Start with the job to be done

The working assignment is to produce a clear, consented thirty-second course welcome in synthetic speech for adult beginners listening on headphones or phone speakers. That sentence is narrower than “use AI voice generation.” It identifies a deliverable and a reviewer. Write a definition of done with three layers: the output must satisfy the audience's need; factual or functional claims must be traceable; and a named person must own the final decision. A technically convincing voice can still be unethical or deceptive. Permission must cover the intended use, duration, audience, storage, and revocation path.

Start by separating tasks. The model may draft, classify, transform, compare, or suggest. It may not silently approve, publish, grade, deploy, cite, or consent on someone's behalf. For this assignment the authoritative material is final script, listener, permitted voice identity, pace, emphasis, pauses, pronunciation guide, disclosure, and transcript. Anything absent from those inputs is either an explicit assumption or an unanswered question.

Use this prompt as a realistic starting contract:

Read this 50-word course welcome in a warm, clear fictional voice at about 135 wpm. Do not imitate any known person. Pause after sentence one; emphasize “try one thing.” Use the approved pronunciation guide for Nguyen. Generate a ten-second sample first and retain a transcript.

Notice what the prompt does: it states the setting, limits the output, names forbidden behavior, and requests evidence that can be reviewed. It does not ask the model to “make it amazing.” If a constraint matters, make it testable. Replace “be accurate” with a source boundary, formula check, test command, rights ledger, or approval step.

A useful response would look like this: A short sample whose pace, emphasis, pronunciation, and disclosure can be reviewed before rendering the full message. That description is intentionally observable. “Looks good” is not acceptance. The operator must listen without reading, compare every word to the script, check names, clipped endings, breaths, volume, pace, phone-speaker intelligibility, and transcript accuracy. Keep the source material beside the draft so review means comparison, not memory.

Run the work through script → rights check → sample → direct → render → listen → publish. Stop when an authoritative input is missing, a high-risk claim lacks evidence, private material cannot be safely removed, or the proposed action exceeds the permission granted. Escalation is successful workflow behavior, not model failure.

Common framing mistakes are celebrity imitation; ambiguous consent; pronunciation drift; robotic pacing; artifact breaths; missing disclosure; audio-only delivery without transcript. Prevent them by writing a one-paragraph job card: user, decision, deliverable, source of truth, constraints, reviewer, and stop condition. This card becomes the anchor for every later prompt.