Client Transparency Framework

The TRUST framework runs the structure. Tier the AI use into "always disclose" (AI generates the primary deliverable, AI handles client data, AI makes recommendations the client acts on), "mention proactively" (AI assists work that humans substantially edit), and "standard tool use" (grammar checks, transcription — disclosed if asked, not unprompted). Then reframe the narrative so the disclosure positions AI as a client benefit rather than a cost-saving secret — the line that works is "AI handles the mechanical 80% so humans focus on the strategic 20%."

The scripts are word-for-word, not summaries. Proactive disclosure for the kickoff. Reactive disclosure for when the client asks directly. The caught-off-guard script for when they notice something and you didn't get ahead of it. The renewal conversation script for when the topic comes up in a procurement review. Each is written to read naturally out loud, because policies the team can't say in a call are policies that don't survive contact with real clients.

Different clients need different defaults. A Series B SaaS expects AI tools in the stack and is more suspicious of an agency claiming purely human production than one being honest about the workflow. A law firm or regulated-industry client needs the opposite — more reassurance, more specifics on data handling, named tools rather than vague references, and an explicit statement that client documents never go into public models. The framework produces both versions when your portfolio mixes them.

You get the disclosure tiers mapped to your actual workflows, three to four conversation scripts written to be spoken, suggested contract clause language (flagged as "review with your lawyer"), an internal AI use log template so you can answer "where did AI touch this deliverable?" without scrambling, and a team FAQ for the questions that come up internally before they come up externally. The skill refuses to invent practices you don't actually run — the policy only works if it reflects reality, so the input is your real workflow, not an aspirational one.