AI Content Moderation ===================== TouchPoint checks chat content with Azure Content Safety before it reaches the conversation. Every new message, every edit to a message, and - where your church has turned it on - every photo is evaluated first. Content that fails the check is never posted: the member sees your blocked-content message instead, and staff can review the attempt in `Blocked & Reported` in `Chat Management`. See :doc:`Chat` for the rest of the chat feature. The settings are found at `Admin > Advanced > Settings > Integrations > Azure`, in the `Content Safety` section, and require the **Admin** role. They apply church-wide, to every conversation - there is nothing to configure on an individual chat or Involvement. The Settings ------------ .. figure:: images/ChatContentSafety.png :target: # The `Content Safety` settings, found by typing ``Content`` in the settings search box +---------------------------------------+-------------------------------------------------------------------+ | **Setting** | **Purpose** | +=======================================+===================================================================+ | Azure Content Safety Key | Your own Azure Content Safety API key, from the `Keys and | | | Endpoint` page of your Content Safety resource in the Azure | | | Portal. Leave it empty to use TouchPoint's managed key | +---------------------------------------+-------------------------------------------------------------------+ | Azure Content Safety Endpoint | The endpoint URL of your own Content Safety resource. Used only | | | when you supply your own key; otherwise TouchPoint's managed | | | endpoint is used | +---------------------------------------+-------------------------------------------------------------------+ | Azure Content Safety Evaluate Images | Whether photos are evaluated as well as text. This applies only | | | when you have supplied your own key and endpoint. Default: False | +---------------------------------------+-------------------------------------------------------------------+ | Azure Content Safety Blocked Message | The message the member sees when their content is blocked. | | | Default: `This content has been blocked because it may violate | | | our community guidelines.` | +---------------------------------------+-------------------------------------------------------------------+ | Abusive Content Threshold | The severity at which hateful or abusive content is blocked | +---------------------------------------+-------------------------------------------------------------------+ | Self-Harm Content Threshold | The severity at which self-harm content is blocked | +---------------------------------------+-------------------------------------------------------------------+ | Adult Content Threshold | The severity at which sexual content is blocked | +---------------------------------------+-------------------------------------------------------------------+ | Violent Content Threshold | The severity at which violent content is blocked | +---------------------------------------+-------------------------------------------------------------------+ Categories and Levels --------------------- **The four categories** Content is scored separately for abusive, self-harm, adult, and violent content, and each category has its own threshold. Content is blocked when *any one* category reaches its threshold, so the strictest setting is the one that decides. **The severity levels** Each threshold is a number from `0` to `7`. Content scoring *at or above* the threshold is blocked, which makes a lower number stricter: - `0` blocks everything in that category. - `7` blocks only the most extreme content. - Each threshold arrives set to `2`, which blocks low-severity content and above. .. caution:: A threshold of `0` blocks every message in that category, including ordinary conversation, because even content scored as safe is at or above the threshold. With `Abusive Content Threshold` set to `0`, a message as harmless as ``I hate vegetables`` is blocked. Microsoft publishes the categories and what each severity level means, with examples, in its `Azure AI Content Safety harm categories `_ documentation. .. warning:: The Microsoft documentation quotes examples of the content each level describes. Some of it is disturbing to read. Reviewing What Was Blocked -------------------------- Blocked messages and photos are recorded, not posted. `Chat Management` counts them on each conversation and lists them in `Blocked & Reported`, where staff can open the blocked content deliberately with `View blocked content`. Blocked photos are not kept - the entry shows that an image was blocked rather than the image itself. .. admonition:: Use Case To be stricter about abusive language than about the other categories, leave `Abusive Content Threshold` at `2` and raise `Self-Harm Content Threshold`, `Adult Content Threshold`, and `Violent Content Threshold` to `4`. Every chat in the church follows the new values: low-severity abuse is refused as it is sent while the other categories are blocked only at a higher severity. The member sees your `Azure Content Safety Blocked Message`, and a leader with the **ManageChat** role reviews the attempts in `Blocked & Reported`. | | +--------------------+------------------+ | **Latest Update** | **09/15/2026** | +--------------------+------------------+ New article documenting the Azure Content Safety settings for Chat, the four content categories, the severity thresholds and their defaults, image evaluation, and how blocked content is reviewed in Chat Management.