Protect My Voice

Voice Pass: Making It Sound Like You

Define your voice profile: formality, humor, emotional intensity, rhythm, vocabulary, signature phrases, banned phrases. The Voice Pass uses a two-stage pipeline: instant deterministic detection of dozens of pattern families, then LLM evaluation that checks each candidate against your own approved chapters before flagging it. Every proposed change shows original, revision, reason, and rule.

Inkloom Voice Pass showing findings with original, revision, reason, and rule
The Problem

Why this is hard

Long manuscripts have problems that short documents do not.

The Problem

AI writing has identifiable patterns: repetitive openings, formulaic contrasts, slogan overload, unearned profundity, uniform paragraph lengths, empty transitions. Left unchecked, your book starts sounding like every other AI-assisted book.

Why It Gets Harder

You cannot catch every instance of a pattern by eye in a 300-page manuscript. And when you do catch one, you need to know whether it is a mechanical failure or your deliberate style. Most tools cannot tell the difference.

How Inkloom Approaches It

You define a voice profile: formality, humor, emotional intensity, rhythm, vocabulary, signature phrases, banned phrases, and examples of what sounds right versus artificial. You write project rules in plain language: no em dashes, no forced triads, name your sources. The Voice Pass runs in three modes (Conservative by default, Balanced, Aggressive) using a two-stage pipeline. First, a deterministic detector scans for dozens of pattern families with frequency thresholds, flagging density and repetition rather than single occurrences. Second, when AI is available, an LLM evaluates each candidate against your voice profile, your active rules, and exemplars from your approved chapters. If a pattern appears consistently in your approved work, it is treated as your style, not a mechanical failure. Proposed revisions must pass a quality bar before they reach you.

What You Control

You accept, reject, or hand-edit each proposed change. Severity is calibrated: severe means it violates an explicit rule you wrote, warn means a mechanical pattern repeated enough to notice, info means borderline and possibly deliberate. Without an AI key, the deterministic detector alone still produces findings as manual-review flags. A global no-em-dashes rule is enforced in every prompt Inkloom sends.

Next in the Inkloom workflow

Review

Whole-book editorial pass across every chapter.

Go to Review

Use AI without giving up your voice.

Let Inkloom find formulaic AI patterns while you decide what actually changes.

The Manuscript Operating System

Your book is bigger than a prompt.

Give it a memory.

Inkloom helps you plan, write, remember, check, polish, typeset, cover, and publish a coherent book from one integrated local-first environment.

$299 one-time purchase. Buy once. It is yours.