Describe style in observable terms
Instructions such as “be authentic” are difficult to evaluate. Specify language, sentence length, greeting style and how to handle uncertainty. Include a few authorised examples of suitable phrasing. Avoid asking the system to invent personal experiences or make factual claims it cannot support.
The brief builder assembles your inputs into a document. It does not learn a voice, call an AI service or send a message. Review the generated text before using it in any product. A template is a starting point for a human decision.
Keep facts separate from preferences
Facts include authorised account information, available content and promises that can actually be fulfilled. Preferences include tone, formatting and how much detail to use. Keeping the two separate helps you change style without accidentally changing the factual ground.
Leave sensitive personal data out of a general-purpose brief. Do not paste a fan’s private messages into this tool. Store necessary operational information in the system approved for that account, with access limited to the people who need it.
Write the exceptions before the ideal reply
Define what must be escalated: requests outside the account’s boundaries, disputes, requests to stop and anything involving uncertainty about authorisation. Say what the system should do while waiting for a person. Silence, acknowledgement and escalation are different actions.
An AI-assisted interaction should not rely on misleading someone about who or what is responding. Set an appropriate disclosure approach and check the platform’s current rules. Test the brief against difficult examples, then revise it from observed failures rather than continually adding vague adjectives.
Write a brief that survives a change of operator
Replace adjectives with examples that can be checked. “Warm” might mean acknowledging the question before giving one clear answer; “concise” might mean using two short sentences unless more explanation is necessary. Avoid arbitrary limits that prevent a complete response. Include both an acceptable example and a near miss so reviewers understand the distinction.
Maintain an authorised fact sheet separately from the tone instructions. Prices, availability, delivery expectations and account-specific information need an owner and a current source. When a fact changes, update that source rather than leaving contradictory versions in several prompts. A conversational system should be able to say that it does not know.
Test a brief with incomplete context as well as routine questions. If a subscriber asks about a previous promise that is not in the available context, the system should ask for clarification or hand over. It should not infer a personal relationship or invent details to maintain the tone. Style is successful when it improves clarity while preserving the facts.
| Layer | A useful instruction | An instruction to avoid |
|---|---|---|
| Voice | Use short, warm sentences and one clear next step. | Be irresistible and sell at any cost. |
| Facts | Use the current approved content list. | Make up a personal story when needed. |
| Boundaries | Ask the named owner about unavailable requests. | Promise that every request can be fulfilled. |
| Transparency | Describe automated interaction honestly. | Insist that the creator is personally typing. |
ILLUSTRATIVE WORKED EXAMPLE
From a vague instruction to a reviewable voice rule
Illustrative starting instruction: “Be friendly, personal and engaging.” This does not say which facts are allowed or how to handle uncertainty.
- Define the voice: “Acknowledge the question, use plain language and offer one relevant next step.”
- Define the fact boundary: “Only mention content, prices and availability from the approved account note.”
- Define the exception: “When the note does not answer the question, ask for clarification or assign it to the account owner.”
The revised brief produces observable criteria. A reviewer can identify an invented fact or an unnecessary promise without debating whether the message feels friendly enough.
Export the brief, add authorised examples and record which version was used during evaluation.
Put it into practice.
Build a practical instruction document. This is a template builder, not an AI model or a live inbox.
Write your brief ↗Persona brief examples for voice, facts and boundaries →
Related questions
Does the builder use generative AI?
No. It formats the text you enter into a structured brief using fixed templates.
Are my entries stored?
The tool keeps its working state in the current browser tab. Downloads are saved on your device; there is no account or server submission.
Can I use someone else’s persona?
Use only information and examples you are authorised to use. The tool does not verify consent or ownership.
Original practical guidance prepared for Onlytool. Worked cases are illustrative, not measured customer results. This publication does not claim independent vendor testing. Methodology and disclosure.