Every draft that comes out of audenAI, the brain inside VoiceMoat, used to carry one number: a voice match score from 0 to 100. It no longer does. Since October 2026 every draft gets two separate checks instead, Style and Stance, and they are never blended into one score.

This post explains what a voice match score was trying to measure, why one number was not enough, what the two checks look at, and how to read them on a real draft.

What is a voice match score?

A voice match score measures how closely a draft matches the way you write. VoiceMoat no longer reduces that to a single 0 to 100 number. It checks every draft two ways: Style, how the draft looks next to your own posts, and Stance, whether what it says fits what you actually say. It is not a quality score and not an AI detector.

Conceptually it is the working version of stylometry, the statistical study of writing style used in authorship attribution. Stylometry says a person’s writing has a measurable fingerprint across features like sentence rhythm, vocabulary and punctuation. A voice check turns that idea into something you can see on every draft, against your own baseline rather than a generic one.

The practical value is the same as it always was: it replaces a vibe check with a reading you can audit. Before any voice check, the only way to know if a draft sounded like you was to read it and decide, which does not hold up across many drafts and cannot be handed to anyone else.

Why one number was not enough

The old score counted how a draft looked: sentence length, openers, emoji, casing, length. It never read what the draft said. That gap showed up in testing. A draft that argued the opposite of a view its writer had posted many times still scored around 80 and read as on voice, because the sentences looked right.

That is the failure a voice tool cannot afford. Readers forgive a slightly long sentence. They notice when you suddenly believe something you have argued against for a year. So the check was split into two parts that answer two different questions, and neither is allowed to hide the other.

Style: how the draft looks next to your posts

Style is a count. It runs a set of checks and tells you how many of them match your own posts, for example 4 of 6 checks match. The checks come from the 10 signals of voice audenAI learns when it trains on you:

• Sentence length. How long your sentences usually run.

• Emoji use. How often you use them, if at all.

• Casing. Your capitalisation habits.

• Post length. How long your posts usually are.

• Opener style. The ways you actually open your posts.

• AI tells. Patterns that make writing read as machine-made, such as the “it’s not X, it’s Y” turn or a bait question tacked on at the end.

• Drift. Signs a draft is sliding away from you: fewer contractions, less “I”, longer words than you normally use.

Every check that is off shows a real line from your own posts next to it, so you can see the difference instead of trusting a number. On Twitter, a reply is compared with your past replies, not your posts, because almost nobody writes replies the way they write posts.

Style is free and updates live as you edit. Read it as a nudge on form: if Opener style is off, the fix is usually one sentence at the top.

Stance: what the draft says

Stance reads the draft line by line against your claim bank, the claims you repeat, the stories you reuse and the framings you avoid, all pulled from your own posts. It does not give a score. It gives one of three results:

• On voice. Every line fits what you say.

• Check one line. A line makes a new claim, or tells a story you have not told. Not wrong, just worth a look.

• Off voice. A line goes against one of your claims, or uses a framing you avoid.

Flagged lines are underlined in the draft. Each one shows what was found, a short reason, and the claim it hit, quoted from your post with a link to it. From there you can ask for a rewrite in your own words, ignore the flag for this draft, or tell audenAI the line is you now, which adds it to your claim bank.

Stance never rewrites anything on its own, and a new claim is a question rather than a failure. People change their minds; the check just makes sure you meant to. Running Stance costs one credit; Style is free.

Where the claims come from

When you train or retrain your voice, audenAI reads your own posts and pulls out claims and stories. It keeps one only if its quote is found word for word in a post you published, so nothing in the bank is a paraphrase it made up. New items wait for you to review: keep them, pin them, edit them into your own words, or remove them with “Not me”, which means they are never suggested again.

Two more inputs sharpen it. You can mark the posts that sound most like you (and the ones that were a bad day) so audenAI leans on the right examples. And when you change a draft audenAI wrote and then post, schedule or save it, it learns from what you changed. Once two of your edits point the same way, it suggests a plain rule you can keep.

Sensitive topics such as politics, religion and health stay out of the bank, and out of audenAI’s drafts, unless you turn them on.

How do you read Style and Stance together?

The two checks are useful precisely because they can disagree. The common combinations:

• Style matches, Stance on voice. Ship it. It looks like you and says what you say.

• Style matches, Stance off voice. The dangerous one, and the case one number used to hide. Fix the flagged line before anything else.

• Style off, Stance on voice. The argument is yours, the delivery is not. Usually an opener or a length fix.

• Style off, Stance off voice. Treat the draft as raw material and rewrite it.

Your own read still wins on any single draft. The checks are statistical views of how you write and what you say, not an oracle. What they give you is a way to catch the drafts that need a second look before your readers do.

Is a voice match score the same as an AI detector?

No, and the difference is the whole point. An AI content detector tries to answer “was this written by a machine?”, a question with known reliability problems and high false-positive rates. A voice check answers a different question: is this recognizably you? It does not care whether you, audenAI or a general model produced the words.

That is why it is useful even on drafts you wrote by hand. Paste in your own writing and a Style miss or a Stance flag tells you that piece drifted from your usual voice, which is a real editorial signal a detector could never give you. It is also the core of scaling your voice rather than cloning it: the goal is not to pass as human, it is to stay yourself, with you as the editor who decides what ships.

How do the checks interact with retraining?

Your voice moves over time, and the checks are only as current as your last training. If Style keeps flagging the same habit on drafts you are happy with, your writing has probably moved on. For the structural version of this, when drift is not benign evolution but the audience-optimization pattern that erodes a voice, see voice drift: why most creators lose their edge after 10K followers.

When that happens, retrain. audenAI rereads your latest posts and rebuilds your style profile and your claim bank, and your choices survive: pinned and edited claims stay as you left them, and anything you removed stays gone. We cover when and how often in the dedicated post on voice retraining.

Want to see both checks on your own writing? Try VoiceMoat on a 3-day trial: Style and Stance come with every draft on every paid plan and in the trial. Or read what is audenAI for the product context, and the 10 signals of Voice DNA for the reference on what each Style check is built on. One subtle drift source worth knowing: drafting on different devices produces different registers, which drafting on X across devices, voice-first covers.