Preparing Your Knowledge Bases for ServiceNow AI: Governance, Structure and Content

[Baris Izci]

Summary

This article explains how to prepare knowledge and the wider ecosystem around it for ServiceNow artificial intelligence (AI) capabilities such as Now Assist and AI Search. It covers three areas. The knowledge ecosystem section (Section 1) shows where ServiceNow knowledge fits within the wider enterprise knowledge ecosystem. The governance section (Section 2) defines the core roles in the knowledge management process and what each role is accountable for. The structural formatting best practices section (Section 3) shows how to build knowledge bases and articles so AI can find, read, and use the

Why this article

The ServiceNow documentation and community already offers many excellent articles on preparing knowledge for AI. Most of them focus on individual technical aspects or capabilities, such as AI Search configuration, Now Assist features, or article formatting. In practice, however, the success of AI depends just as much on organizational factors: how ServiceNow fits into the wider knowledge landscape, who is accountable for knowledge, and how knowledge is governed. This article brings these perspectives together to provide a holistic set of recommendations.

1. Positioning ServiceNow in the Enterprise Knowledge Ecosystem

Stakeholders measure the success of ServiceNow AI capabilities by the reliability of their answers, and that reliability depends on the quality of the knowledge being retrieved. In most organizations, knowledge is stored across several platforms and owned by different teams. More mature organizations govern this landscape through an enterprise wide knowledge management methodology. Before scaling AI, ServiceNow needs a clearly defined place within this ecosystem. This is especially important when ServiceNow is positioned as the enterprise AI front door.

The first step in defining this place is to look at how knowledge is already managed in the organization. Where an enterprise knowledge management methodology is in place, ServiceNow initiatives should be connected to it from the beginning. This alignment is a key factor in their long-term success.
Building on this, the scope should be defined early. ServiceNow can serve the entire enterprise as the central knowledge repository, or it can focus on a specific domain and user group. Each option requires different capabilities, governance, and content ownership. For example, an IT service desk scope can work with IT owned content and a single approval flow. An enterprise-wide scope brings HR, Finance, Procurement, and other domains into one place. That requires domain specific knowledge bases and owners, user criteria for access control, and governance across all entities.

When several platforms coexist, clear platform boundaries are essential. Most organizations store knowledge in tools such as ServiceNow, SharePoint, Confluence, and Dropbox. It should therefore be decided which types of knowledge belong in ServiceNow and which belong elsewhere. Where an enterprise architecture team exists, these boundaries should be defined and protected together with that team. Without architectural guidance and governance within each tool, the same information ends up documented differently in several places, each version following its own lifecycle.

This leads to a real risk: contradictory answers for users. Assume an organization uses ServiceNow EmployeeWorks, powered by Moveworks, as its enterprise-wide AI front door, connected to several platforms beyond ServiceNow. Bringing knowledge from all these systems into one conversation is a major strength, but it also exposes every inconsistency between them. For example, an employee asks EmployeeWorks how many days of parental leave they are entitled to. A current HR article in ServiceNow says 90 days. An outdated page in SharePoint says 60 days. Moveworks retrieves from both sources. It may return either answer or combine both into one confusing reply. The employee cannot tell which is correct and may plan their leave on the wrong basis. When they find out, trust in AI suffers, even though the root cause is inconsistent knowledge, not the technology. Every type of knowledge therefore needs one authoritative source, and other platforms should link to it instead of keeping their own copy.

A single authoritative source only works if the people behind each platform keep it that way. This is why a connected knowledge community matters. Leading organizations link knowledge management roles across domains and platforms and keep them in regular contact. When a policy changes in one place, the owners of the other platforms know about it and update or remove their copies. This shared ecosystem helps AI return the same answer, even when several systems run in parallel.

2. Defining Roles and Guardrails for Knowledge Management

A connected knowledge community depends on clearly defined roles, and this is where governance comes in. The ecosystem defines where knowledge lives, while governance defines who is accountable for keeping it accurate, current, and useful. Without clear roles, even well-structured knowledge bases lose quality over time. Users stop trusting the content, and AI answers become less reliable.

2.1 Process Owner

The Process Owner sits at the center of the governance model and is accountable for the effectiveness and efficiency of the Knowledge Management process as a whole.

A large part of this role is keeping the process visible. The Process Owner regularly communicates the mission, goals, objectives, and changes of the process to all entities. Over time, this helps Knowledge Management become part of daily work rather than a one-time initiative. The Process Owner also keeps the process aligned with organizational goals such as the AI vision.

The Process Owner reports upward as well. The Knowledge Management Overview Dashboard in ServiceNow shows how the process performs, and the Process Owner shares these insights with senior management on a regular cadence. This keeps Knowledge Management on the leadership agenda. The same reports reveal bottlenecks, which become the starting point for targeted improvements. In addition, the Process Owner makes sure Knowledge Managers and Contributors receive proper training on standardized templates, keywords, metadata, and formatting best practices.

Training sets expectations, but guardrails in the platform make sure they are met. The Process Owner therefore puts the following controls in place:

• Review before every change: Instant publishing and instant retirement are switched off, so no article is published or retired without review.
• Ownership and descriptions: It is worth considering making the knowledge base owner and the knowledge base and category descriptions mandatory. This helps ensure that every knowledge base has someone accountable for it and that its purpose is clear to users and AI.
• Expiry notifications: Knowledge Managers receive an automatic notification before an article expires.
• Article flagging: Users can flag articles that look wrong or outdated.
• Search dictionaries: The synonym dictionary and the stop word dictionary are kept up to date, so search continues to return relevant results.

2.2 Knowledge Manager

While the Process Owner sets direction and guardrails, the Knowledge Manager makes them work in practice. This role owns the knowledge management plan and is mainly responsible for approving, publishing, and regularly reviewing articles. In this way, Knowledge Managers act as the quality gate between what is written and what users and AI rely on.

2.3 Knowledge Contributor

Contributors create the knowledge itself. They work closely with Knowledge Managers and peers to close knowledge gaps in three ways:

• New articles: They write new articles when top search queries return no results.
• Updates: They update existing articles based on user feedback.
• Retirement: They retire content that is no longer valid.

3. Structuring Knowledge Bases and Articles for AI

With clear roles and guardrails in place, the focus shifts to the content itself. Governance defines who manages knowledge, but the content defines how good the answers are. Now Assist and AI Search build their answers on published articles. The way knowledge bases are set up, and articles are written therefore shapes every answer users see.

3.1 Active Knowledge Base Ownership

Every knowledge base needs an active owner. Without one, content slowly drifts: articles remain incomplete, duplicates accumulate, and outdated information stays published. When AI surfaces this content, users lose trust not only in the article but in the AI capability as a whole. Active ownership is the simplest and most effective protection against this.

3.2 Knowledge Base Structure

A clean knowledge base structure supports AI in two ways.

The first is security. User criteria control who can see which content. Search profiles apply these rules when AI retrieves content, and the rules are defined at knowledge base level. A clean setup ensures that AI only returns content a user is allowed to see.

The second is retrieval. Clear knowledge base definitions improve the quality of the AI Search index, and well-defined categories with meaningful descriptions improve it further. This helps the search engine identify quickly which knowledge base and category hold the right answer.

3.3 Metadata Completeness

AI Search uses hybrid retrieval, which combines lexical signals (the exact words) with semantic signals (the meaning). Complete article metadata feeds both. Together with clear knowledge base and category descriptions, it makes retrieval more precise.

3.4 Article Formatting

Formatting may look like a cosmetic topic, but for AI it is structural. It determines how easily an article is found, interpreted, and summarized.

• Search-friendly titles: Use the words users actually search for and describe the symptom rather than the technical cause. For example, “Cannot access the internet” works better than “Network is down.” The title is one of the strongest signals AI uses to match an article to a query.
• Abstracts: Start longer articles with a short abstract that states which question or problem the article answers. AI uses this summary to decide whether the article matches a query, which improves the precision of answers in Now Assist and in AI Search Genius Results.
• Headings: Use real heading styles instead of manually formatted text, such as bold or enlarged fonts. Headings give AI a clean outline of the article and also support screen readers.
• Numbered steps: Write step-by-step instructions as real numbered lists rather than typed numbers. This keeps the sequence clear for both users and AI.
• Short paragraphs: Keep paragraphs short, with one idea each. This makes it easier for AI to extract and summarize the right information.
• Templates: Use shared templates to keep structure and style consistent across articles. A predictable pattern helps AI interpret every article in the same way.
• Knowledge blocks: Manage recurring content, such as service desk contact details, support hours, and compliance notices, through knowledge blocks. The block is written once, and every article that uses it stays current automatically. For example, if the service desk phone number changes, updating one block corrects it in every article that uses it. This prevents outdated copies of the same text from surviving in individual articles and appearing in AI answers.
• Text formatting: Use basic formatting such as bold, italic, and underline, which AI interprets well, and avoid custom fonts and colors. Use emphasis sparingly, because overuse weakens its meaning.
• Tables: Keep tables simple and use them only for genuinely tabular data. Complex tables often lose their structure when AI processes them.
• Emojis and special characters: Avoid emojis and special symbols and spell out the concepts in words instead. AI interprets plain text more reliably than symbols.
• Images and videos: Back every image and video with text, such as alt text, captions, transcripts, and key steps written into the article body. For example, instead of “Follow the steps in the screenshot below,” write “Open Settings, select Network, and then select Connect VPN,” and keep the screenshot as visual support. AI can process images, but not as reliably as text, and text alternatives also support users with visual or hearing needs.
• Plain language: Avoid jargon and write out abbreviations in full the first time they are used. For example, write “Restart your laptop” instead of “Power-cycle your endpoint.” Plain language helps people and AI alike.
• Descriptive links: Write link text that tells the reader where the link leads. For example, write “See the Parking Pass Policy” instead of “Click here.” AI and screen readers often read links on their own, without the surrounding sentence, so generic text such as “click here” or “more info” tells them nothing about the destination.
• Self-contained FAQ answers: In frequently asked questions (FAQs), make every answer stand on its own, because AI may show the answer without the question. For example, to the question “Can I use my personal laptop for work?”, instead of “No,” write “No, personal laptops cannot be used for work. Request a company laptop through the Service Portal.”

Each of these practices is small on its own. Together with a well-positioned ecosystem and clear governance, they make sure that what reaches users and AI is knowledge they can trust.