GenAI only improves the customer experience you can deliver when it draws on what your business actually knows. A model that writes fluent text but has never seen your policies, claims history, or customer records can produce a polished letter that is simply wrong.

Most, if not all, enterprises have already tried a generic AI writing tool. The drafts come quickly, then someone spends an hour checking every figure, rewriting the tone and confirming nothing violates a regulation. The time saved at the start is lost at the end.

This blog explores a different approach, knowledge-driven GenAI, and why it matters most for teams that send high-stakes customer communications.

Why generic LLMs fall short in customer communications

A general-purpose large language model (LLM) knows a great deal about the world and almost nothing about your customers. It was trained on public text, not on your claim files, product terms, or last week’s call with a policyholder.

That gap shows up in three predictable ways:

Accuracy. Without access to the real record, the model fills gaps with plausible guesses. In a billing notice or claim letter, a plausible guess is a compliance problem.
Context. A good response depends on history: what the customer already asked, what was promised, which policy applies. A generic model starts every request from zero.
Brand and governance. Regulated industries need approved language, consistent tone and an audit trail. Free-form generation offers none of these by default.

The result is that generic AI often shifts work rather than removing it. Drafting gets faster; reviewing gets slower.

Buyers already feel this. In a June 2026 Foundry survey commissioned by OpenText, 212 enterprise CX and technology decision-makers named data quality and access (40%) and compliance or risk concerns (36%) as the top two barriers to expanding AI use.

What knowledge-driven GenAI means

Knowledge-driven GenAI connects the model to your governed enterprise content, so every draft starts from the facts on file rather than from general training data. The model still writes; your systems of record supply the substance.

It draws on governed content, including claims, policies, call logs and customer workspaces, to produce communications that are accurate, rich in context and on-brand.

A grounded request typically runs in four steps:

An employee starts a task, such as replying to a customer inquiry.
The AI retrieves the relevant governed content for that customer and case.
It drafts a response using those facts and the organization’s approved style.
The employee reviews, adjusts and sends.

The key word is governed. Governed content already controls who can see what, which version is current and how records are retained. Grounding AI in that content means the AI inherits those controls instead of working around them.

Humans keep the judgment

The goal is to take manual work off employees’ plates, not to take employees out of the loop. Grounded AI acts as a co-pilot: it handles the gathering and first draft, and people keep the judgment.

That division of labor plays to each side’s strengths:

When the repetitive work shrinks, the same team can respond faster and spend more of its time on the conversations that build loyalty. Productivity goes up, and so does the quality of engagement.

In practice: answering a claims inquiryConsider an insurance customer who emails to ask why part of a claim was not paid. It is a routine question with high stakes: a wrong answer creates a complaint, and a slow one erodes trust.

Without grounding, a service agent opens the claim system, the policy document and the call log in separate windows. They piece together the history, then write a reply from scratch or paste a generic AI draft that needs heavy correction.

With knowledge-driven GenAI, the agent asks for a draft reply. The AI pulls the claim details, the relevant policy terms and the notes from the customer’s last call. It drafts a clear explanation that cites the actual coverage limit and reflects what the customer was told before. The agent checks it, adds a personal line and sends it.The agent’s expertise is still in the message. What disappears is the hunting and retyping.

Questions to ask when evaluating GenAI for customer communications

If you are comparing AI options for customer communications, these questions separate grounded tools from generic ones:

Where does the AI get its facts? Look for direct connection to your governed content, not just a prompt box.
Does it respect existing permissions? An agent should only see, and the AI should only use, what that agent is allowed to access.
Can it follow brand and regulatory standards? Approved language and tone should be built in, not bolted on during review.
Who has the final word? A co-pilot model keeps a person accountable for every message that goes out.
Does it fit how your team already works? AI that lives inside existing workspaces gets used; a separate tool often does not.

Take a self-guided tour of OpenText™ Experience Aviator and see how knowledge-driven GenAI drafts customer communications from your governed content, at your own pace.

Dana Morio

Dana Morio is a Product Marketing Manager at OpenText with 20+ years expertise in Voice and Unified Communications. She is a technology veteran, with experience in messaging, mobility, cloud, on-premises and hybrid applications for secure enterprise communications.