Answered

Agent sending multiple answers to one prompt

(4) Share

Report

Posted on by VJ-22041215-0

To a prompt to receive a summary of meetings / events scheduled for today, the agent sends multiple answers that differ from each other as it tries to load the events many times. Each of the responses is quite different, there are missing meetings and instead it includes some meetings scheduled for next week. I think the reason is that it uses the tool to get events multiple times. Do you have any tips for how to limit this?

Categories:

Calling actions from Copilot Studio

I have the same question (0)

All responses (5)

Answers (2)

Sort by

Nanit Khanna 56 on at

Like (1)

Report
Copy link

Link copied!
Hi @VJ-22041215-0,

This usually happens when the agent repeatedly calls the Outlook connector/tool during a single conversation turn. Each tool execution may return slightly different results depending on timing, pagination, time zone interpretation, or incomplete filtering.

From your screenshot, the agent is calling:
• Outlook.Read.Events multiple times
• Re-querying calendar events instead of reusing previous results

This can cause:
• Duplicate or inconsistent answers
• Missing meetings
• Meetings from incorrect dates
• Different summaries in each response

Things you can do to reduce this:

• Add stricter instructions in the agent prompt like this:

“Call Outlook calendar tools only once per request”

“Use the first returned event list and do not re-query”

“Only summarize events for today”

• Add explicit date filtering:

Use Today() boundaries

Specify time zone clearly

Limit results to current date only

• Reduce agent autonomy/tool reasoning:

If using generative orchestration, try limiting unnecessary tool retries

Use deterministic topics/actions where possible

• Return structured event data first:

Fetch events once

Store results in variables

Generate summary from stored data instead of calling the connector again

• Add max-result limits:

Example: top 20 events for today only

• Verify timezone configuration:

Incorrect timezone handling often causes next-week events to appear

• Test with a simpler prompt:

“List today’s meetings only”

Then add summarization later

This behavior is fairly common with Copilot Studio agents using generative orchestration and Outlook tools, especially when the model decides to retry tool calls to improve confidence.

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

Please ✅ Does this answer your question if my post helped you solve your issue. This will help others find it more readily. It also closes the item.

If the content was useful in other ways, please consider answering Yes to Was this reply helpful? or give it a Like! 🩷

Was this reply helpful? Yes No
Verified answer

deepakmehta13a 355 on at

Like (1)

Report
Copy link

Link copied!

Hi @VJ-22041215-0,

The issue is likely happening because the agent is calling Outlook.Read.Events multiple times for a single prompt.

You can try these improvements:
• Add instruction: “Call the Outlook tool only once per request”
• Store the first response in variables and reuse it
• Add strict filtering for today’s date only
• Verify timezone settings
• Reduce generative orchestration/retries if enabled

Multiple connector calls can return inconsistent meeting lists, which is why summaries differ.

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

If this helps resolve your issue, please consider marking the response as Verified so it can help others facing a similar scenario.

If you found this helpful, you can also click “Yes” on “Was this reply helpful?” or give it a Like.

Was this reply helpful? Yes No
Suggested answer

sumit_artesian 261 on at

Like (1)

Report
Copy link

Link copied!
Hi @VJ-22041215-0,

What you’re seeing is usually caused by the orchestrator calling the calendar tool multiple times while trying to “improve” the answer, which can lead to inconsistent results, duplicate responses, or meetings outside the intended timeframe.

I would personally create a dedicated topic for the daily summary and call the tool explicitly from there instead of letting generative orchestration decide when to invoke it. Restricting the tool so it only runs from within the topic should also help stabilize the behavior.

Basically, you need to:

Create a new topic and give it instructions on when to trigger.

Optionally make the topic accept input parameters that it will recognize on its own from the user's prompt.

Call the tool to perform operations.

After that, navigate to your Tools tab and configure your tool.

You will see an option like, "Decide when this tool will run" or something else on those lines. There you can configure the tool to only ever run from within a topic.

That way, your tool will only ever run once per prompt.

It’s also worth making the tool output as strict as possible before the model formats the response. For example, explicitly filtering for only today’s events and sorting them before returning them to the agent.

Please ✅ Does this answer your question if my post helped you solve your issue. This will help others find it more readily. It also closes the item.

If the content was useful in other ways, please consider answering Yes to Was this reply helpful? or give it a Like! 🩷

Was this reply helpful? Yes No
Suggested answer

rezarizvii 352 on at

Like (1)

Report
Copy link

Link copied!

Hi @VJ-22041215-0, hope you are doing well.

I would probably avoid relying entirely on generative orchestration for this scenario and instead move the logic into a dedicated topic.

You can create a topic specifically for meeting summaries and define clear trigger instructions so the agent knows exactly when to use it. If needed, you can also add input variables so the topic can intelligently pick up things like “today”, “tomorrow”, or “this week” directly from the user’s message.

Inside that topic, call your calendar/events tool directly rather than letting the orchestrator decide when to invoke it. Then, in the Tools configuration, set the tool so it only runs when explicitly called from a topic. That usually prevents the repeated tool invocations and inconsistent responses.

Please ✅ Does this answer your question if my post helped you solve your issue. This will help others find it more readily. It also closes the item.

If the content was useful in other ways, please consider answering Yes to Was this reply helpful? or give it a Like! 🩷

_______________________________

Reza M. Rizvi
LinkedIn | Email

Was this reply helpful? Yes No
Verified answer

Valantis 5,268 on at

Like (1)

Report
Copy link

Link copied!

Hi @VJ-22041215-0,

The core fix is moving from generative orchestration deciding when to call the tool to a dedicated topic that calls it exactly once.

The specific setting that locks this down: in your tool configuration, find the option that controls when the tool runs and set it to only run when called from a topic. Once that's set, the orchestrator can no longer invoke it autonomously during reasoning, which stops the multiple calls entirely.

In the topic itself, call the tool once, store the output in a variable, and build the summary from that variable. That way even if the model wants to retry, the data comes from the stored variable not a new tool call.

For the date filtering specifically, pass today's date explicitly as a parameter to the tool call rather than relying on the model to interpret 'today' correctly. That also fixes the timezone drift causing next-week events to appear.

Best regards,

Valantis

✅ If this helped solve your issue, please Accept as Solution so others can find it quickly.

❤️ If it didn’t fully solve it but was still useful, please click “Yes” on “Was this reply helpful?” or leave a Like :).

🏷️ For follow-ups @Valantis.

📝 https://valantisond365.com/

💼 LinkedIn

▶️ YouTube

Was this reply helpful? Yes No