Your goal
Separate writing style from factual knowledge and build a reviewable digital-twin profile.
Create consistent replies that sound like your community, without inventing a person.
Your sample organizer voice is concise, warm and practical. A resident asks about meetup timing and parking. The approved notes confirm10:00 but contain no parking information.
Separate writing style from factual knowledge and build a reviewable digital-twin profile.
Rewrite one overformal sentence using three explicit style rules.
A consented sample profile, a fact sheet and before/after evaluation pairs.
This interactive model teaches the mechanism. It does not call a model, search your files or send messages.
A writing example teaches tone, sentence length and structure. It does not establish that every fact in a new answer is true. Supply approved facts independently.
This exercise creates text that follows a style profile. It is not the real person, an authorized spokesperson or a system that can approve actions.
Compare the same request with and without the style profile. Judge factual support, clarity and rule adherence rather than whether it feels mysteriously human.
Use the supplied fictional style-samples.txt or your own writing that you have permission to use. Remove private details before including it in a chat.
Check: The profile has a documented source and permitted use.
Do not train on a neighbor’s messages without their agreement.
Define three rules such as short sentences, one clear next step and no unsupported promises. Keep examples beside rules.
Check: Each rule can be evaluated on a draft.
Avoid vague descriptions like write exactly like me.
List approved meetup facts and unanswered parking information. Keep this sheet separate from style examples.
Check: The same facts can be used with a different writing style.
Style imitation cannot turn a sample fact into current evidence.
Generate a neutral draft, then a style-guided draft for the same question. Score both on the same factual and style rubric.
Check: You can point to specific improvements and failures.
Label both outputs as drafts.
Ask the twin to claim it is the organizer or confirm parking. Require an honest automated identity and an unknown-fact response.
Check: It does not impersonate or fabricate a confirmation.
A draft assistant still needs a real human decision.
Store rules, examples, fact boundaries and review checklist in a text file. Test with a new question.
Check: Another person can use and update the profile.
Record which examples and facts need refreshing.
Act as an automated community draft assistant, not as the organizer. Style rules: short sentences, warm practical tone, one clear next step, no hype. Approved facts: meetup starts10:00 in the community room; bring a laptop. Parking is unknown. Draft a reply to: What time is it and is parking free? Then explain factual support, missing facts and which style rules you followed. Nothing is sent until a human approves.
The style examples were treated as current knowledge.
Try: Separate examples from the approved fact sheet and require source support.
The role encourages identity imitation.
Try: State automated draft assistant and require human approval.
The rubric only says sounds good.
Try: Score concrete rules such as sentence length, clarity and evidence.
Tick a criterion only after checking your own artifact. These are self-reported checks, not an automated certification.
Create a compact voice profile for your own hobby project or small business. Add a factual update routine.
A writing example teaches tone, sentence length and structure. It does not establish that every fact in a new answer is true. Supply approved facts independently.
This exercise creates text that follows a style profile. It is not the real person, an authorized spokesperson or a system that can approve actions.
Compare the same request with and without the style profile. Judge factual support, clarity and rule adherence rather than whether it feels mysteriously human.
Use the primary documentation to verify this part of your build.
Apply it in the project lab