AI Adoption Has a Knowledge Problem
Generative AI has made it easier for organizations to experiment with automated customer and employee support.
Many companies already have product manuals, FAQs, SOPs, technical documentation, contracts, and internal policies. The obvious next step appears to be uploading those materials to an AI system and asking it to answer questions.
Documents can be ingested, indexed, divided into retrievable sections, and used by modern AI systems. The problem is not that documents are unsuitable for AI.
The problem is that uploading documents alone does not create a reliable knowledge service.
A large document collection may contain useful information, but it may also contain outdated procedures, duplicate instructions, conflicting policies, unclear terminology, or content intended for a different audience.
The quality of the AI experience depends on how that information is prepared and managed.
Documents Can Be Sources of Knowledge Without Being Ready for Every Question
A document and a knowledge base are not opposites.
A product manual can remain an important source document. A retrieval system may use it directly. In other cases, selected sections may be rewritten into focused articles, troubleshooting steps, decision trees, or internal reference content.
The appropriate structure depends on the information and the way people need to use it.
The practical question is not whether content is stored as a document or an article. It is whether the content is fit for the intended retrieval and service use case.
Customer Service Starts With User Intent
Customers rarely ask an AI system to summarize an entire manual.
They ask service questions such as:
- How do I reset my password?
- Why is this feature not working?
- How do I update my account?
- What should I do when this warning appears?
These are service intents.
To answer them consistently, an AI system needs access to relevant information and enough context to determine which information applies.
A Practical Method for Creating AI-Ready Knowledge
Organizations can improve existing content through a structured process.
1. Inventory and Classify Source Documents
Begin by identifying where relevant information exists.
Classify each source by topic, product, market, audience, language, sensitivity, and likely service use.
This first inventory often reveals duplicate files, obsolete versions, and information that has no clear owner.
2. Identify Authoritative Sources and Content Owners
Every important topic should have an accountable source and owner.
For example:
- Product procedures may belong to Product or Engineering
- Refund policies may belong to Finance or Customer Operations
- Privacy information may belong to Legal or Compliance
If two sources conflict, the organization needs a defined method for deciding which one is authoritative.
Without ownership, AI systems may retrieve information that is technically available but no longer approved for use.
3. Analyze Real Customer and Employee Questions
Review support tickets, search queries, chat transcripts, call reasons, escalation records, and unresolved inquiries.
This analysis connects the knowledge program to actual service demand.
4. Improve or Restructure the Content
Some content can be used with limited changes. Other content needs revision.
Possible improvements include:
- Clarifying ambiguous language
- Identifying applicable products or versions
- Adding titles that reflect real user questions
- Removing duplicate or conflicting instructions
- Creating focused articles where a long source is difficult to retrieve
AI can assist with classification, summarization, drafting, and translation workflows, but subject-matter owners should review the result.
5. Approve and Publish the Knowledge
Before content is made available to an AI agent or customer-facing help center, define its status.
Publishing should be treated as an operational decision, not merely a formatting step.
6. Test AI Answers Against Real Questions
Use representative questions from actual customers and employees.
Check whether the AI:
- Retrieves the correct source
- Gives a complete answer
- Includes relevant conditions
- Avoids unsupported claims
- Cites or links to appropriate content where supported
Testing should include difficult cases, not only straightforward questions.
7. Monitor Gaps, Accuracy, and Failed Resolutions
After launch, review:
- Unanswered questions
- Repeated escalations
- Negative feedback
- Search terms with no results
- Articles that are frequently opened but do not resolve the issue
- Conflicting answers across channels
- Questions that require human intervention
These signals show where the knowledge system needs improvement.
8. Review, Update, and Retire Content
Products change. Policies change. Software is updated. New markets launch.
Knowledge therefore needs a lifecycle.
A knowledge base becomes more reliable when maintenance is part of normal operations rather than an occasional cleanup project.
How Gemini Can Support the Knowledge Process
Google currently provides Gemini capabilities within supported Google Workspace applications, including Docs and Drive.
Depending on the organization’s Workspace edition, language, administrator settings, and current feature availability, Gemini can assist with drafting, refining, summarizing, and working with information in Workspace files.
Zendesk Knowledge Sources Need Configuration and Governance
Zendesk’s AI-agent documentation describes knowledge sources as information that AI agents can use when generating answers. Depending on the Zendesk product, plan, add-on, configuration, and current feature availability, organizations may be able to use Zendesk Help Center content and connect selected external knowledge sources through supported connectors or web crawling.
A well-configured AI service should define:
- Which sources are authoritative
- Which products and markets each source covers
- Which content is current
- Which sources should be excluded
- How permissions are handled
Zendesk Guide Can Connect Self-Service and Agent Workflows
Zendesk Guide provides knowledge-base and help-center functionality for organizations that want to publish customer-facing content and make knowledge available to support teams. Exact capabilities depend on the Zendesk product, plan, configuration, and current feature availability.
A shared knowledge foundation can support several service paths:
- A customer searches the help center.
- An AI agent uses configured knowledge sources to respond.
- Unresolved questions are recorded as signals for future knowledge improvement.
This creates a connection between self-service, AI assistance, and human support.
The value comes from the quality and governance of the knowledge behind those channels.
Multilingual Knowledge Requires Governance, Not Just Translation
International companies often need to support customers in English, Chinese, Thai, Vietnamese, Indonesian, Malay, and other languages.
Translation is only one part of the problem.
A localized article must also preserve:
- Product terminology
- Safety instructions
- Legal or policy conditions
- Market-specific processes
- Version and approval status
Knowledge Architecture Is an Operating Capability
AI models will continue to change, but organizations will still need to manage the information those models use.
A reliable knowledge program connects three layers:
- Knowledge — accurate, relevant, structured content.
- AI — technology that can retrieve, interpret, summarize, and generate responses from configured sources.
- Governance — ownership, approval, permissions, testing, localization, monitoring, updating, and retirement.
The model is only one part of the service system. If the source content is unclear or outdated, a more capable model may produce a more fluent version of the wrong answer.
How DEMETER ICT Can Help
DEMETER ICT’s service portfolio includes Google Workspace, Zendesk, AI business automation, and related implementation and support services. Our Zendesk Guide service describes help-center and knowledge-base capabilities for organizing information for customers and agents.
Based on these services, DEMETER ICT can discuss and support areas such as:
- Google Workspace and Gemini adoption
- Zendesk Guide implementation
- Help-center and knowledge-base structuring
- AI business automation
- Enterprise information search and Q&A use cases
- Knowledge workflows connected to customer-service operations
For organizations preparing customer or employee service automation, the practical starting point is not simply choosing an AI model or uploading every available file.
It is assessing whether the organization’s knowledge is accurate, governed, retrievable, and connected to real user needs.
Ready to assess and improve your AI-ready knowledge foundation?
About the Author
Mr. Carl Aldrich Wang is an International Marketing Specialist at DEMETER ICT, a Premier Partner of Google and Zendesk in the APAC region. DEMETER ICT serves over 4,600 business customers across APAC, including Greater China, with the largest customer base for Google and Zendesk services in the region. His expertise is in customer experience and global digital strategy with work that emphasizes aligning business goals with customer needs, enabling organizations to strengthen engagement, streamline workflows, and drive measurable growth.
