There is a predictable moment in many digital transformation projects. The vision is clear: fewer support cases, better self-service, useful AI-generated answers, and conversations that help customers get things done.

Then someone looks at the content. Thousands of documents. Inconsistent metadata. Old product pages. Duplicate knowledge articles. Different teams describing the same subject in different ways. The project stalls: “We need to fix the content first.”

For knowledge managers, the concern is real. Outdated guidance, conflicting versions, and unclear ownership can undermine any experience built on that knowledge. Generative AI makes those concerns more visible because it presents information as an answer, not a link to sort through.

You still need reliable content for the experience you launch. But you do not need to resolve every content issue across the organization before delivering value.

(Why waiting for clean content rarely pays off is its own discussion.)

A practical starting point is to choose a useful, defined use case, identify the knowledge you trust, and control what that experience can retrieve. Then learn from real questions, improve the content that matters most, and expand as confidence grows.

Put knowledge decisions into the delivery experience

Knowledge managers make the decisions that give an experience its credibility: which sources are authoritative, who owns them, which audiences they serve, and when they need review. Those decisions need to carry through to what people can find and what AI can use to answer.

At Coveo, we provide a shared retrieval and relevance foundation for putting those decisions into practice. Content can remain in its source systems while being indexed for discovery. Different experiences can retrieve different subsets of that knowledge, with relevance tailored to their purpose.

For example, an organization may have current product documentation, employee troubleshooting notes, archived policies, and community discussions. Each can have value, but their suitability depends on the audience and the question.

Our query pipelines let teams apply rules and machine learning to different search experiences. Filters can limit which content is returned, while ranking rules can influence prominence. Inclusion-based filtering lets teams introduce new sources deliberately, rather than automatically exposing everything they index.

Security works differently: it is built in, not bolted on. When Coveo indexes a source, it also indexes that repository’s permissions, so every query and every generated answer respects them automatically. Users only find, and AI only answers from, content they can already access in the source system. Pipeline filters narrow an experience’s scope on top of that; they never widen it.

Define readiness for a specific use case

Content readiness becomes more useful when it describes what a piece of knowledge is ready to do. A long technical reference may work well as a search result. A focused troubleshooting article may be a better source for a generated answer. An internal escalation procedure may help an employee but have no place in a public experience.

A simple classification can help:

Ready for search: useful and discoverable, even if the structure could be improved.
Ready for generated answers: authoritative, current, and focused on the questions in scope. The full solution can be spread across several documents, as long as each piece can be found.
Restricted to a specific audience: appropriate only for users with the necessary permissions.
Needs review: potentially useful, but awaiting an owner’s assessment before inclusion in the intended experience.

These categories can overlap. A document may be ready for both search and generated answers while remaining restricted to employees. The aim is to make deliberate choices about use, ownership, and access.

Consider a customer who asks how to configure a product, then follows up: “Does that also apply to the older version?” A useful response requires retrieving guidance for the right version, audience, and task. Familiar knowledge-management disciplines — clear ownership, version information, and authoritative sources — become essential to the conversation.

Build generated answers and conversations on trusted knowledge

Coveo Relevance Generative Answering brings answers grounded in enterprise content into the search experience, with citations that let users explore supporting sources. It retrieves relevant items and passages from the index to ground the response, while enforcing content permissions during retrieval.

The question for a first deployment is specific: which knowledge can reliably support the questions we want this experience to answer? Start with that scope and validate the results before adding more sources or topics.

Coveo’s content guidance for generative answering favors focused, current material that answers questions clearly. Removing boilerplate, checking indexed content, and clarifying ambiguous instructions can improve the material used for grounding. This gives the content team a targeted improvement program tied to an actual experience.

When a question needs further exploration, Coveo Conversational Search lets users ask follow-up questions while retaining conversational context.

For knowledge managers, each follow-up is another opportunity to deliver the right guidance — and another reason to be deliberate about which sources the experience can use.

At Xero, Coveo reported approximately 40% less customer search time and more than 20% fewer search sessions requiring additional support with generative answers in Xero Central. Those results illustrate the practical value of helping customers use knowledge through an answer within their existing support journey.

Improve the content that users need most

Content quality and business importance vary. A newly reviewed guide may deserve greater visibility than an older document on the same topic. A critical policy update may need attention before it has accumulated much usage data.

Metadata and measured ranking adjustments can help reflect those differences. Use them selectively, preserving relevance to the user’s question. Promoting an article is not a substitute for reviewing its accuracy or deciding whether it belongs in the experience at all.

Once the experience is live, knowledge managers can use evidence to prioritize the next round of work. Search behavior, answer review, user feedback, and conversation analysis show the team:

Which high-demand questions have no authoritative answer?
Where do users repeatedly ask for clarification or missing steps?
Which product versions or conflicting sources cause confusion?
Where does a generated answer fall short even when relevant content exists?
Which gaps should go back to a content owner, and which sources are ready for broader use?

Clicks alone cannot tell the whole story when a user may receive a useful answer without opening an article. Combine usage signals with answer-quality reviews and appropriate measures of successful self-service. Check that the evidence supports the outcome you are trying to assess.

Zoom’s experience connects knowledge coverage with customer outcomes. Coveo’s 2025 customer awards announcement reports an 89% reduction in content gaps and a 20% increase in self-service success across Zoom’s AI search and Relevance Generative Answering program.

For knowledge managers, the connection is valuable: improving the available knowledge and improving how people use it can advance together. Each round of delivery and review gives the content program a more concrete set of priorities.

Clear headings, focused topics, useful metadata, and current information can also help other discovery channels. When content is published appropriately, these improvements may support external search and answer engines as well. The immediate priority remains making the knowledge useful for the audience and task in scope.

From deflection to resolution

In service, the payoff of good content is easy to measure: every question it answers is a case nobody has to open. That is how support teams lower the cost to serve, and it is why self-service is usually where they start with AI. Forcepoint is a good example. It brought Coveo Generative Answering into its Customer Hub, Salesforce, and in-product experiences, and Coveo reports a 70% self-service success rate and more than a 40% reduction in time to resolution.

None of this requires the whole solution to sit in one article. Relevance Generative Answering already builds answers from passages across several documents. The Coveo Search Agent, which powers Conversational Search, takes it further. It works out what the user is asking, decides what to look for, retrieves content, and checks whether what it found is enough. If it isn’t, it searches again before answering. Everything it has found stays in context as the conversation goes on, so it can put together a solution from pieces spread across your documentation.

Think about someone with no inside knowledge of your products, digging through your documentation to fix a problem. If they would eventually piece the answer together, an AI agent can too. It doesn’t have your support team’s instincts, but it can keep searching, the way a patient person would. Your content still has to be accurate, current, and findable. It doesn’t have to be perfect.

This is also where things are going. We are working with select customers on using the same loop for complex support cases, where the agent keeps reasoning and searching for longer, with more context, until it can deliver a resolution rather than an answer. Further out, we expect agents to draw on context specific to each company and to show solutions in richer formats, like step-by-step layouts with images instead of a wall of text.

Expand through progressive trust

A practical rollout can begin with one audience and one valuable set of questions:

Choose the use case and define what a successful outcome looks like.
Identify the trusted sources, responsible owners, and review expectations.
Configure the retrieval scope for that experience, knowing source permissions already carry through from indexing.
Test representative questions and review the supporting content and answers.
Launch to a defined audience and measure quality, usefulness, and outcomes.
Use the evidence to improve knowledge and expand coverage deliberately.

Generated answers can serve questions with clear supporting knowledge. Conversational search can help when users need to clarify their situation or explore further. The choice should follow the user’s task and the quality of the available knowledge.

That foundation can also support broader adoption. In its July 2025 business update, Coveo reported that SAP was expanding its use of Coveo following its generative AI implementation at SAP Concur, supporting a global customer-experience strategy that included agentic deployments. The progression illustrates how an initial implementation can become part of a wider knowledge-delivery strategy.

Make content improvement part of delivery

When an organization says, “Our content isn’t ready,” the next question should be: “Which content can we trust for a useful first experience, and how will we know it is working?”

That question gives knowledge managers a concrete role in moving forward. It calls for clear scope, accountable owners, appropriate access, and evidence about where the knowledge helps or falls short. One simple test applies throughout: if a person could find the answer in your content, so can an AI agent.

Knowledge changes. Products evolve, policies are updated, and users ask questions you did not anticipate. Coveo serves as a shared foundation for carrying knowledge decisions into search, generated answers, and conversations as the content behind them improves.

Start with the knowledge you trust. Learn from how people use it. Make each improvement count toward the next experience you are ready to deliver. For the deeper reason “clean content first” holds so many programs back, read why that instinct is often backward.