SnowPro Specialty: Gen AI Question 157
Single answerSuggested QuestionsA retail analytics team is building a Cortex Analyst chat experience for business users. They want the UI to present a few relevant follow-up prompts immediately after a user opens the app and after each answered question, so users can quickly explore the semantic model without manually typing every query. The team also wants to minimize application-side prompt engineering and use a capability designed specifically for this purpose. Which approach should they use?
- A
Call the Cortex Analyst Suggested Questions endpoint for the semantic model and display the returned prompts as clickable follow-up questions in the application
- B
Use AI_COMPLETE to generate a list of generic retail questions from the model's YAML file and treat those as official suggested questions for Cortex Analyst
- C
Query the semantic model with a SELECT statement to retrieve suggested questions stored automatically by Snowflake in hidden metadata tables
- D
Use a search service over prior chat logs to return the most frequent user questions, because Cortex Analyst does not provide a built-in suggested questions capability
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use the Cortex Analyst Suggested Questions capability directly. In Snowflake Cortex Analyst, Suggested Questions is designed to help applications present relevant starter prompts and follow-up prompts grounded in the semantic model, improving discoverability and user adoption. This is preferable to building a custom workaround with a general-purpose LLM function such as AI_COMPLETE, because the dedicated capability is aligned to the Analyst experience and reduces custom orchestration. It is also incorrect to assume suggested questions are stored in hidden SQL-accessible metadata tables or that developers must approximate the feature by mining historical chat logs. Best practice is to use the Analyst feature built for semantic-model-guided question suggestions and then present the returned prompts in the application UI as selectable questions.
- A. Correct.
Correct. Cortex Analyst includes a Suggested Questions capability intended to generate relevant starter or follow-up questions based on the semantic model. This is the most appropriate approach when the goal is to surface clickable prompts in an Analyst-powered application while avoiding unnecessary custom prompt engineering.
- B. Incorrect.
Incorrect. AI_COMPLETE can generate text, but using it as a substitute for the dedicated Suggested Questions capability is not the best fit here. It would require custom prompting, may produce less consistent results for this use case, and does not use the purpose-built Analyst workflow intended for suggested questions.
- C. Incorrect.
Incorrect. Suggested questions are not retrieved by directly querying hidden metadata tables with SQL. This reflects a common misconception that every AI feature in Snowflake exposes its generated artifacts as queryable system tables. Suggested Questions is accessed through the appropriate Cortex Analyst interface, not by selecting from hidden storage.
- D. Incorrect.
Incorrect. Prior chat-log mining could be a custom enhancement, but it is not the best answer because the scenario explicitly asks for a built-in capability designed for this purpose. Cortex Analyst does provide a Suggested Questions feature, so claiming it does not exist is factually wrong.