Power Platform Community Forum Thread Details

Dear all,

I have a problem with the sudden appearance of [REDACTED] messages in our ConversationTranscripts dataverse table.

To give you context: We have rolled out our internal chatbot through Copilot Studio for roughly a month by now. It is available in Sharepoint, as well as MS Teams and the M365 Copilot App.

Since the beginning we have been analyzing the conversation by using the aforementioned dataverse table. So we have one month of data more or less.

Now since April we are seeing [REDACTED] messages from Copilot Chatbot in the transcripts. It appears that those messages were appended by the bot, but also appear in the middle of the questions. Additionally, it does not look like that the bot failed to answer or that an appropriate answers was redacted. The opposite is rather true, the question was answered successfully and the [REDACTED] message slid in afterwards.

Example of previous conversation:

User: Asks Questions

Bot: Gives Answers, with or without Citations, depending on the questions

Conversations now:

User: Asks Questions
Bot: [REDACTED]

Bot: Gives Answers, with or without Citations, depending on the questions

User: Asks Questions
Bot: Gives Answers, with or without Citations, depending on the questions
Bot: [REDACTED]

Does anybody know what causes this? I looked into the news for March but I couldn't spot an update that could cause this.

Any help is highly appreciated

Categories:

Bot administration

Bot analytics

Hi @TF-07040736-0

Since March/April, Microsoft introduced enhanced safety and compliance processing in Copilot Studio. As part of this, some internal or filtered messages (e.g. safety checks, hidden system prompts, or potentially sensitive content) are now stored in the ConversationTranscripts table as [REDACTED].

These entries:

Do not mean the bot failed

Are not visible to end users

Are added at the logging level (Dataverse only)

That’s why you see correct answers, but [REDACTED] appears before or after them.

This behavior cannot be disabled, so the recommended approach is to handle it in your reporting layer:

Filter out [REDACTED] messages in Dataverse queries or Power BI

Exclude them from analytics and KPIs

(Optional) implement custom logging if you need fully clean transcript

This is expected behavior after recent platform updates, not a bug. The solution is simply to exclude [REDACTED] entries from your analysis.

------------------------------------------------------------

If this solved your issue, please mark it as ✅ Accepted Answer.
If it helped, feel free to give it a 🩷 Like!

Best regards,
Piotr