Responsible AI Practices for Client Work

Handle client data like it is radioactive

The fastest way to lose a client is to paste their unreleased roadmap into a consumer chatbot. Use business tiers of AI services where inputs are excluded from training, and know each vendor's retention terms before any client material touches the tool. Anything under NDA deserves an explicit decision, not a habit.

Disclose in a way that builds confidence

Clients rarely object to AI-assisted work; they object to discovering it. A short clause in your agreement covering which categories of tools you use, how client data is protected, and the fact that humans review all deliverables turns a potential surprise into a professionalism signal.

Accountability stays human

• Every deliverable gets a named human reviewer before it ships.
• Facts, statistics, and citations get verified at the source; models fabricate all three convincingly.
• Generated code goes through the same review and testing as handwritten code.
• Check licensing terms on generated assets, especially imagery, before commercial use.

Write it down

A one-page policy covering approved tools, prohibited inputs, review requirements, and disclosure language is enough. Revisit it quarterly, because the tools change fast and your commitments to clients should not depend on who happens to remember the rules.