A starts with complete sample dialogues, B checks staff-written scripts, and C tests out-of-scope requests before a final paper rehearsal. All outputs remain text; voice tool availability is not claimed.
Option A: ChatGPT: drafts the sample dialogues
Best when: The task list exists but the wording is not written.
What it does: ChatGPT drafts the dialogues; you review them with staff.
Strong at: Each task gets normal and stop-path wording together.
Limit: Text samples do not prove people will speak the prompts naturally.
Free plan: Yes, limited (images, voice, deep research, Codex)
Option B: Claude: flags ambiguous lines in staff scripts
Best when: Staff already have wording that fits how they ask.
What it does: Claude flags ambiguous lines; you accept or reject edits.
Strong at: The review starts from staff language, not invented style.
Limit: Gaps may remain if staff do not supply missing details.
Free plan: Yes
Option C: Gemini: tests unsafe requests before rehearsal
Best when: Staff need clear behaviour for ambiguity and out-of-scope pressure.
What it does: Gemini drafts the tests and pack; staff rehearse on paper.
Strong at: Boundaries are approved before dialogues are assembled.
Limit: A paper pass cannot establish real speech or integration safety.
Free plan: Yes