Admins can use AI Answer Generation to create a conversational bot that pulls information from multiple sources, including the knowledge base. This feature integrates a large language model (LLM) with Retrieval-Augmented Generation (RAG), enabling the bot to generate accurate, relevant responses based on customer-approved content. By using RAG, the bot ensures answers align with your organization's specific information while leveraging the LLM's natural language abilities. When enabled, the bot uses the five most recent interactions and up to five relevant knowledge base articles to answer user queries. Admins can enable AI Answer Generation at the bot level, controlling its use and functionality.
AI Answer Generation involves three key stages: query processing, external LLM interaction, and response generation.
When you submit a question, the system analyzes it to understand the intent and context.
To generate a response, the following information is sent to an external Large Language Model (LLM):
The LLM generates a response based on the provided information, ensuring that the answers are relevant, context-specific, and maintain conversational continuity, all while prioritizing data privacy and security. You can adjust the tone of your bot's responses to match your brand's voice or desired style.
You can customize the tone of your bot's responses by selecting a predefined tone for AI-generated answers.
For example, a technical support organization may choose a Technical tone for their Tech Support bot. This tone prompts the AI to use more specialized language and precise terminology when responding to inquiries.
For instance, if a user asks, "Why is my computer running slowly?", the response could be: "Several factors can contribute to reduced computer performance, such as insufficient RAM, fragmented hard drive, or resource-intensive background processes."
You can enhance your bot's responses by adding source citations to improve credibility and the quality of information. By referencing knowledge base sources, AI-generated answers become more reliable and transparent.
For instance, a financial services company enabling inline citations for their virtual agent discussing investment products might see responses like:
Question: "What are the benefits of diversifying my investment portfolio?"
AI Response: "Diversifying your investment portfolio can help mitigate risk (1) and potentially improve returns over time (2). It involves spreading investments across various asset classes, such as stocks, bonds, and real estate (3)."
This approach enhances the credibility of the information and helps users quickly identify the sources of specific claims.
You can use AI disclosure notices to inform users that they are interacting with an AI-powered virtual agent.
For example, a tech support service might use a custom notice like:
"I'm an AI assistant trained to help with technical issues. While I strive for accuracy, please verify critical information with our human support team."
This notice appears after AI-generated responses, setting clear expectations for users. When answering a complex technical question, the bot displays the notice, reinforcing transparency and accuracy.
By implementing these features, organizations can enhance the question-answering capabilities of their Zoom Virtual Agent, providing more accurate, brand-aligned, and trustworthy responses to user queries.
You can use dynamic variables to personalize your bot to give more relevant and engaging responses. These variables incorporate contextual information into AI-generated answers for a tailored experience.
For example, an e-commerce company can use variables such as the customer’s name and their most recent purchase. If Sarah asks, "How do I set up my new laptop?", the AI could respond:
"To set up your new XYZ Laptop, Sarah, first ensure it's fully charged. Then, follow these steps..."
This feature enhances the bot's ability to deliver customized, user-centric responses.
You can customize your bot’s responses to reflect your company’s brand identity by incorporating brand-specific information into Zoom Virtual Agent’s knowledge base.
For example:
"You are an AI assistant for EcoTech Solutions, a leader in sustainable technology. Our mission is to provide innovative, eco-friendly products that reduce environmental impact without compromising performance. We are committed to sustainability, cutting-edge research, and exceptional customer service."
If a customer asks, "What makes your products different from competitors?" the AI might respond:
"EcoTech Solutions' products stand out due to our commitment to sustainability and innovation. Our eco-friendly designs minimize environmental impact without sacrificing performance. For instance, our latest laptop model is made with recycled materials and consumes 30% less energy than standard models, demonstrating our dedication to sustainable technology."
For more tips to effectively implement brand information, see Tips for crafting effective brand information.
You can define specific guidelines and policies to shape the bot's responses, ensuring they align with your organization's priorities and requirements.
For example, a healthcare provider might include the following guideline:
"Always prioritize patient safety. If a question involves potential health risks, advise consulting a healthcare professional and provide the emergency helpline number."
If a user asks, "What should I do for severe chest pain?" the AI might respond:
"Severe chest pain can be a serious symptom. While it may have various causes, prioritizing your safety is essential. We strongly recommend consulting a healthcare professional immediately or calling our emergency helpline at [number]. Please note that this AI assistant cannot provide a medical diagnosis or treatment."
For best practices on creating effective response rules, see Tips for creating effective response rules.
You can ensure transparency and trust in your bot's responses using source citations and an AI disclosure notice.
When crafting inputs for the Brand Information and Response Rules sections, follow these tips to create clear and effective prompts that Large Language Models (LLMs) can easily interpret, especially those used in AI Answer Generation. This section provides guidance on writing impactful prompts, selecting relevant variables, showcasing your brand’s identity, and defining guidelines to shape your bot’s behavior effectively.
To maximize the benefits of AI Answer Generation and reduce potential risks, it’s important to follow certain best practices. When creating inputs, focus on clarity, brevity, and relevance to enable the AI to process and apply the information effectively.
Personalize your bot's responses by using dynamic variables. Select variables from your bot's preconfigured settings to add context and make AI-generated responses more relevant and engaging. Use clear and descriptive labels for variable names to ensure the LLM understands and applies them correctly without needing additional explanation.
To effectively present your brand information, use a structured format that includes key details such as your company name, industry, mission, and other defining attributes. Summarize your brand's core elements into concise, factual statements that highlight your identity and unique value proposition.
| Category | Description |
|---|---|
| Company | EcoTech Solutions |
| Industry | Sustainable Technology |
| Mission | Provide eco-friendly tech products without compromising performance |
| Unique Selling Point | 30% more energy-efficient than standard alternatives |
| Core values | Sustainability, Innovation, Customer-centric, Quality |
You can create clear response rules that align with your brand policies and customer service standards. Create a clear set of guidelines that direct the AI's behavior to focus on critical do's and don'ts that align with your brand policies and customer service standards.
AI Answer Generation offers many benefits, yet it's important to be aware of a few key considerations.
To address these considerations, the following is recommended:
Follow these practices to maximize the benefits of AI Answer Generation and reduce potential risks and limitations. For any questions or help to optimize the AI Answer Generation, contact our Support team.
Zoom's federated AI approach uses third-party models, such as those from OpenAI and Anthropic, and its own AI models for certain features to deliver high-quality results.
Zoom ensures its subprocessors meet obligations equivalent to those in its Data Processing Agreement.
As part of its third-party risk management program, Zoom conducts security assessments of subprocessors at least annually. Independent audit firms evaluate Zoom's risk management controls for various security certifications, which are available to customers on the Zoom Trust Center.
When a Zoom AI feature uses a third-party model, the provider may retain content in the U.S. for trust and safety purposes for up to 30 days, unless the law requires a longer retention period.