Privacy & limits
AI features are only trustworthy if the boundaries are explicit. Here is the whole boundary set — what the AI sees, what it can never do, where the limits bite, and how they degrade.
What the AI sees
| Feature | Sees |
|---|---|
| Content generation | What you select / paste / describe in the prompt |
| Chatbot | Your PUBLISHED content + platform docs (retrieved per question) |
| Ticket suggestions | The ticket text + your published content |
| Design agent | The canvas you are editing |
What NONE of it sees: your password, other users' sites, unpublished drafts (for the chatbot), or anything outside the task's scope.
What it never does
- Never publishes. Generation fills drafts; publishing is your click.
- Never trains on your content. Your content grounds YOUR conversations; it is not used to train models. Platform AI runs on configured providers with their own data policies — bring-your-own keys are encrypted at rest.
- Never edits invisibly. AI changes flow through the normal editor with undo; the design agent returns a canvas through the same pipeline as your own edits.
- Never invents your facts. Where your content is silent, the honest answer is "I don't know" — or an escalation to a human (see AI in the chatbot).
Quotas and honest degradation
Plan meters count assistant + chatbot usage monthly, with remaining quota shown next to each AI tool. When a meter empties, tools degrade HONESTLY: the button shows the limit and why — nothing fails silently, nothing deletes, nothing downgrades your published content. The upgrade (see Plans compared) or the month's reset restores it.
Model and provider choices
| Setting | Where | Notes |
|---|---|---|
| Default model | AI → Providers | Applies to new conversations |
| Provider keys | AI → Providers | Bring your own; encrypted at rest |
| Fallback | Automatic | Falls back to the platform provider if yours errors |
Choosing a provider is a trust decision made once and revisited rarely; the fallback exists so a provider outage never becomes YOUR outage.
Caution
Do not paste into any AI tool what you would not paste into a form: client secrets, private keys, anything covered by an NDA. The AI is scoped and private, but the habit of discretion is free.
The one-paragraph summary
The AI layer is scoped, reversible and honest: it reads what the task needs, writes only where you allow, spends what your plan meters, and tells you plainly when it cannot help. The rest — judgment, facts, the publish button — was never its job.
Note
Questions about data handling for a specific compliance need (client contracts, institutional rules) are best asked as a support ticket — you will get the current, specific answer rather than a doc that may have aged (see Contacting support).