Current Behavior
When Ask or AI Draft retrieves multiple sources, one Q&A pair marked Verbatim causes all retrieved sources to be returned verbatim. This includes non-Verbatim Q&A pairs and documents—even though documents do not have a Verbatim setting. As a result, document content may be copied exactly as stored, including tables of contents, headings, duplicated information, and unrelated text.
Product confirmed this is the current behavior.
Client Use Case
Blue Yonder needs certain approved Q&A responses to remain unchanged while allowing Responsive AI to synthesize supporting information from other Q&A pairs and documents.
Currently, marking one response Verbatim can turn the entire AI-generated answer into a collection of copied source content instead of a concise, cohesive response.
Requested Enhancement
Apply Verbatim only to the individual Q&A pair on which it is enabled:
- Preserve the Verbatim Q&A response exactly as approved.
- Allow non-Verbatim Q&A pairs and documents to be synthesized normally.
- Use only relevant excerpts from documents and exclude tables of contents or unrelated document structure.
- Clearly identify which portion of the generated answer came from a Verbatim source.
Business Impact
This behavior affects more than Blue Yonder. It can impact any client using Ask or AI Draft with mixed knowledge sources and at least one Verbatim Q&A pair.
The current behavior:
- Overrides the administrator’s source-level intent.
- Produces lengthy, duplicated, or irrelevant answers.
- Creates accuracy, legal, compliance, and content-governance risks.
- Requires substantial manual cleanup and review.
- Makes results unpredictable because users do not know which sources will be retrieved.
- May cause clients to remove necessary Verbatim controls or stop using documents as AI sources.
- Reduces trust in and adoption of Responsive AI.
Clients should not have to choose between protecting approved wording and receiving a usable AI-generated response. Verbatim should remain a source-level instruction and should not change how every other retrieved source is processed.
