GenSearch persona prompt best practices and examples
Expert’s out-of-box settings are intended to capture a baseline persona for a variety of use cases. We recommend testing your content first before making adjustments.
Personas differ based on the task that is being performed. This information is intended for CXone Mpower Expert’s generative search tool. Steps and recommendations for Expert may vary from those for other generative AI platforms or tools.
Best Practices
- Create a persona document. This will help you outline the personality traits you value, and you can use that document to construct effective persona prompts.
- Establish your GenSearch personality in the fewest words possible. Too much information and instructions can lead to slower processing time and attention issues. If you include too many topics, your guidance about which to include or exclude may be ignored.
- Use the Assistant field to expand GenSearch's understanding. Clarify your terminology and the kinds of site visitors you engage with.
- Example: We are a storage facility and most of our site traffic is prospective customers looking for a unit to rent. When we say "unit" we mean a storage unit for personal belongings.
- Set the target response length to 250 words or less. Longer responses take more time to generate, which can leave customers frustrated with the GenSearch UX.
- Limit the AI's knowledge resources. GenSearch will use "general knowledge" unless otherwise specified. To avoid this, indicate in the output instructions to only use information pulled from available articles.
- Consider adjusting your threshold. If precise answers are critical, set a higher threshold. If broader context is helpful, a lower threshold can allow for more information. Expert recommends a 0.7 threshold.
- Keep your Kernel limit optmal. Focusing on the top 5 most relevant Kernels gives the LLM enough context to confidently answer a query, without being hampered by irrelevant or extraneous content. Because LLMs tend to remember the first and last pieces of content best, too much information could exclude critical details. To avoid this, we do not support more than 10 Kernels.
Persona examples
Your persona can be set up however you prefer. The following are some examples to help you set up and refine your persona.
Assistant
This tells GenSearch who it is, what we liken to its "personality." This can include, among other things, its expertise, responsibilities, and voice or tone.
- You are a polite and knowledgeable subject matter expert trained to summarize articles found on a website.
- Never forget you are a helpful and legally responsible agent who wants to answer questions related to filing taxes in the United States.
- You are a polite and knowledgeable subject matter expert on Product X. Product X is the new brand name for Product A. Product A is a SAAS XXX product.
- You are a legal advisor working for Company Y. You provide positive and negative statements on questions related to an account holder.
User
This gives GenSearch context about your users. This can include information such as why the user might be asking questions, or how familiar they are with your products or services.
- I am a customer of Company Y that needs help resolving my problem.
- I am trying to understand and customize the Software Z platform.
- I am a brand new customer looking to learn and looking for support.
Output instructions
This tells GenSearch how to format responses.
- Be brief, use bullet points and step by step instructions when necessary respond with language a 6th grader will understand.
- For queries asking for instructions, provide a numbered list of steps to achieve the intended outcome. In your response, it is important to avoid mentioning which articles were summarized. Do not explain your answer or mention what it is based on. Only answer the query if it is related to the information provided. If the query can be answered, start your reply by answering the query immediately without any introduction.
- Provide direct, accurate answers without additional context or commentary. Skip pleasantries and opinions, and use unambiguous language.
- Do not include or generate HTML of any kind.
Missing information response
This tells GenSearch how to respond to a query when it cannot find relevant information in your knowledge base.
- I could not find that in the knowledge base.
- I could not find an answer to your question in the Help Center. I can only answer questions related to [specified topics].
Number fields
Target output word length: 50
The target output is the maximum number of characters a response may contain.
Score threshold: 0.75
The threshold is the relevancy distance between Kernels in the vector database. The more closely related Kernels are, the smaller their threshold will be.
Kernel limit: 5
The Kernel limit is the maximum number of relevant Kernels used to create a response.