AI Voice System
The AI Voice system is the core differentiator of Bloomberry. It builds a persistent linguistic model of how you specifically write — your sentence patterns, vocabulary choices, pacing, tone, and argumentation style — and applies this model to every piece of content it generates.
Generic AI tools write in a generic AI voice. Bloomberry writes in your voice.
Voice Training
Voice training is the process of teaching Bloomberry your writing style by providing actual examples of your writing. These samples are analyzed for linguistic patterns that define your voice.
Adding samples
To add a voice sample:
- Go to Settings → AI Voice → Voice Training
- Click Add sample
- Paste your writing (minimum 50 words per sample)
- Add a title for reference
- Save
You can also:
- Upload file — PDF or DOCX, auto-extracted
- Import from LinkedIn — Connect LinkedIn and import posts directly
Sample requirements
| Requirement | Value |
|---|---|
| Minimum samples to analyze | 3 |
| Minimum total words | 500 |
| Minimum words per sample | 50 |
| Recommended samples | 5–10 |
| Recommended total words | 1,500+ |
| Supported upload formats | PDF, DOCX |
Voice analysis
After uploading samples, click Analyze voice. The system processes your samples and generates a voice fingerprint covering:
- Average sentence length and complexity
- Vocabulary range and preferred word types
- Structural patterns (how you open and close ideas)
- Rhetorical devices you use (questions, lists, contrasts)
- Formality register
- Punctuation patterns
Knowledge Base
The Knowledge Base stores factual context about you, your audience, and your writing preferences. This context is injected into every generation to ground the AI in your real world.
Categories
| Category | Purpose | Example entries |
|---|---|---|
| About Me | Background, expertise, experience | "I founded two SaaS companies", "I have 15 years in growth marketing" |
| Audience | Who you write for, their pain points | "My audience is B2B founders at Seed to Series B", "They struggle with..." |
| Writing Style | Explicit style instructions | "I never use corporate jargon", "I always open with a surprising fact" |
Maximum 20 knowledge base entries. Each entry has a title, category, and content. Entries are surfaced contextually during generation using semantic matching — not injected wholesale.
Voice Tuning
Voice Tuning provides fine-grained control over your voice profile via 9 independently adjustable axes.
| Axis | Range | Description |
|---|---|---|
| Temperature | Conservative → Creative | Controls how closely the AI follows conventions vs. takes risks |
| Directness | Diplomatic → Direct | How bluntly ideas are stated vs. hedged |
| Warmth | Detached → Warm | Emotional tone and personal connection in the writing |
| Wit | Serious → Witty | Use of humor, wordplay, and unexpected observations |
| Energy | Calm → High energy | Pacing, punctuation intensity, and momentum |
| Formality | Casual → Formal | Vocabulary register and structural formality |
| Verbosity | Concise → Elaborate | How much the AI expands on ideas |
| Edginess | Safe → Edgy | Willingness to take controversial positions or challenge consensus |
| Confidence | Tentative → Confident | Use of hedge language vs. definitive statements |
Learning System
AI dialect filter
The AI dialect filter is a post-generation processing step that scans output for linguistic patterns common to AI-generated text and removes or replaces them automatically.
Default blocked phrases include:
delvelandscapelet's unpackat its corelet's dive init's worth notingin conclusionin summaryfurthermoreleverageutilizeparadigmholisticrobustelevateCustom banned phrases
Add your own banned phrases in Settings → AI Voice → Banned Phrases. Any phrase you add will never appear in generated content. Use this to block:
- Competitor brand names
- Internal jargon you've moved away from
- Phrases that don't match your brand personality
- Words you personally dislike
Edit learning
Every edit you make to a generated post teaches Bloomberry. When you change a word, restructure a sentence, or delete a paragraph, this edit is logged and factored into future generations.
Over time, the system learns your correction patterns and self-adjusts. Posts require fewer edits with each session.