PowerLeap Logo

Copilot Studio Friction

Get notified when this record changes

One email when the status or the fix changes — double opt-in, no tracking, unsubscribe in every email.

Question repeats endlessly even after a valid answer

Mitigatedsince 8 July 2026

Last verified

Details & related

Assessment

Confidence
CorroboratedMultiple independent sources describe the same behaviour.
Severity
Blocking
Typical time lost
Hours

Identification

Topics & variablesAll channels

Verification & changes

  1. Verified

    Doc check by human: Q&A thread and slot-filling documentation re-read during seed migration; the limitation remains undocumented in official docs.

  2. Change

    Provisionally approved by the Product Owner; external LLM quality review pending.

  3. Change

    Initial record created from the seed dossier (migration wave 1).

Related

Are you in the right place?

  • A question node in a topic asks the user for a value.
  • The user answers with a valid, complete value.
  • The agent asks the identical question again — sometimes several times in a row.
  • The topic never advances past the question node.
  • Your agent runs with generative orchestration enabled.

If the agent switches to a different topic mid-question → see Agent jumps out of the topic mid-question. If the wrong topic fires after an "End all topics" node → see Wrong topic fires after "End all topics". If the captured value has the wrong type → a silent type change after a toggle (record planned).

What's happening

With generative orchestration enabled, two mechanisms compete for the user's answer. The question node expects the answer for its slot filling — filling the variable behind the question. The orchestration planner reads the same message and re-plans the conversation. When the planner wins, the topic re-asks its question instead of accepting the value. Microsoft support calls this expected behavior: with generative AI active, the system may re-ask for details. With orchestration disabled, the identical question and input work correctly. Picture two receptionists sharing one desk: you answer one, and the other asks again. This limitation is not documented in the official docs; makers learn it from support answers.

For technicians

The Q&A thread shows the loop reproducibly across multiple questions and topics. A second maker confirmed the identical loop after deleting and recreating the topic. The Microsoft answer states the system prioritizes conversational flow under generative AI and may re-prompt. Disabling generative orchestration restores strict slot filling on question nodes. The workaround therefore trades orchestration benefits for reliable guided dialogs.

How to fix it

Solution 1

Community workaround

Switch the agent to classic orchestration

  1. Open your agent in Copilot Studio.
  2. Select "Settings".
  3. Select "Generative AI".
  4. Under the orchestration option, select the classic mode instead of the generative one.
  5. Save the settings.
  6. Retest the question path in the test pane.

✅ You should now see: the question accepts a valid answer on the first try and moves on.

  1. Weigh the trade-off: classic mode removes generative topic selection for the whole agent.

Check that it worked

In the test pane, trigger the topic and answer the question with a valid value. Expected: the agent accepts the answer once and continues to the next node. Repeat the test in your published channel after publishing.

If it didn't work

  • The channel still runs the old orchestration setting. Publish and propagation delay is suspect number one. See Teams answers with an old version of your agent.
  • An entity mismatch rejects the answer. If the answer never matches the question's entity, the node re-asks by design. Check the entity and its options.
  • A trigger steals the message. Another topic's trigger phrases can match the answer text and interrupt the question. Review overlaps between triggers and expected answers.

Prevent it next time

  • Build critical guided dialogs under classic orchestration, or test them under generative orchestration before you commit.
  • Test every question node with realistic answers early, in the mode your agent will ship with.
  • Retest question-heavy topics after platform updates; orchestration behavior is a known drift area.

Evidence

  • Community threadlearn.microsoft.com

    Multiple makers report question nodes re-asking after valid answers under generative orchestration; a Microsoft response calls it expected behavior.

  • Official documentationlearn.microsoft.com

    The entities documentation describes slot filling as the mechanism that fills a question node's variable directly from the user's answer.