Lesson 1: The Four Layers of Voice DNA

When you connect an account or upload samples to Bolta, the system isn't just guessing your "tone." It's building a structured snapshot across four distinct layers.

Layer 1 — Vocabulary

The words you use, and the ones you don't.

Do you say "use" or "leverage"?
"Folks" or "people" or "everyone"?
"Dope" or "awesome" or "great"?
Are there technical terms you use comfortably (e.g., "p99 latency") that wouldn't fit a marketing audience?

The vocabulary layer also captures the words you avoid. If you've never used "synergy" in 200 posts, the system learns that's not a word in your dictionary.

Layer 2 — Sentence Structure

How sentences are built.

Average length (8 words? 20?)
Are fragments common? ("Ship it. Ship it now.")
Do you start with conjunctions? ("And here's the thing…")
Do you ask rhetorical questions, then answer them?
Active voice vs passive?

This layer is what makes a paragraph feel like you, even if every word is different from one you've ever used.

Layer 3 — Tone Signals

The emotional register.

Confidence (declarative vs hedged)
Warmth (intimate vs detached)
Playfulness (puns, asides, jokes)
Punctuation choices (em dashes, ellipses, question marks)
Use of emojis (none, sparingly, freely)

Layer 4 — Patterns

Repeated structural choices.

How you open posts (hook style, opening line patterns)
How you close posts (CTA, reflection, cliffhanger)
Whether you use lists, paragraphs, or one-liners
How you transition between ideas

Lesson 2: Voice vs Persona

These are the two most-confused concepts in Bolta. Get the difference right and everything else clicks.

Voice = how you write

Voice is mechanical. It's vocabulary, structure, tone signals, and patterns. Voice doesn't care what you're writing about — it cares how you write whatever you write.

Voice should stay relatively constant. Your voice tomorrow looks like your voice today.

Persona = who is writing and why

Persona is identity. It's the character behind the keyboard — the role, the worldview, the motivations, the topics they gravitate toward.

Persona can shift more — you might run a "founder voice" persona for one set of posts and a "engineering deep-dive" persona for another. Both could share the same Voice DNA.

Why this separation matters

If you bundle voice and persona into one blob, you lose the ability to:

Run multiple personas for the same brand without retraining voice every time.
Refresh your voice as you evolve as a writer, without losing your persona.
Catch which layer is actually causing a problem when output goes wrong.

Bolta keeps them separate by design. Use the separation.

New to Bolta agents overall? Start with Mastering Bolta Agents — it covers presets, autonomy, and blueprints before you dive into Voice DNA specifics.

Lesson 3: Why "Friendly and Professional" Isn't Enough

Most "AI for social" tools ask you for a tone — "friendly and professional," "casual," "expert" — and then generate posts using that label. The output sounds generic because the input is generic.

Why it fails

"Friendly" and "professional" describe nothing specific. Two perfectly friendly writers can sound nothing like each other.
Tone is one of four layers. Skipping the other three (vocabulary, structure, patterns) leaves the model to guess — and it'll guess generic.
Without examples, the system has no way to ground "friendly" in your version of friendly vs everyone else's.

What works instead

Examples beat descriptions. Ten samples of your writing tell the system more than 1,000 words of "describe your voice."
Specifics beat generalities. "I never use exclamation points" is a usable rule. "Be enthusiastic but not over the top" is not.
Constraints beat permissions. Tell the system what to avoid, not just what to do.

Common myths

"AI will always sound generic." Not when given the right inputs. The bottleneck is almost always the training data, not the model.
"My voice is too unique to capture." Probably not. You'd be surprised how much of "your voice" is recoverable from 20-30 samples.
"I'll lose authenticity if AI writes for me." You won't, if the voice is properly captured. The agent is reproducing patterns you already use, not inventing new ones.

Action Steps

Open your current voice profile. Was it trained on samples, a connected account, or a description? Note the answer.
Read 5 recent agent-written posts. For each one, mark whether it sounds like you (✓) or sounds generic (✗). Count the ✗'s.
If more than 1 in 5 sounds generic, your voice training is under-fed. Plan a refresh — we'll cover the right way to do it in the next two lessons.

Voice DNA done well is invisible. Voice DNA done poorly is painfully obvious. The next lesson digs into how to train it from samples — the highest-fidelity option.

What Voice DNA Is (and Isn't)