Keep unreleased and proprietary details out of consumer tools

Documentation work runs right up against things that are not public yet: the feature shipping next quarter, the internal endpoint, the customer name in a support transcript, the API key in a config example.

A personal or consumer AI account has very different data terms from an approved company tool. Content you paste can leave your control and, depending on the tool and its settings, may be retained or used to improve a model. For unreleased or proprietary material, that is a leak.

What to keep out of consumer tools:

• Unreleased features, roadmap details, and internal codenames
• Real API keys, tokens, connection strings, and passwords — use placeholders such as YOUR_API_KEY
• Customer names, account numbers, and real personal data from tickets or logs
• Internal-only endpoints, hostnames, and architecture details

How to still get the help:

• Redact and abstract. Replace the sensitive specifics with placeholders and describe the shape, not the secret. AI can structure a doc perfectly well around SERVICE_NAME and [INTERNAL_ENDPOINT].
• Prefer the approved tool. If your organization provides an enterprise or private AI with a data-processing agreement, use that for anything non-public.
• Draft public, fill private. Do the AI-assisted drafting on the public-facing description, and add the sensitive real values yourself, offline.

The habit is simple: before you paste, ask whether this could appear on the public docs site today. If not, redact it first.