Voice and Communication Rules
Voice and Communication Rules
Section 3 asks how you communicate: your detail preferences, how bad news should reach you, your tone in conflict, the point where you walk away from a conversation, the tones you have banned, and — the strategic choice — whether the twin should write in your voice or a neutral professional voice, plus whether it admits it is AI. These are not style questions. They decide whether people trust what the twin produces.
Detail defaults and bad news
Start with the detail default. Do you want the headline first, then two bullets? Numbers without narrative? A one-line summary with detail behind a link? Write the formula, because the twin will apply it hundreds of times. Then the bad-news rule, which protects you from the twin's most dangerous instinct: softening. LLMs are trained to please, and an unconfigured one will bury a problem under reassurance. Your rule might read: "Tell me bad news in the first sentence, with the three options you see. Never wrap it in background. If I would be angry in two days, tell me today." Give the twin the exact shape, and it will never cushion a crisis into a footnote.
Tone, conflict, and the walk-away point
Define your conflict tone: calm, direct, and about the issue rather than the person — or whatever your real style is, so long as it is described as behavior. Then define the walk-away point: the specific behavior that makes you end a conversation. Common entries: repeated lying, disrespect toward you or your staff, or arguing in circles after the decision. The twin needs both rules because it will meet hostile messages on your behalf: it should hold your tone and end exchanges at your line, not at its own frustration threshold.
The banned-tone list
Most owners can name tones they would never use: sarcasm at a client's expense, false cheer, passive-aggressive questions, guilt-based urgency, or corporate phrases nobody actually says. List yours literally, in words. This is one of the few profile rules an LLM obeys with near-perfect consistency, because tone is the thing language models control best; a tone you ban today disappears from every draft tomorrow.
Voice or neutral — decide deliberately
Now the crucial decision. Should the twin write as you, or as a neutral professional assistant? Mimicking your voice is fragile: style without your reasoning underneath reads odd the moment a situation is unusual — like an impressionist improvising your policy decisions. The strong default is a neutral professional voice carrying your judgment in the content: your priorities, your exact trade-offs, your standards of proof. Save voice mimicry for narrow, low-risk message types if at all, and decide once, in writing — later lessons build on the choice.
Then answer the honesty rule: does the twin say it is AI? Owners differ — some accept a line like "Prepared by the owner's AI assistant, reviewed and sent by the owner," while others prefer a human handoff. What is not permitted anywhere: passing as you live. Decide the identification line now; Lesson 9 wires it into every external message so the choice is enforced, not remembered.
Key Takeaways
- Section 3 sets detail defaults, bad-news delivery, conflict tone, the walk-away point, and banned tones.
- Bad news arrives first, unwrapped — write the exact shape you want the news to take.
- A banned-tone list is enforced near-perfectly by LLMs; write your tones out literally.
- Neutral-professional voice with embedded judgment beats voice mimicry by default: carried thinking beats carried style.
- Decide how the twin identifies itself as AI, and never permit live impersonation of you.
Voice and Communication Rules
Section 3 asks how you communicate: your detail preferences, how bad news should reach you, your tone in conflict, the point where you walk away from a conversation, the tones you have banned, and — the strategic choice — whether the twin should write in your voice or a neutral professional voice, plus whether it admits it is AI. These are not style questions. They decide whether people trust what the twin produces.
Detail defaults and bad news
Start with the detail default. Do you want the headline first, then two bullets? Numbers without narrative? A one-line summary with detail behind a link? Write the formula, because the twin will apply it hundreds of times. Then the bad-news rule, which protects you from the twin's most dangerous instinct: softening. LLMs are trained to please, and an unconfigured one will bury a problem under reassurance. Your rule might read: "Tell me bad news in the first sentence, with the three options you see. Never wrap it in background. If I would be angry in two days, tell me today." Give the twin the exact shape, and it will never cushion a crisis into a footnote.
Tone, conflict, and the walk-away point
Define your conflict tone: calm, direct, and about the issue rather than the person — or whatever your real style is, so long as it is described as behavior. Then define the walk-away point: the specific behavior that makes you end a conversation. Common entries: repeated lying, disrespect toward you or your staff, or arguing in circles after the decision. The twin needs both rules because it will meet hostile messages on your behalf: it should hold your tone and end exchanges at your line, not at its own frustration threshold.
The banned-tone list
Most owners can name tones they would never use: sarcasm at a client's expense, false cheer, passive-aggressive questions, guilt-based urgency, or corporate phrases nobody actually says. List yours literally, in words. This is one of the few profile rules an LLM obeys with near-perfect consistency, because tone is the thing language models control best; a tone you ban today disappears from every draft tomorrow.
Voice or neutral — decide deliberately
Now the crucial decision. Should the twin write as you, or as a neutral professional assistant? Mimicking your voice is fragile: style without your reasoning underneath reads odd the moment a situation is unusual — like an impressionist improvising your policy decisions. The strong default is a neutral professional voice carrying your judgment in the content: your priorities, your exact trade-offs, your standards of proof. Save voice mimicry for narrow, low-risk message types if at all, and decide once, in writing — later lessons build on the choice.
Then answer the honesty rule: does the twin say it is AI? Owners differ — some accept a line like "Prepared by the owner's AI assistant, reviewed and sent by the owner," while others prefer a human handoff. What is not permitted anywhere: passing as you live. Decide the identification line now; Lesson 9 wires it into every external message so the choice is enforced, not remembered.
Key Takeaways
- Section 3 sets detail defaults, bad-news delivery, conflict tone, the walk-away point, and banned tones.
- Bad news arrives first, unwrapped — write the exact shape you want the news to take.
- A banned-tone list is enforced near-perfectly by LLMs; write your tones out literally.
- Neutral-professional voice with embedded judgment beats voice mimicry by default: carried thinking beats carried style.
- Decide how the twin identifies itself as AI, and never permit live impersonation of you.
There are no comments for now.