How voice learning works
The longer you use SparkVox, the better Sparky knows your voice - and how Voice Confidence tracks it.
The longer you use SparkVox, the better Sparky knows your voice. Sparky builds a structured voice profile from your transcripts: tone, sentence style, vocabulary, hooks, banned words, and how you open posts. Posts sound like you because they are built from what you actually said - not scraped from LinkedIn or guessed from a template. Source material and professional perspective from onboarding still shape the angle of each post.
When your profile updates
- After every project transcript is stored (including transcript file uploads).
- For Host, Guest, and Advisor projects, only your selected speaker lines are analyzed.
- When you confirm a post with Schedule or Schedule for now after editing - Sparky may add style correction rules from the diff.
- When edit rate is high and rising, Sparky may cluster repeated edits across recent posts and auto-apply voice profile rules (voice regression detection). See sparkvox.io/help/voice-health-panel.
- When you discard drafts, Sparky records which topics and post styles you rejected and may deprioritize similar moments on future projects.
Draft Feedback in the Voice Profile drawer
Open the Voice Profile drawer from the top bar and open the Draft Feedback tab to see edit-intensity counts, a proof ledger (facts you reinforced by hand), on-brand patterns, and learned corrections - each tied to a source post when available. The Analytics tab shows a theme map: publish vs edit vs discard rates by topic and funnel column.
Voice Confidence score
The top bar shows Voice Confidence as a percentage. It measures how much signal Sparky has across four sources - not how good your LinkedIn posts are. Open the Voice Profile drawer to see each slice: Brief Sparky, Transcripts, Your posts (approvals), and Analytics.
The total is capped at 99% - Sparky is always learning. Re-processing the same project does not add transcript credit again. Deleting a project does not remove voice profile patterns already learned.
Four slices (10 / 40 / 40 / 5)
| Slice | Max | How it grows |
|---|---|---|
| Brief Sparky | 10% | Filling in your brief (Settings → Brief Sparky) |
| Transcripts | 40% | +2% per distinct processed transcript (20 transcripts to max) |
| Your posts | 40% | Approving drafts as-is or after edits (diminishing per approval) |
| Analytics | 5% | +0.5% per published post with synced LinkedIn performance (10 posts to max) |
If your score dropped after a recalibration, transcript credit is now flat (+2% each) instead of front-loaded. Approving posts and publishing through SparkVox so Post Analytics can sync is the fastest way to recover.
Typical transcript-only path
Transcript credit alone tops out at 40%. A brand-new account with one processed project is often around 2% from transcripts plus whatever Brief Sparky credit you earned. Ten transcripts without approvals land around 20% transcript credit - not 80%. The full score needs approvals and published posts with analytics.
| Transcripts counted | Transcript slice only |
|---|---|
| 0 | 0% |
| 1 | 2% |
| 5 | 10% |
| 10 | 20% |
| 20 | 40% (max) |
How profiles merge
Each new transcript strengthens patterns Sparky already knows, adds new ones, and drops anything clearly contradicted. Your profile is never replaced unless you reset it from the Voice Profile drawer in the top bar.
Voice Health and regression detection
Open the Voice Profile drawer from the top bar (Voice Confidence). Voice Health tracks your 30-day approval rate and edit rate trend. When edits are rising, Sparky may run regression detection: it compares your last edited posts, clusters what you keep changing (tone, structure, vocabulary), and writes rules into your voice profile automatically. See sparkvox.io/help/voice-health-panel.