How automated assistants are disclosed

AIVAS agents can write in an account owner's voice. They are still automated assistants. Meta recommends telling people when they are in an automated chat, and some laws require it.

This is how AIVAS discloses that, when enabled, and what a lead hears if they ask.

Optional setting: Automated assistant disclosure is disabled by default. When disabled, agents maintain the owner persona and escalate AI questions silently to human review.

Disclosure is configured in the organization's app_settings:

• One agent: set agent.assistant_disclosure.<agentId>.enabled to true. For example, agent.assistant_disclosure.instagram-manager.enabled applies only to Instagram Manager in that organization.
• Organization default: set agent.assistant_disclosure.enabled to true. This applies to eligible agents without an agent-specific setting. agent_assistant_disclosure_enabled and assistant_disclosure_enabled are also organization-wide keys; they do not target one agent.

An agent-specific false overrides an organization-wide true. Patty V1 and email agents are excluded.

Important: Meta can change its policies. Its current documentation takes priority over this article.

What the lead is told

On the moments below, the agent includes a short disclosure in the DM, in the account's usual voice. The shipped line is:

Quick heads up: this is an automated assistant. I write in this account's voice, but the owner didn't type this.

The agent can paraphrase that into the persona's texting style. It must not claim the owner personally typed the message.

Product and legal own that copy and the moments it appears. Changing them is a product decision, not something each conversation improvises.

When the disclosure appears

The disclosure is included when:

1. Conversation start — the first automated outbound on the thread.
2. Significant lapse — 24 hours since the last automated outbound.
3. Human-to-automation — a person just wrote from Inbox, and the agent is taking the thread again.

It is not repeated on every reply. After a recent automated message, the agent keeps talking in voice without restating the line.

If someone asks whether they are talking to AI

They get a plain, truthful answer in the same voice, then the conversation continues. The agent does not stay silent, joke it off, or refuse to say.

Asking "are you a bot?" is not, by itself, a reason to hand the thread to a person.

When a person takes over

If the agent hands the thread to a teammate (trolling after one deflection, anger, a report threat, a missing answer that needs a human, or a booked live session), it sends:

A person on the team is taking over this chat now.

Then the agent stops until someone marks the thread handled.

What stays the same

Owner voice can stay first-person. Disclosure is a short honest line, not a switch into a corporate bot persona.

Patty V1 (the frozen Personal Trainer) still uses the older dodge-and-escalate rule for AI questions. Conversations on Patty V2, James, and Instagram Manager follow this article when disclosure is enabled.

Official source

Messenger Platform policy: Informing users about your automated experience