Copilot Studio bot behavior changed without topic updates

(1) Share

Report

Posted on by SK-23051105-0

Hi everyone,

I’m facing unexpected behavior in Copilot Studio and wanted to check if others have experienced something similar.

I built several scenarios/topics in a bot that were working correctly. After spending about a month developing new, separate scenarios, I tested the older ones again. I did not change any configuration, global variables, trigger phrases, or sub-agent assignments for the old topics.

However, now I’m seeing the following issues:

The bot sometimes jumps between topics
Incorrect workflows are being triggered
In some cases, multiple responses are returned
Generative AI nodes behave differently than before (output style and flow behavior changed)

Important details:

New topics do not reuse variables
Trigger phrases/prompts are different
Old scenarios use different sub-agents than the new ones
Old scenarios previously worked consistently
Behavior changed without direct edits to those topics

Has anyone noticed similar behavior recently?

Could this be related to backend model updates, topic routing changes, or generative AI behavior updates in Copilot Studio?

Any insight or best practices to stabilize topic routing and generative nodes would be appreciated.

Categories:

Autonomous agents

Calling actions from Copilot Studio

Topic creation & management

I have the same question (2)

All responses (3)

Answers (0)

Sort by

MichaelFP 2,001 Moderator on at

Like
a
(1)

Report
Copy link

Link copied!

The model has changed? I saw this behaviour when change from GPT-4 to GPT-5.

Was this reply helpful? Yes No
RichAI 26 on at

Like
a
(1)

Report
Copy link

Link copied!

Hi @SK-23051105-0,

This normally happens with me when Copilot Studio gets updated. You can fix/troubleshoot this by using following best practices per my view:

1. Reduce dependency on generative nodes for control flow, and using them only for language generation
2. Add trigger phrases more distinct and explicit
3. Add validations and guardrails in topics
4. Last and most important, introduce evals and regression-style tests to detect drift early

Hope this helps.

Thank You.

Was this reply helpful? Yes No
Suggested answer

Valantis 6,735 on at

Like
a
(1)

Report
Copy link

Link copied!

Hi @SK-23051105-0,

What changed on the platform: Between October 27-31, 2025, Microsoft retired GPT-4o as the default orchestration model in Copilot Studio and replaced it with GPT-4.1. This is confirmed from Microsoft docs. GPT-4.1 has different routing behavior, topic selection logic, and generative node output characteristics compared to GPT-4o. This is almost certainly what changed your behavior without any edits to your topics.

Since your agent was built on GPT-4o behavior and GPT-4.1 is now the default, the orchestration model interprets trigger phrases and selects topics differently. This explains the topic jumping, incorrect workflow triggering, and changed generative node output.

To check which model your agent is using: go to Settings > Generative AI > Orchestration. You can see the current primary model and change it if needed.

If you want to temporarily validate this is the cause: switch back to GPT-4o (it remains available by turning on "Continue using retired models" option), test your old scenarios, and confirm if they work correctly. This will tell you definitively whether the model change caused the regression.

For stabilizing going forward, the most impactful thing from Microsoft docs: the primary factor in topic selection with generative orchestration is the topic description, not just the trigger phrases. Each topic should have a clear, specific description that explains when it should be used and what it does. If topics have no description or vague descriptions, GPT-4.1 is more likely to route incorrectly than GPT-4o was.

Also confirm whether your agent uses generative orchestration (Settings > Generative AI > Orchestration > Yes). If you switch to classic orchestration, topic selection is deterministic based on trigger phrase matching and the model change has no effect on routing.

Ref: https://learn.microsoft.com/en-us/microsoft-copilot-studio/whats-new

Ref: https://learn.microsoft.com/en-us/microsoft-copilot-studio/authoring-select-agent-model

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