# AI spam checks for incoming conversations

Classify eligible new text and social conversations as spam before they reach your operators.

*Release 3.3 · admin, operator*

**Settings › AI › Features**

## What it is

**Spam Check** uses AI to decide whether an eligible message needs your team's attention. A block decision includes a reason, and INSIDE records the message for spam review. This helps keep unwanted new text and social conversations out of your operators' queues.

## Who uses it and why

Users with the **Settings** or **Developer** role choose which platforms and visitors to check and define the instructions. Authorised operators review detections and correct legitimate messages marked as spam.

## Where to find it

Open **Settings › AI › Features** and choose **Spam Check** as the feature type. Configuration requires the **Settings** role with **AI** allowed under **Settings Permissions**, or the **Developer** role.

Review detections under **Conversations › Spam**. The **Settings** role or **Manage Spam Messages** under **Manager** permissions grants review access.

## How it works

1. A customer sends eligible text that starts a new conversation.
2. INSIDE selects a matching enabled **Spam Check** feature using its account, priority and visitor filters.
3. The AI assesses the text and returns an allow or block decision. Classification does not call additional functions.
4. Before applying a block, INSIDE checks whether the message text has already been corrected as a false positive.
5. A block records the message and reason in **Spam** and stops that new conversation from entering normal handling. An allow decision lets normal handling continue.

The built-in instructions cover unsolicited offers, phishing, abuse, irrelevant or malicious content, and messages consisting only of unrecognised characters or emoticons. They are written for social comments, so review their suitability for your selected channels.

## Configure it

Ask Powerfront to enable AI and prepare an account. Create a **Spam Check** feature, choose its **Account**, review **Instructions**, and add platform conditions in **Visitor Filters (drag and drop)**. Select at least one supported text or social platform. The editor rejects **Chat** and **Email**.

Check **Enabled** and give overlapping features distinct **Priority** values. Test representative legitimate enquiries as well as unwanted messages before expanding the filters.

### Settings

| Setting | Where | Applies to | Default | Who can change it | What it does |
|---|---|---|---|---|---|
| Instructions | Settings › AI › Features | website | Built-in rules blocking unsolicited offers, phishing, abuse, irrelevant messages and unrecognised-character or emoticon-only content | Settings role, Developer role | Defines which messages the AI should allow or block and asks for a reason when blocking. |

### Permissions

| Permission | Lets you |
|---|---|
| Settings | Configure Spam Check when AI is allowed under Settings Permissions. |
| AI | Allows AI configuration within Settings Permissions. |
| Developer | Configure Spam Check. |
| Manage Spam Messages | Review detections and correct false positives through Manager permissions. |

## Limits and good to know

- The normal check runs at conversation creation. It does not reclassify every message in an ongoing conversation.
- Attachment messages are excluded, and some channel interactions bypass classification.
- A failed or unavailable classification is not itself a spam verdict. Without a final block decision, normal intake continues.
- **Not Spam** corrects the detected text. It does not automatically remove separate visitor or sender blocking.
- There is no dedicated numeric confidence threshold for this feature.

## Depends on and works with

### This feature uses

- **Requires** [AI accounts and shared defaults](AIP-001.md): Classification needs a configured AI account.
- **Configured in** [Roles and permissions](ADM-008.md): Roles and permissions control configuration and spam review.
- **Configured in** [AI feature settings and change history](AI-001.md): Feature enablement, account, priority and visitor filters control selection.
- **Configured in** [AI instructions and templates](AI-002.md): Instruction sources define the classification rules.
- **Configured in** [Spam review and blocking](CHAT-020.md): Spam review lets authorised operators whitelist text that was incorrectly blocked.
- **Configured in** [Visitor event targeting conditions](ENG-002.md): Platform and visitor conditions determine eligible conversations.
- **Works better with** [AI guardrail routing and fallback actions](AI-025.md): Configured guardrails can stop a spam check without returning a spam verdict.
- **Needs licence or module** Admin settings and feature switches (managed by Powerfront): Powerfront must enable AI for the website.

### Used by

- **Enhances** [Spam review and blocking](CHAT-020.md): AI spam classification supplies detections for review.

## FAQ and troubleshooting

### Why can I not save a feature for Chat or Email?

**Spam Check** is restricted to text and social platforms in this release. Remove **Chat** and **Email** from the platform filter and select a supported platform.

### Why did a later message in a conversation get through?

The normal intake check assesses eligible conversation-starting text. It is not a check of every subsequent message.

### How do I correct a legitimate message marked as spam?

Open **Conversations › Spam**, find the detection and choose **Not Spam**. Check any separate visitor or channel blacklist if the sender remains blocked.

### Does an AI service error block the customer?

An error does not count as a spam classification. INSIDE may try a fallback; if it obtains no block verdict, normal intake continues.