu items into lists, and so on. No more chicken nugget suggestions!
The lesson? When you improve the structure of your content, you improve the experience for machines and people alike.
How content quality shapes AI performance and reduces costs
Our experience with the dining chatbot underscored a familiar principle in chatbot development: Garbage in, garbage out. When the source material is poorly structured, even the most cutting-edge AI tool will struggle to provide useful responses.
The challenge lies in how AI interprets content. When indexing large chunks of content, AI applications use a process called “chunking,” also known as “text-splitting.” This involves breaking content down into smaller, more manageable pieces.
These chunks need to overlap a little bit with what comes before and after so the AI understands each chunk in context. Without overlap, individual chunks of text may not make sense to the AI. However, the degree of overlap is a balancing act. More overlap improves accuracy, but it also increases the cost — both computational and dollar cost — of each AI prompt or query. Your organization must strike a balance between cost-effectiveness and chatbot performance.
This is where content structure becomes a crucial advantage. Well-organized, clearly structured content is easier for both machines and humans to digest and understand — and it can save you a bundle in query fees!
Tactics for improving content structure
The principles for optimizing your content for AI chatbots are no different from those aimed at helping human users. The content your chatbot processes should be:
- Scannable: Break up dense text into digestible paragraphs.
- Grouped in related sections: Organize content with similar material.
- Organized under meaningful headings: Create descriptive and hierarchical headings that reflect the content below.
- Sequenced in bulleted lists: Use when possible to present information concisely.
The wider benefits of well-structured content
Readability
Clear organization, concise paragraphs, and descriptive headings make your content more engaging and easier to understand. This improved readability can lead to increased time on page, better information retention, and a more positive user experience.
Usability
When content is well-structured, users can more easily find the information they need. This enhanced usability can increase satisfaction and lead to improved website performance.
Accessibility
Websites with properly structured content that use clear headings, lists, and logical organization are more accessible for website users with disabilities.
SEO performance
Search engines favor websites with well-structured content and assign them better rankings. Clear headings and logical content flow also help search engines index your content more effectively. This can lead to improved visibility in search results and increased organic traffic to your website.
A smart content strategy unlocks the power of AI
Developing and implementing a comprehensive content strategy enables your organization to better serve your website users and any AI application. Start by thoroughly auditing your current website content and identifying areas where each page’s structure and organization can be improved.
Organizations aiming to create more effective AI agents and chatbots confront a fundamental truth: The quality of your content directly impacts the quality of your digital experiences. Focusing on improving content structure not only improves the performance of AI applications, but also creates a more user-friendly digital environment for everyone.
Well-organized content is the foundation for accurate AI responses, enhanced usability, better accessibility, and stronger SEO performance. It’s a holistic approach that pays dividends across your organization’s digital experience. By improving the structure of your content to improve AI performance, you improve the experience for your users as well.
Making the web a better place to teach, learn, and advocate starts here...
When you subscribe to our newsletter!
* indicates required field