Why is my Copilot agent returning explanation_of_tool_call JSON instead of topic responses?

Question

Why is my Copilot agent returning explanation_of_tool_call JSON instead of topic responses?

Anonymous

Hi everyone,

The agent was working fine this morning. However, when I started exploring topic outputs in a structured format—for example, scenarios like meeting prep and company overview—the responses stopped returning the expected topic text.

Instead of the topic response, I’m seeing outputs like:

{"explanation_of_tool_call":"This action needs to be done to acknowledge the user's greeting and initiate the conversation in a polite and professional manner."}

Here’s my setup:

OnRecognizedIntent triggers for greetings.

SetProperty nodes to store topic responses in dialog.topicResponse.

SendActivity nodes referencing these stored responses.

Optional tool/function nodes that return internal metadata like explanation_of_tool_call.

My questions for the community:

Has anyone seen fallback responses appear like this after trying to structure topic outputs in Copilot Studio?

What’s the best way to map a tool or lookup output to a user-facing property without sending internal metadata?

Are there known issues with multi-line expressions or YAML parsing in SetProperty that can cause fallbacks?

I’d appreciate any guidance, examples, or best practices for structuring topic outputs without triggering these fallback responses.

Thanks, Kabilan

Karan Shewale 2,585 Reputation points Microsoft External Staff

2025-10-10T04:40:44.92+00:00

Thanks for bringing this issue to our attention. We will check and update you soon.
Patrick Herrington 0 Reputation points

2026-01-13T15:38:14.3633333+00:00
This seems like a bug in CopilotStudio. Shouldn't the Agentic Orchestrator output whatever the sub-agent(s) output? I don't mind the intent in testing but not in production. It only shows in production, not in test though.

I would vote that it behaves the same in the CopilotStudio test as in a production scenario. I would also vote that this functionality be a toggle, or removed completely.

If you want to see what agent your orchestration agent is using this works just fine, and isn't json

Transparency: Always inform the user which agent you are consulting before or while providing the answer. For example, "I'll ask the Compliance Agent about that."

2 answers

Your answer

Karan Shewale 2,585 Reputation points Microsoft External Staff

2025-10-10T04:40:44.92+00:00

Thanks for bringing this issue to our attention. We will check and update you soon.
Patrick Herrington 0 Reputation points

2026-01-13T15:38:14.3633333+00:00

This seems like a bug in CopilotStudio. Shouldn't the Agentic Orchestrator output whatever the sub-agent(s) output? I don't mind the intent in testing but not in production. It only shows in production, not in test though.

I would vote that it behaves the same in the CopilotStudio test as in a production scenario. I would also vote that this functionality be a toggle, or removed completely.

If you want to see what agent your orchestration agent is using this works just fine, and isn't json

Transparency: Always inform the user which agent you are consulting before or while providing the answer. For example, "I'll ask the Compliance Agent about that."

Answer 1

Karan Shewale 2,585 Microsoft External Staff

Hi Kabilan Rajendiran,

The “explanation_of_tool_call” output appears when Copilot Studio accidentally returns internal reasoning metadata instead of the intended topic response. This usually happens when the tool or function node returns its raw output object instead of the actual text meant for the user. To fix this, ensure that your topic stores only the readable text (for example, dialog.topicResponse = toolResult.responseText) and that your SendActivity node references this text variable directly, not the entire object. Also, double-check your YAML or structured authoring syntax — indentation or variable mapping issues can cause the system to fall back to this JSON response. This behavior started after the October 2025 Copilot Studio update, which changed how structured tool calls are parsed.

For more details, refer to:

Use variables and expressions in Copilot Studio

Create and manage topics in Copilot Studio

Use actions and connectors safely

Thank you.

Karan Shewale.

If this response resolves your issue, please accept the answer and, if helpful, click the “Upvote” button. Your feedback helps us improve and assist others more effectively.

David Farrell 0 Reputation points

2025-10-14T07:11:53.9833333+00:00

Having a variant on this but ultimately the same problem.

Our situation:

GPT-5 powered agent with an MCP tool (added via url to the MCP server).
MCP server exposes two tools.

There are no specific flows or anything - it's just a Copilot Agent with one MCP added.

When user asks their question in Copilot Studio (in the testing box), Agent fires off the tool, interprets the result and communicates effectively.

When I publish and access the Agent via Copilot or via Teams the agent shows the justification instead of executing the call. If I say 'go ahead then' it does and continues for rest of conversation.

User: Please show me all xxxxxx

Agent:
{"explanation_of_tool_call":"The user has requested to see all xxxxxx, so a search filtered by type 'xxxxxx' will be performed to retrieve up to 50 relevant results.","type":"xxxxxx","max_results":50}

User: looks good go ahead

Agent: (At this point the agent works correctly for rest of convo in Copilot)
Karan Shewale 2,585 Reputation points Microsoft External Staff

2025-10-15T08:49:43.7466667+00:00
Hi David,

The explanation_of_tool_call JSON appears in published Copilot or Teams because the agent first shows its reasoning before executing a tool. In the Studio sandbox, it works correctly. To fix this, map the tool output to a user-facing variable (e.g., dialog.topicResponse = toolResult.responseText) and ensure your SendActivity node uses that variable. If supported, enable auto-execute for the MCP tool to remove the confirmation step. This will make the agent return the intended response directly.

References:

Use connectors in Copilot Studio

Work with variables

Topics in Copilot Studio

Thank you.

Karan Shewale.

If this response resolves your issue, please Accept the answer and, if helpful, click the “Upvote” button. Your feedback helps us improve and assist others more effectively.
Even lauvsnes 20 Reputation points

2025-10-23T09:59:28.5233333+00:00

I have a similar problem like this, where the agent outputs the "explanation_of_tool_call" reasoning for each time it uses its connected agent. However, I do not use any custom topics to handle the connected agents response, and am therefore not able to access the something.responseText. What do you suggest to do?
Varun Patkar 0 Reputation points

2025-11-14T03:28:32.47+00:00

I have a similar issue on this where the agent shows that on saying Thanks. Is this still a bug and being tracked? This is for the default Thanks topic after another topic has been executed. This is present in both the Teams published agent and the Test
Karan Shewale 2,585 Reputation points Microsoft External Staff

2025-11-14T04:45:01.6166667+00:00

Hi Varun,

The agent shows explanation_of_tool_call in the default “Thanks” topic because the previous topic is passing a full tool-output object into the next step instead of a clean text value. When the “Thanks” topic runs, Copilot surfaces that leftover object as internal reasoning JSON. This is a known product issue, but not publicly visible. The mitigation is to ensure each topic stores only the user-facing text (for example, dialog.topicResponse = toolResult.responseText) so the default “Thanks” topic does not inherit tool metadata.
Candi Carrera 0 Reputation points

2025-12-07T07:10:49.45+00:00

I have also the same issue. any update on the bug resolution timeline please since Nov 14th, 2025 ?
Jose Valdivia Leon 10 Reputation points

2025-12-11T11:23:32.8033333+00:00

I have a similar situation. My agent uses Generative Orchestration, it triggers correctly a custom topic which essentially uses a Generative Answer node to get the response from the KB. I got the correct answer, but after the answer I get also the JSON:

{"explanation_of_tool_call":"The user’s question has been fully addressed with the available information, and no further action is required.","}

Where should I configure dialog.topicResponse = toolResult.responseText in the Topic?
Matteo Guarneri 5 Reputation points

2026-02-04T15:52:52.9433333+00:00

Siamo a gennaio e il problema persiste. Potremmo avere indicazioni su Bug ID CSUP-10877.?

Mohammad Z Khan 0

I have tried all the suggestion by my relatively simple topic using Generative Answers from a a few PDF is still leaking the

explanation_of_tool_call

. I have pasted the YAML if that helps, I cannot see why its still showing despite following your suggested approach of explicitly assigning the text response generated by the Generative Answer step into the SendActivity. Would appreciate any insights on this.

kind: AdaptiveDialog
modelDescription: |-
  This Topic must be used as the entry point for all questions users have related to the ui. 

  where do I find call history
  how do I submit feedback in CRM
  how do I use the timeline
   how do I log a phone call
  how do I use the Sales Hub
  where is the command bar
  how do I create a new contact
  how do I use quick create
  how do I switch between apps in Dynamics
  how do I find recent records
  how do I navigate the site map
  how do I use dashboards in Dynamics
  how do I track activities in Dynamics
  how do I use the business process flow
  how do I find my recent activities
  how do I use the Sales app interface
  help me with Dynamics 365 Sales UI
  how do I use the command bar in a record
  how do I find related records
beginDialog:
  kind: OnRecognizedIntent
  id: main
  intent: {}
  actions:
    - kind: SearchAndSummarizeContent
      id: XNGq8u
      displayName: Draft generative answers
      autoSend: true
      variable: Topic.DraftUserGuidanceResponse
      userInput: =System.LastMessage.Text
      additionalInstructions:
      fileSearchDataSource:
        searchFilesMode:
          kind: SearchSpecificFiles
          files:
            - msdyn_salesCopilot.file.-UserGuidesPhase1.pdf_bwA
            - msdyn_salesCopilot.file._crm_user_guide_phase_1a_20260218_085653.pdf_Rym
            - msdyn_salesCopilot.file.dynamics_365_sales_ui_guide_20260218_093709.pdf_dCd

      knowledgeSources:
        kind: SearchSpecificKnowledgeSources

      responseCaptureType: FullResponse

    - kind: SendActivity
      id: sendActivity_DmhP3F
      activity:
        value: =Topic.DraftUserGuidanceResponse.Text.Content
        attachments:
          - kind: AdaptiveCardTemplate
            cardContent: |-
              {
                "type": "AdaptiveCard",
                "$schema": "https://adaptivecards.io/schemas/adaptive-card.json",
                "version": "1.5",
                "body": [
                  {
                    "type": "RichTextBlock",
                    "inlines": [
                      {
                        "type": "TextRun",
                        "text": "Is there any other questions I can help answer for you? ",
                        "color": "Good",
                        "weight": "Bolder",
                        "size": "Medium"
                      }
                    ],
                    "horizontalAlignment": "Left",
                    "spacing": "ExtraSmall"
                  },
                  {
                    "type": "Table",
                    "columns": [
                      {
                        "width": 1
                      },
                      {
                        "width": 1
                      },
                      {
                        "width": 1
                      }
                    ],
                    "rows": [
                      {
                        "type": "TableRow",
                        "cells": [
                          {
                            "type": "TableCell"
                          },
                          {
                            "type": "TableCell"
                          },
                          {
                            "type": "TableCell"
                          }
                        ]
                      },
                      {
                        "type": "TableRow",
                        "cells": [
                          {
                            "type": "TableCell",
                            "items": [
                              {
                                "type": "Image",
                                "url": "",
                                "backgroundColor": "#FFFFFF",
                                "size": "Medium",
                                "spacing": "Small",
                                "horizontalAlignment": "Left",
                                "verticalContentAlignment": "Bottom"
                              }
                            ]
                          },
                          {
                            "type": "TableCell",
                            "items": [
                              {
                                "type": "RichTextBlock",
                                "inlines": [
                                  {
                                    "type": "TextRun",
                                    "text": "User Guidance Agent",
                                    "color": "Attention",
                                    "weight": "Bolder",
                                    "size": "Small"
                                  }
                                ]
                              }
                            ]
                          },
                          {
                            "type": "TableCell",
                            "items": [
                              {
                                "type": "Image",
                                "url": "",
                                "size": "Medium",
                                "horizontalAlignment": "Right",
                                "verticalContentAlignment": "Bottom"
                              }
                            ]
                          }
                        ]
                      },
                      {
                        "type": "TableRow",
                        "cells": [
                          {
                            "type": "TableCell"
                          },
                          {
                            "type": "TableCell"
                          },
                          {
                            "type": "TableCell"
                          }
                        ]
                      }
                    ],
                    "spacing": "ExtraSmall",
                    "showGridLines": false
                  }
                ]
              }

    - kind: EndDialog
      id: 7zHrgR

inputType: {}
outputType: {}

Answer 2

Deleted

This answer has been deleted due to a violation of our Code of Conduct. The answer was manually reported or identified through automated detection before action was taken. Please refer to our Code of Conduct for more information.

Comments have been turned off. Learn more