Chat Prompts | Yext Hitchhikers Platform

When a user submits a message, Chat runs through a series of prompts to produce a relevant response. Prompts tell the bot what to do and how to format its response.

The series of prompts that a bot runs through for a user’s message will vary depending on the goal triggered. You can review which prompts are utilized for a particular message within the Response Logs of your bot.

• summarizeConversation: This prompt is designed for generating a concise summary of an ongoing conversation between a bot and a human. It focuses on creating a summarized version, with more details on the later parts of the conversation and a more abstract representation of the earlier parts. If there is an outdated summary available, it needs to be updated to incorporate new messages, ensuring that important information from the outdated summary persists.
• suggestAlternatives: This prompt is used to write a terminating message in a conversation where a clear answer was not found in the provided data. The bot needs to acknowledge the user’s question, clarify the lack of a definitive answer, and then suggest alternative or helpful options based on the data. The response should be concise, use markdown formatting, and invite the user to ask more questions or refer them to additional resources if a URL is provided.
• selectOptions: This prompt assists in generating a list of options or steps for the user based on a specific context or goal. It’s meant to guide the user by presenting a structured list of choices or actions they can take, helping to direct their next steps in a clear and organized manner.
• provideDirectAnswer: This prompt is aimed at directly answering a user’s question based on the provided data. It involves crafting a response that directly addresses the user’s query, using the information available and ensuring the answer is clear and to the point.
• predictSearchQuery: This prompt is designed to generate a concise search query based on the latest message in an ongoing conversation. It guides the creation of a relevant query in a specified language, focusing on capturing the essence of the user’s request without including sensitive or irrelevant information.
• predictGoal: This prompt, combined with its system message counterpart, is used to predict the goal a user is trying to achieve in their conversation with a bot. It involves analyzing the user’s last message, the provided list of goals with descriptions and example messages, and the context of the conversation to identify the user’s intent and align it with a specific goal ID. The reasoning for the selected goal ID must be clearly articulated.
• parseSelectedData: This task involves analyzing a chat history and a set of potential answers to determine if these answers have been presented before and which one the user might have chosen. The focus is on the current set of options, disregarding any past unrelated selections. The analysis requires identifying if the user’s choice aligns with one of the current options based on the messages.
• parseFields: Combined with its system message counterpart, this prompt is about parsing and extracting specific pieces of information from provided data or text. The task involves identifying relevant fields or data points within a larger set of information, ensuring accurate and context-sensitive extraction.
• parseConfirmation: This prompt centers on identifying confirmations or affirmations within a conversation. The task involves analyzing the conversation to detect instances where a user confirms or agrees with something, interpreting the context to accurately assess confirmation.
• noGoal: The prompt focuses on situations where a user’s intent or goal is not aligned with predefined objectives or is unclear. The task involves analyzing the conversation to determine that the user’s actions or queries do not match any set goals, indicating a lack of a specific intent.
• noData: This prompt is used when there is an absence of data or information in response to a user’s query. The task involves acknowledging the lack of data and effectively communicating this to the user, providing clarity and managing expectations regarding information availability.
• followupQuestion: This prompt is about crafting a follow-up question in a conversation when additional clarification is needed. It involves using the context of the conversation and any available analysis to ask targeted, relevant questions that help gather more specific information or clarify the user’s needs.
• evaluateCondition: This task involves evaluating the truthfulness of a condition based on a conversation and any provided data. The evaluation must consider the context, any relevant data, and the current date if time-dependent, leading to a reasoned analysis and a clear true/false conclusion.
• detectResultContent: This prompt requires evaluating a conversation and accompanying data to determine if they provide a clear answer, require further clarification, or are irrelevant to the user’s question. The task involves assessing the relevance and completeness of the information in relation to the user’s query, considering factors like the user’s location if specified. The outcome should include a reasoned analysis and a recommendation for the next steps.
• confirmCollectedData: This prompt focuses on confirming the accuracy and relevance of data collected during a conversation or interaction. It involves validating the collected data against the user’s inputs or requirements, ensuring that the information aligns with the user’s needs or queries.
• confirmChoice: Combined with its system message counterpart, this prompt is about confirming a choice or decision made by the user or suggesting within a conversation. The task requires understanding the context, the options available, and the user’s response to affirm or clarify the chosen option.
• collectFields: This prompt is designed to extract and organize specific pieces of information from a conversation or provided data. It involves identifying and collating relevant data points or fields, structuring them in a way that’s accessible and useful for further processing or decision-making.
• casualResponse: Along with its system message counterpart, this prompt is for crafting casual, engaging responses in a conversation. The response should be helpful, professional, and polite, tailored to the user’s last message and the provided context or instructions. The aim is to maintain a friendly and conversational tone while addressing the user’s needs or queries.
• askForSelection: This prompt is designed to ask the user to make a selection from a set of options that the bot has fetched. The bot needs to present these options clearly and concisely, encouraging the user to indicate their choice. The prompt includes additional instructions on how to format and present the options to the user effectively.
• answerQuestion: Combined with its system message counterpart, this prompt is for responding to a user’s question using information from search results. The bot must use the provided search results to construct an accurate and informative response, explaining how the answer was derived from the search data. The response should be tailored to the user’s language if specified, and it should avoid restating the question or including irrelevant or fabricated information.