How we use AI

The help center assistant (Help assistant)

When you describe a problem in the help center, an AI assistant may answer from our help articles before you file a request. Every answer is badged as AI, it can be wrong, and "Talk to a person" is always one click away. If you file the request, our team sees what the assistant told you.

What the assistant tells you: “You're chatting with an AI assistant. It answers from our help articles and can be wrong — a person can take over at any time.”

Replies from our team

Our agents may start a reply from an AI draft. The AI never sends anything on its own: a person reads, edits and sends every reply, and stays responsible for it.

How AI-generated content is marked

• Email: an AI-Disclosure header (and X-AI-Assisted) on the message, and a matching ai-disclosure tag in its HTML.
• Help center: assistant answers and AI-drafted replies carry a machine-readable data-ai-disclosure attribute.
• SMS, chat apps and social messages cannot carry hidden labels, so a reply sent there exactly as the AI drafted it ends with a short written note.
• Our API and webhooks include an ai_disclosure field on each message, so systems that receive our replies can tell too.

Behind the scenes

AI also helps our team sort requests: it suggests a category, priority and summary, and gauges sentiment, so urgent problems are seen sooner. These suggestions are only for our team; they do not decide anything about you on their own.

The AI may suggest an action on your account, such as resetting a mailbox password or restarting a service. A person must approve it, and only after confirming it is really you — by your portal sign-in, a code we send to the address or number on file, or your support PIN.

Which AI we use

No external AI model is connected at the moment; the assistant answers only from our help articles.