Copilot Returns ‘Content Filtered’ Without Showing Output: Fix

Official documentation checked October 14, 2026.

If Copilot displays “Content Filtered,” start with the message and the task it blocked. A safety refusal, an organizational data restriction, and a connection failure require different responses. The aim is to establish whether a legitimate request was mistakenly blocked and report that problem, rather than conceal the same request from the safeguard.

Why Copilot Shows ‘Content Filtered’ Instead of Output

Microsoft’s Copilot application card describes checks on prompts and generated responses. A harmful-content classification can stop an answer even if the wording you typed seems benign, because the supplied material or potential answer also matters. The visible message alone does not reveal which internal check fired or prove a particular classifier made a mistake.

Organizational data controls are separate. Microsoft Purview DLP for Copilot can restrict processing of specified labeled sources, block certain sensitive-information prompts, or restrict their use in external web searches, depending on the policy action. A “Confidential” label name does not by itself prove that a matching Copilot rule exists. Use the displayed policy notice and administrator evidence rather than assuming every restriction is harmful-content filtering.

Steps to Diagnose and Report the ‘Content Filtered’ Issue

1. Record the exact context. Note the full message, account type, host, task, time and time zone, and any policy tip or request ID. Record whether the request involved an uploaded file, a referenced work source, an agent, or only typed text. Preserve the original prompt in an approved location; keep sensitive material out of general support screenshots.
2. Read any explanation or access notice. If Copilot cites an organizational restriction, follow its access or policy-review route. If it explicitly refuses an unsafe task, respect that boundary. A missing answer does not justify lowering DLP, removing a sensitivity label, changing accounts, or copying protected material into a personal service.
3. Try a clearly benign, unrelated test. Start a new conversation and ask for something ordinary, such as a short meeting agenda. This helps distinguish a task-specific block from a broader service problem. A successful test does not prove which filter blocked the original. An unrelated test that also fails is evidence to give support, not proof of a tenant-wide policy.
4. Clarify a legitimate purpose without hiding the content. If your original request was ambiguous, describe the permitted task, audience, and desired safe output. For example: “Summarize this authorized training passage for a workplace safety briefing, without graphic detail or identifying personal information.” This changes the requested output and explains its scope. It is different from substituting code words, repeatedly resubmitting, or asking Copilot to ignore a safeguard.
5. Minimize the data appropriately. For a test, use invented names and a synthetic passage you are authorized to submit. Do not remove labels, warnings, or sensitive words from a restricted document just to get the same blocked content processed. If the real work requires protected data, ask the owner or governance team whether that use is permitted before proceeding.
6. Report a suspected false positive. Use response feedback when available, or the appropriate Microsoft support route when no feedback control is shown. Explain the legitimate purpose, expected safe response, and exact reproduction steps. Microsoft’s feedback guidance explains that organization policy can affect the available options and that you can review whether content samples are included. Inspect attachments and share only information you are permitted to disclose.

If Copilot Still Has Issues After the Main Checks

Have the administrator identify the applicable service and actual policy, rather than looking for a universal Copilot “High/Medium/Low” filter page or a pane-level “Data Protection” switch. Clearing cookies cannot change server-side safety rules or grant missing source access.

Microsoft does document a narrowly scoped harmful-content protection policy for authorized specialist work such as investigations or legal review. It is an organizational governance choice for selected users, not the normal fix for a filtered answer, and core safeguards remain enforced. The administrator guide excludes agents from its scope, while the user guide describes agent experiences. Have IT confirm current support for the exact host before considering any approved specialist deployment. These guides do not establish an unrestricted override for all Copilot apps.

For a suspected DLP mistake, give the authorized policy owner the notice, source, and business purpose for review. For a suspected safety-classification mistake, give support a sanitized reproduction. Neither route guarantees an answer or a policy exception.

Copilot Safety, Access, and Data Restrictions: Comparison