Passing Record Output Between Generative Answers Actions in Copilot Studio

Generative Answers actions and topics in Copilot Studio only accept specific input types (such as text/string) and expose specific output types. When an action returns a record/table and the next action expects a string, the record can't be passed directly; a string value must be extracted or defined as an output.

Use the following approach:

1. Define topic outputs as strings
   • When designing topics that will feed other actions, define explicit output variables of type text/string that contain exactly what the next action needs.
   • For example, if a Generative Answers action returns a record with multiple fields, add a topic output like SummaryText and map it to the specific field or composed text that should be passed on.
   • In generative orchestration, the orchestrator then uses these outputs as structured data to build the next step or the final answer, instead of passing the entire record.
2. Avoid passing citations/records as tool inputs
   • Citations and other structured outputs from knowledge sources are exposed as tables/records (for example, Answer.Speech.Citations in voice scenarios) and are meant for display or branching, not as inputs to other tools.
   • Current limits explicitly state that citations returned from a knowledge source can't be used as inputs to other tools or actions. Instead, extract or summarize the needed information into a text output variable and pass that.
3. Use topic input/output parameters for orchestration
   • Configure topic inputs as text parameters with clear names (for example, questionText, contextSummary) so the orchestrator can auto-fill them from conversation context or ask the user if missing.
   • Configure topic outputs as text parameters that represent the final data points needed by downstream topics or actions (for example, NearestStoreLocation, AnswerSummary).
   • The orchestrator then passes these string outputs between topics and tools in the generated plan; records remain internal to the topic/tool and are not passed directly.
4. Pattern to handle record → string
   In practice, this means:
   • Use the Generative Answers action to query knowledge sources and get a record/table internally.
   • Within the same topic, compose the desired text (for example, a summary, a specific field, or a formatted string) and assign it to a text output variable.
   • Configure the next Generative Answers action (or topic) to take that text output variable as its string input.

This pattern aligns with how topic inputs/outputs and generative orchestration are designed: structured outputs (records/tables) stay inside the topic/tool, while explicitly defined text outputs are used to connect steps.

References:

• Use generative answers in a topic
• Use generative answers in basic voice agents
• Manage topic inputs and outputs
• Apply generative orchestration capabilities
• Configure real-time voice agents
• Work with variables