How PostKing works:
I create a statistical model of your existing content - blogs, social posts, university assignment anything you've written.
It learns your brand voice, tone, and positioning - not just surface-level writing style, but the actual way you think
Generate content across any channel that sounds authentically like you, not generic AI slop
I take these fingerprints from your content:
Stylometric: - Function word frequency (the, a, and, etc.) — normalized ratios - Lexical diversity: Matthews (type-token), Yule's K, Simpson's - Signature Phrases (TF-IDF): - Top 10 bigrams - Top 10 trigrams - Top 15 distinctive words (length > 3)
Sentiment: - Overall score - Emotional markers: joy, sadness, anger, fear, surprise (normalized) - Sentiment variation (sentence-level deviation)
Patterns: - Collocations: frequent two-word pairs (≥2 occurrences, top 20) - Recurring phrases: three-word phrases (≥2 occurrences, length > 10, top 10) - Structural templates: questions, exclamations, lists, quotations - Rhetorical: - Metaphor density (like/as/is indicators per sentence) - Alliteration frequency (adjacent same-letter starts per sentence) - Parallelism instances (repeated structures per sentence) - Anaphora instances (repeated sentence beginnings per sentence)
It's a simple model, but I hope it's a better one than the "prompt and pray" approach most AI writing tools use. Fundamentally I feel like most AI content tools improve for volume over authenticity, which creates a race to the bottom. That misalignment is what drove me to make this.
Because this is a soft launch, the voice training algorithm is still being refined. You won't get bad output or anything, but the more diverse your input content, the better it captures nuance. I've opened up free alpha accounts in the meantime if anyone wants to test it with their own brand.
Happy to send more credits your way if you use the provide feedback link in the app.
Edit: Happy to share more about the technical approach to voice preservation if folks are interested - it's not just fine-tuning, there's a lot of contextual layer work involved.