---
sourceDocument: Yokohama Conversational Interfaces
sourceDocumentLink: https://servicenow-prod.fluidtopics.net/r/yokohama/conversational-interfaces

 Release :

    - yokohama

ft:locale :

    - en-US

ft:publication_title :

    - Yokohama Conversational Interfaces

ft:clusterId :

    - convint

bundleId :

    - convint

workflow :

    - Platform


---

# Profanity Filter for Agent Chat overview

# Profanity Filter for Agent Chat overview {#ariaid-title1}

Release version: Yokohama  
Updated January 30, 2025  
![](https://www.servicenow.com/docs/portal-asset/ico-clock) 3 minutes to read
Summarize  
![AI sparkle icon](https://servicenow.com/docs/portal-asset/ai-sparkle-icon) Summarized using AI  
This content was generated using new OpenAI-powered functionality. Results are provided on an as is basis and are not guaranteed to be accurate or complete.  

## Summary of Profanity Filter for Agent Chat overview

The Profanity Filter for Agent Chat in the Yokohama release prevents agents from sending messages containing profane language or forbidden keywords to requesters.
When an agent attempts to send such a message, the filter blocks it from being delivered, highlights the forbidden keywords in the agent's chat window, notifies the agent about the violation, and flags the usage to the chat manager.
This feature helps maintain professional communication and protects requester experience by preventing offensive language exposure.
Show full answer Show less  

## Key Features

* **Message Blocking and Notifications:** Messages with forbidden keywords are blocked and highlighted for the agent, who receives immediate notification.
* **Chat Manager Alerts:** Chat managers receive real-time or summary notifications if an agent exceeds a configured threshold of forbidden keyword usage within a specific time frame, allowing proactive oversight.
* **Transcript Marking:** Forbidden keywords are recorded in chat transcripts for audit and review purposes.
* **Machine Learning and Keyword Models:** The filter uses either a machine learning-based model or a keyword-based model:
  * The **machine learning model** evaluates keywords contextually within the entire message, reducing false positives (e.g., allowing "UNIX for Dummies" but blocking "You are a dummy"). This model is centrally managed and cannot be customized per instance.
  * The **keyword-based model** allows explicit configuration of allowed and forbidden keywords or phrases, enabling customization to permit certain phrases (like book or movie titles) or block specific terms regardless of general profanity status.
* **Domain Separation Support:** The filter respects domain separation settings, applying keywords according to domain scope. Administrators can configure domain visibility via child domain inclusion, access grants for sister domains, or setting keywords under a global domain.

## Practical Benefits for ServiceNow Customers

* Ensures professional and respectful communication in agent-requester chat interactions.
* Provides real-time oversight and control to chat managers to address inappropriate agent behavior promptly.
* Offers flexibility to customize keyword filtering according to organizational needs and contextual exceptions.
* Supports domain-specific configurations to align with multi-domain environments and security models.  
The Profanity Filter for Agent Chat prevents
agents from sending messages to requesters that include profane language (also known as forbidden
keywords). If an agent is upset and tries to send a message with offensive language, the Profanity Filter prevents the requester from seeing the message.

If an agent tries to send a message with profane language, the Profanity Filter
blocks the message and flags the forbidden keywords. The Profanity Filter warns
the agent that the message contains forbidden keywords and alerts the chat manager to the agent's
use of profane language.  
When the Profanity Filter detects forbidden keywords in a message entered by the agent:

* the message containing the forbidden keywords is highlighted in the agent's chat window.
* the agent receives a notification that they have used a forbidden keyword.
* the message with the forbidden keyword is not sent to the requester.
* if the agent has exceeded the number of forbidden keywords in a specified amount of time, a message is sent to the chat manager.
* the forbidden keyword is marked in the transcript of the chat conversation.
{#ci-profanity-filter__ul_bwj_srq_yrb}

## Notifying the chat manager {#ci-profanity-filter__section_q4w_ktq_yrb}

You can configure the Profanity Filter to notify the chat manager if an agent
tries to use more than the specified number of forbidden keywords over the specified time. The
Profanity Filter considers all conversations owned by the agent when determining
the number of forbidden keywords used. If the agent tries to send a message with a forbidden
keyword, the chat manager receives a notification. If the Manager Workspace or Agent Workspace
is the active window, a real-time notification displays. If the Manager Workspace or Agent
Workspace isn't the active window, a notification displays on the summary view in the Workspace
notification menu. In the incident entry, the chat manager can review the incident and the chat
log and privately message the agent.

## Machine learning-based or keyword-based {#ci-profanity-filter__section_rjk_v5q_yrb}

The Profanity Filter uses a machine learning-based model or a keyword-based
model to determine whether a message includes forbidden keywords. The machine learning-based
model analyzes keywords and phrases in the context of the agent's entire message before
determining whether any forbidden keywords in the message should be considered profane. For
example, the statement "I made a silly mistake" would be considered not profane while "You sound
so silly" would be considered profane. The machine learning-model is not on each instance
individually -- you cannot change the model directly to account for customized keywords and
phrases.

The keyword-based model compares a message against a list of forbidden keywords to determine
whether something should be considered profane. You can allow certain keywords or phrases even
if those keywords are considered profane in a generic sense. Common examples are book or movie
titles that might contain forbidden keywords and but should still be allowed in a conversation.
Conversely you can configure the Profanity Filter to block certain keywords or
phrases even if the words may not always seem profane. The Profanity Filter
analyzes each message against the list of allowed and forbidden keywords before running it
through the model-based mode.  
{#ci-profanity-filter__table_t4b_mtq_wsb__entry__4}

| Allowed keyword | Forbidden keyword | Message to be analyzed | System behavior |
|-|-|-|-|
|   |   | You are a dummy. | Message is blocked. |
| UNIX for Dummies |   | The reference book "UNIX for Dummies" might be useful. | Message is allowed. |
| UNIX for dummies | stupid | "UNIX for Dummies" is a stupid book. | Message is blocked. |
|   | foolish | I feel so foolish for not knowing the answer to your question. | Message is blocked. |
|   |   | I feel so foolish for not knowing the answer to your question. | Message is allowed. |
[Table 1. Example of machine learning-based model]

{#ci-profanity-filter__table_t4b_mtq_wsb} {#ci-profanity-filter__table_u4h_ztq_wsb__entry__4}

| Allowed keyword | Forbidden keyword | Message to be analyzed | System behavior |
|-|-|-|-|
|   | dummy | You are a dummy. | Message is blocked. |
|   |   | You are a dummy. | Message is allowed. |
[Table 2. Example of keyword-based model]

{#ci-profanity-filter__table_u4h_ztq_wsb}

## Domain separation {#ci-profanity-filter__section_fw2_2gb_jsb}

The Profanity Filter works with domain separation. If the domain is set to Global, then the keywords apply to both the agent and requester domains. For the agent and requester to see each other's domain, the administrator must do one of the following:

* Include the domain where the regular expression should be visible as a child domain to the corresponding parent domain.
* Grant necessary access to the appropriate domain if it's a sister domain.
* Set the keywords under a global domain.
{#ci-profanity-filter__ul_akq_zgb_jsb}