Your voice on LinkedIn.
Not a copywriter's.
A 30-minute meeting becomes a LinkedIn post you'd actually want to post. In your voice. Not the AI version of you. No blank-page prompts, no “leveraging synergy in today's fast-paced business environment.”
Drop transcripts. Skip the questionnaire.
The 20-question wizard works. Transcripts work dramatically better. One 30-minute meeting transcript populates more of your identity file than the wizard ever could. Voiceprint reads it twice — once for the post material, once for the voice tics it catches and folds back in.
Variety beats volume. 5 varied transcripts > 25 of one type.
Six posts in three days.
All written by Voiceprint from real artifacts (Fireflies meeting transcripts, client L10s, training calls). All published under Ruben's name on LinkedIn. Drafts went through edits — and those edits got fed back into the identity file, so each round shipped with fewer corrections than the last. The voice tunes itself in real time.
Then I started showing her the things I've been building.
That's how a client opened our marketing Level 10 meeting last month.
He sent one last month to the founder of an eight-figure SaaS company. Two hours later: “100%, let's meet up.” No emoji. No throat-clearing.
A copywriter on my team was telling the group why she was stuck on an AI foundations course. The course assumed a baseline tech vocabulary she didn't have. Every video had her pausing with her husband to Google acronyms.
A young filmmaker came into our pipeline last month. He wants to scale to 2-4 commercial projects a month. Wants the cash flow to fund the feature films he actually cares about.
Ask it to write a LinkedIn post about leadership. It says no. Ask it for a thought leadership piece on the future of B2B marketing. Again, it says no. Ask it to “generate content about EOS.” It says no (politely, like a Catholic school librarian) and tells you to come back with a transcript.
Most AI LinkedIn tools start from “give me a topic.”
Voiceprint starts from “give me what already happened.”
Hallucinated thought-leadership about leveraging synergy in today's fast-paced business environment. Em dashes. Bullet lists of 5 tips. “Here's the thing nobody tells you about leadership.”
Source-grounded posts in your actual voice. Pulled from your Fireflies transcript, your weekly review, the client moment that happened on Tuesday. Banned phrases caught before they ship. Voice protected.
Voice fingerprint, banned phrases, gold-standard reference.
The system reads three things on every run: who you are, what you refuse to sound like, and what good output actually looks like in your voice. The model defers to those rules every time. That's why two SMEs in the same job title can run the same workflow and sound nothing alike.
Who you are, in your words
Your audience, your voice pillars, the verbatim phrases you actually say. Pulled from real conversations, not invented. The model can't generate voice details that aren't in this file.
What you refuse to sound like
AI-slop banned phrases, your personal close-cousins, em dashes (the biggest tell). Every draft passes through the rules check before you see it.
Gold-standard reference
Side-by-side BAD vs GOOD outputs in your voice. The model matches the GOOD shape; the BAD examples are training-by-counterexample. Three populated identities ship in the folder.
Drop a transcript. Voiceprint reads it twice.
First pass for the post material you asked for. Second pass for the voice tics, signature phrases, and AI-slop cousins it spots inside your own words. New patterns get logged, cross-referenced across transcripts, and surfaced for your approval. Approve them and they fold into your identity file automatically. The voice gets sharper without you having to maintain it.
Every transcript, twice
Once for the artifact moments. Once for the voice signal. Openers, transitions, coined phrases, banned cousins, archetype moves. Each candidate logged with source pointer and example sentence.
Cross-transcript repeats
A phrase in one transcript is interesting. A phrase in three transcripts (sales call + L10 + testimonial) is a signature. Cross-transcript repeats get flagged as high-confidence additions.
You stay in the loop
When candidates accumulate, Voiceprint surfaces them: approve, reject, refine. Approved tics ship into your identity file. The voice tunes itself in real time. You never have to remember to update it.
Three steps. Fifteen minutes. Then run it weekly.
Build your identity
Drop in 1-5 transcripts (testimonials, L10s, client calls, parody clips) and Voiceprint populates your identity file from the source material in minutes. No transcript handy? The 20-question wizard works as a fallback.
Drop in an artifact
A meeting transcript, a weekly review, a client moment, a V/TO entry. Anything that already happened. The system refuses to write from a blank topic.
Get posts in your voice
Three to five posts per artifact, ready for review. Voice check runs on every draft. You publish, edit, or kill — the system learns either way.
Ready to build your voice?
Twenty questions. One markdown file. A LinkedIn rhythm that finally sounds like you.
Start the intake →Under the hood: a folder.
Voiceprint is built on Jake Van Clief's ICM (Interpretable Context Methodology). Each file does one job. The folder drops into any Claude project and runs cold. Maya, Dale, and Sarah are populated identity examples that prove the structure travels across niches.
voiceprint/
├── README.md ← cold-start in <5 min
├── identity.md ← who you are (generate this here)
├── rules.md ← banned AI-slop, voice check, Rule 11
├── voice-tics-pending.md ← new tics caught from your transcripts
├── examples.md ← what good looks like
├── reference/
│ ├── intake-interview.md ← 20 questions (fallback)
│ ├── post-formats.md ← 6 formats + hooks
│ ├── source-protocols.md ← artifact → post
│ └── identity-examples/
│ ├── maya-fulcrum.md (fictionalized)
│ ├── dale-second-seat.md (fictionalized)
│ └── sarah-whitcomb-negotiation.md (fully fictional)
└── workflows/
├── transcript-to-posts.md
├── vto-to-30-day-calendar.md
└── weekly-review-to-post.md