How your default voice works

How Agent Smith and Toto talk with you

Both agents use simple, everyday language for questions, explanations, editorial feedback and change summaries. They use professional or technical terms when those terms help, with explanations when needed.

This shared application voice keeps conversations clear and approachable. Changing your writing preferences does not change the agents’ everyday conversational voice.

How they write for you

Articles, channel adaptations and newsletter content follow your account’s editorial guidelines. Smith uses them when drafting, revising and assessing article style. Toto uses them alongside the audience, purpose and requirements of the selected channel.

The agents’ chat style is not the standard your article must match. For example, Smith can explain an edit in plain language while preserving a more formal passage that fits your writing and audience.

Your starting guidelines

Eligible new authors start with Talkbase Editorial Foundation v2 after meeting the verified-email and author-access requirements. The app creates an account-owned copy of these guidelines, giving you a starting point before you personalize them.

The Foundation does not mean the app has already learned your distinctive style. It provides general guidance for clear writing that respects your thinking.

What each layer contributes

The two voices sit within a wider set of instructions and context. Each part has a different job; they are not blended into one general tone.

Shared working principles

These guide how both agents collaborate: respect your decisions, distinguish evidence from assumptions, and preserve your meaning. A style preference does not give an agent permission to invent facts or speak for you.

Agent role

Smith helps develop thinking and articles. Toto adapts an existing article for a particular audience and channel. Their roles define the work they do, not a separate personal writing style for each agent.

Application voice

This shapes what the agents say to you: questions, explanations, feedback and descriptions of changes. It favors familiar words and clear explanations.

Your editorial guidelines

These shape writing produced for you: vocabulary, rhythm, formality, structure, audience and evidence standards. Foundation v2 supplies the initial guidelines for eligible new accounts. Your approved revisions become your account’s own version of those guidelines.

Pattern suggestions

These are possible refinements drawn from eligible feedback, writing samples and references you select for style or structure. They stay separate from your active guidelines. Even a well-supported pattern has no authority to change your voice until you review and explicitly approve the proposed wording in Voice Settings.

The current piece and channel

Your developed argument, notes, examples and corrections determine what this piece says. Audience, purpose and channel requirements guide its presentation. Retrieved material can support the work; it does not automatically become a writing preference or an instruction for the agent.

How the agents interpret the layers

The app supplies instructions for the task being performed and identifies which parts of the response are commentary and which are authored content. In ordinary Smith conversation, your saved editorial guidelines are not loaded as the voice for his reply. In article drafting, revision and review, the app supplies the full account guidelines for article text.

A single response can use both voices. An article follows your editorial guidelines, while its change summary uses the application voice. Toto’s post or newsletter follows your guidelines and the channel requirements, while an explanation of the approach uses the application voice.

The model interprets these written instructions in context. They are not a word substitution list, a fixed template or a numerical style score. “Use accessible language,” for example, means making the reasoning understandable without removing useful technical depth. “Be concise” means removing unnecessary material without cutting an essential qualification.

The Foundation is the starting content of your account’s guidelines, not an extra style imposed over every personalized version. Other task instructions still apply, including preserving meaning, handling evidence honestly and returning the format supported by the app.

When instructions point in different directions

• Chat language and article style: an agent can explain a change simply while keeping your article formal, technical or deliberately lyrical. The chat voice should not overwrite your prose.
• A standing preference and a request for this piece: if your guidelines favor detailed articles but you request a short introduction, that request guides this piece. It does not permanently change your length preference.
• Clarity and distinctive style: a deliberate fragment or repeated phrase can serve your voice. The agent should consider its purpose and your preferences before treating it as a problem.
• Persuasion and evidence: asking for a stronger opening should not turn “may help” into “will work” without support. The agent should improve expression while preserving the strength of the original claim.
• Channel conventions and source meaning: Toto can change the presentation for a post or newsletter, but should preserve the article’s position and important qualifications rather than inventing a more dramatic claim.

If a meaningful choice depends on your intent, the agent should ask a focused question. A suggestion from an agent is not treated as your belief simply because it appeared in the conversation.

How interpretation is checked

Smith’s article workflow includes a separate editorial pass that considers the supplied draft, your guidelines and your developed thinking. It looks for problems such as changed meaning, unsupported claims, lost qualifications and a mismatch with your writing preferences.

These instructions guide model behavior; they do not guarantee that every output captures your voice perfectly. The review also includes general checks for clarity, flow and repetitive writing. Those checks are meant to be interpreted through your preferences, rather than used to erase intentional choices.

Specific guidance helps: instead of only “make it sound like me,” explain what matters, such as “keep my cautious conclusions” or “preserve the repeated phrase because it connects the sections.” Use feedback for the current piece, and approve a Voice Settings revision when you want a preference to carry into future work.

Make the guidelines your own

Voice Settings has four sections: Your voice, Writing samples & feedback, Suggested changes, and History. You can edit your guidelines directly or review suggestions grounded in your writing and feedback. Previous versions remain available in History.

A correction to one article does not automatically become a permanent writing rule. Your active guidelines guide new writing; work already underway can retain the voice version associated with it.

How ongoing work can help refine your voice

Pattern suggestions are a built-in way to learn from your work over time. Ongoing observation is off by default. When you turn it on in Voice Settings, the app can look for lasting preferences in new author messages to Smith and feedback submitted to Toto. Checks run while the app is open and visible; they are not continuous background monitoring. Past work is not included automatically—you choose a past topic when you want it analyzed.

You can also add writing samples, describe a preference, or select a reference for writing style or structure in Writing samples & feedback. A facts-only reference is not used to infer style. An outside reference can show an influence you want to explore; it is not proof of how you naturally write.

Suggestions refine your current approved guidelines. Foundation is the starting point for eligible new accounts, not a template that replaces your personalized voice whenever a new pattern appears. The aim is to preserve your distinctive wording and ways of developing an argument without inventing your views or evidence.

What makes a suggestion worth considering?

• A stated preference: you explain a choice you want to carry into future writing.
• Repeated feedback: a preference appears in your corrections across different pieces.
• A writing pattern: independently written samples suggest a characteristic choice.
• A possible preference: limited or context-specific evidence needs your confirmation.

Each suggestion shows supporting excerpts, an explanation, its scope and any uncertainty or conflict with your guidelines. Repeating an edit within one piece does not make it a pattern across pieces. Casual chat wording, silence, accepting a draft and the agents’ own prose are not evidence of your lasting style. AI-assisted or heavily edited samples are treated more cautiously.

For example, “Keep the qualification in this conclusion” may be a request for one article. Similar feedback across different pieces could support a suggestion to preserve cautious conclusions more consistently. Neither case changes your saved guidelines on its own. Stronger evidence helps you judge a suggestion; it never replaces your approval.

You decide what becomes part of your voice

1. Open Suggested changes in Voice Settings and inspect the evidence and proposed wording.
2. Edit a suggestion or add it to your review draft. You can also decide later, dismiss it, or choose “Don’t suggest this again.”
3. Use Review changes to compare the proposed guidelines with your current version. Saving a draft does not activate it.
4. Choose Approve for future writing only when the exact changes reflect what you want.

Approval creates a new account-wide voice version for future writing by Smith and Toto. There are no per-piece voice trials in this flow. Existing articles and adaptations keep their associated voice version, and the agents’ conversational language stays the same. Restoring earlier wording from History also requires review and approval as a new revision.

You can turn observation off, exclude a topic, or remove a saved source and its related suggestions. Full samples are saved only when you choose to keep them; supporting excerpts are retained with suggestions. Removing evidence does not undo approved guidelines or erase wording in a separately saved draft. Change approved guidelines through the same review-and-approval process.