The AI assistant uses Retrieval-Augmented Generation (RAG). Rather than relying primarily on general outside knowledge, it first searches the text analysis project for information relevant to the question.
How grounding works
When you ask the AI assistant a question, the process has two main parts:
Retrieve relevant information
The AI assistant searches the text analysis project for records related to your question. This retrieval step searches the project data and does not generate the answer.Generate an answer from that context
The relevant information is then provided as context for the language model, which generates its response based on that retrieved text analysis project data. This approach is designed to keep the responses anchored in the information available within your text analysis project.
Why grounding matters
General-purpose AI tools may draw on broad knowledge when generating responses. The AI assistant in Journey Insights Hub is designed to focus its question-and-answer experience on the dataset currently being analyzed.
This is especially useful when working with large feedback datasets, where you want the answer to reflect the experiences and information contained in the text analysis project rather than unrelated external information.
For example, if you ask:
"What is associated with negative feedback?"
The AI assistant searches the project records for relevant evidence and generates its response using that retrieved context.
What if the information is not in the text analysis project?
The AI assistant is intended to answer within the scope of the available project data. When a requested topic is not meaningfully represented, the response may indicate that there is insufficient information or that no relevant matches were found.
A lack of supporting data should not be treated as evidence that the topic could never be relevant. It means the topic was not meaningfully represented in the records available for the response.
Try:
Rephrasing the question
Using broader or related terminology
Confirming that the relevant records are included
Searching the project for related concepts