AI Adoption in Documentation

Introduction

Since I am very interested in AI in general and especially in ways to use it in technical writing, I attended this interesting webinar organized by the Expert Team. The purpose of the webinar was to share valuable insights on AI adoption in documentation with the community.

This is a brief summary of some parts of it, and for those interested in listening to it, there is a video recording on YouTube.

About Zoomin

Zoomin is a content delivery platform, not a documentation authoring tool. It enables technical communicators to choose the tool they want to use. This means that you can match your documentation type to the tool that best suits you, and then with Zoomin just seamlessly publish to your documentation portal.

In March 2023, Zoomin introduced Zoomin GPT, an AI-assisted search feature for their documentation portal.

Note: At Zoomin at the time, they needed to rapidly deploy new documentation as their company was in a hyper development phase. They trained the product management team to write their product documentation.

The docs team moved from tools such as DITA to Confluence so that the managers could use the tools they felt the most comfortable with. Then they created five different KB (knowledge base) formats as templates, enabling the managers to create many different knowledge-based articles with the docs team serving as editors, trainers, and writers.

The decision of whether to use AI as a tool was something that at Zoomin could do on their own, but after they tested it, they brought AI into their product offering, which was something much larger.

They had two main use cases in mind:

Making end users’ lives easier – because people today are spending less time reading, and they want fast, accurate answers.
Making technical writers’ lives easier – by launching the Editorial Co-pilot which gives technical writers AI-based recommendations on how to improve the content they’re writing.

Content Improvement Guidelines

When using AI to enhance your content, you have two options:

Using a Generic Prompt where the AI is asked to provide suggestions and ideas for improvement without specifying any particular guidelines.
1. Example: “I’m Mira, I’m a technical writer. I want you to help me improve this topic page. Give me suggestions, bullet points, and so on.”
Using Company Authoring Guidelines to ensure AI recommendations align with company standards.
1. Example: “Here are my guidelines. Please provide very clear recommendations on how I can improve my work to align with these standards.”

AI has enabled tech writers to create more documents, enhancing the breadth and depth of their documentation.

AI can also helps in formulating more focused questions for subject matter experts (SMEs). For instance:

Example prompt: “As a technical writer, I value a, b, and c. Tell me why I would want to enable this feature which enhances x, y, and z.”

The Importance of Tech Writers in the Age of AI

Three points to consider:

No application can be built, and no AI functionality can be implemented successfully, without quality content.
Conveying the Value of Documentation
Embracing AI Adoption – technical writers are at the core of any successful implementation of AI applications..

First Point

If your development or IT teams haven’t already asked, “Is our knowledge base ready for AI?”, it’s likely to happen soon.

Second Point

Technical writers are no longer just a “cost center” but can be a “revenue growth engine” for their companies. There is clear data showing that the quality of documentation has a significant impact on users’ purchase decisions. Companies need to choose and track the right KPIs to measure the impact of documentation, such as:

Website traffic to the documentation portal
Returning users
Self-service rate

Measuring the number of visitors to the documentation portal is a crucial metric, as it indicates user interest and engagement.

At Zoomin they have seen that if you visit documentation portal twice in the past month, you are actually more likely to become a paying customer. It’s a clear indication that you’re interested in the product and that you’re actually going to become a paying customer.

Optimizing Search with Synonyms and Boosting

Search Synonyms

Search synonyms are crucial for optimizing search. They allow users to find relevant results even when their terminology differs from the internal terminology used in the documentation.

Identifying and adding more synonyms is an ongoing process.

Boosting Specific Content Types

Boosting is another technique used to enhance search results.

When a user searches for a general concept, like “uploading content”, the system can prioritize the results for “About” topics that provide an overview of the concept. If the user adds more specific keywords, like “How to upload content”, the system can then prioritize the task-oriented topics that provide step-by-step instructions.

These techniques enhance user experience and increase the likelihood of users finding the information they need.

Data-Driven Documentation Improvement

The technical writing team closely monitors the performance and usage metrics of the documentation. If they notice any unusual trends or anomalies, they jump on it immediately.

For example, if they notice that users are consistently asking a particular question but not finding the relevant answer, this is a clear indication that there is a gap in the documentation. The team’s quickest solution is to add the frequently asked question to the FAQ section. This provides the AI-powered search system with additional content to reference.

Handling Gated Content with GPT and AI

Gated content is not accessible through search engines like Google. It can only be accessed by authenticated users of the system.

When it comes to AI, you want to ensure that the AI engine is not being fed information from the gated content that you don’t want public users to see. This is currently handled properly by Zoomin.

The key is to have control over the visibility and accessibility of your content, whether it’s for the public or specific user groups, and to ensure that AI-powered search respects those content access rules.

Keeping Gated Information Out of the LLM (Large Language Model)

The main idea is to either completely avoid sending gated content to the LLM, or to carefully scrub any sensitive information before doing so. This helps maintain the security and privacy of your gated content.