--- title: "Voiceprint — AI That Writes In Your Voice | Lucid Labs" description: "Voiceprint analyses your own sent mail, Teams messages and meeting transcripts to build a drop-in voice profile, so AI drafts sound like you. Consent-first and sensitivity-label aware." url: "https://lucidlabs.com.au/products/voiceprint" --- SaaS product Live on the Microsoft commercial marketplace # Voiceprint AI that writes in your voice, because it read your writing first Voiceprint builds a linguistic profile from your own sent mail, Teams messages and meeting transcripts, then gives you a drop-in prompt that makes any AI assistant write the way you actually write. Your people approve their own analysis, and the corpus is deleted once it is done. [Ask for a Voiceprint walkthrough](https://lucidlabs.com.au/#contact)[All products](https://lucidlabs.com.au/products) ## Everyone can tell when the email was written by AI The tells are consistent. Three tidy bullet points where you would have written a sentence. An opening line that restates the question. A closing that offers further assistance nobody asked for. It is competent, and it is obviously not you. So the draft gets rewritten. The time the assistant was meant to save goes into making its output sound less like an assistant, and people quietly stop using it for anything that has their name on it. The gap is not capability, it is evidence. A model asked to write professionally has no idea what you sound like, and a style guide written from memory describes how you think you write rather than how you do. People are poor witnesses to their own habits. Voiceprint closes that gap with the writing already sitting in Microsoft 365. It reads what someone has genuinely sent, measures the patterns, and produces a profile any assistant can use. The result is a draft that needs editing for content, not for voice. ## How it works 1. ### The subject approves their own analysis Nobody is profiled because an administrator decided so. The person signs in, reads what will be read, and grants or declines scope by scope. That grant expires, and they can withdraw it at any time. 2. ### Voiceprint reads what they have written Sent mail, Teams chats and channel posts, meeting transcripts. Read-only, in your tenant, on their delegated permission. Anything carrying a Confidential or higher sensitivity label is skipped before it is read, and counted so you can see how many. 3. ### It measures, then writes the profile Sentence rhythm, vocabulary, punctuation habits, and how the register shifts between a Teams message and an external email. Those measurements become a drop-in prompt, plus a report showing the evidence behind it. 4. ### The corpus is deleted Once the profile exists, the raw messages are purged. What remains is the aggregate profile and a handful of short, redacted examples — enough to anchor the voice, not a copy of the mailbox. ## What you get ### A profile that travels One analysis produces a drop-in prompt, artefacts for Claude Code, and a Microsoft 365 Copilot agent. Use the assistant you already use — the voice comes with you. ### Register, not just style How someone writes in a Teams message is not how they write to a customer. Voiceprint keeps those apart and tells the assistant which one the task in front of it calls for. ### Fidelity you can inspect Every profile ships with a measured fidelity score, how it was calculated, and what it cannot tell you. A number without its caveat is worse than no number at all. ### Drift detection People change how they write. A weekly check compares recent messages against the profile and tells you when it is worth rebuilding, rather than leaving you to notice. ## Pricing Voiceprint is sold through the Microsoft commercial marketplace, so it draws down your Azure commitment and bills on your existing Microsoft invoice. A free trial is available. ### Standard $1,000AUD / month A full voice analysis across your sent mail, Teams messages and meeting transcripts, your drop-in voice profile, and the complete communication voice report. Re-run your analysis whenever your voice evolves. List price on the Microsoft commercial marketplace, excluding GST. [View the Voiceprint listing](https://marketplace.microsoft.com/en-au/product/lucidlabsau.voiceprint). ## Built on - Microsoft 365 - Microsoft Graph - Microsoft Purview - Azure AI Foundry - Azure Container Apps ## Reading someone’s writing is a serious thing to ask Voiceprint is built on the assumption that the person being analysed is the one who gets to say yes. Each control below exists because the alternative was not acceptable, not because a framework asked for it. ### Consent belongs to the subject The person being profiled grants it themselves, from their own sign-in, scope by scope. It expires after twelve months and can be withdrawn at any point — including part-way through a run, which then stops before reading anything. ### Sensitivity labels are obeyed Microsoft Purview labels are read before the content is. Anything at Confidential or above is skipped and counted, an unrecognised label is treated as sensitive, and encrypted content is never decrypted. ### The corpus does not linger Raw messages exist only for the length of the analysis and are deleted at the end of it. Subscription data is purged within twenty-five days of termination, ahead of the thirty-day commitment. ## Common questions Does the person being analysed have to agree to it? Yes. Voiceprint cannot profile someone who has not granted consent from their own Microsoft 365 sign-in, and an administrator cannot grant it on their behalf. The grant is scoped, expires after twelve months, and can be withdrawn at any time — a run already in progress stops before it reads anything. What happens to the emails and messages after the analysis? They are deleted. Voiceprint reads the communications in order to measure them, then purges the raw corpus once the profile is built. What is retained is the aggregate profile and a small number of short, redacted excerpts used to anchor the voice. How do you know the voice profile is actually accurate? Every profile is scored by comparing generated drafts against the person’s real writing, using a blind judge alongside two objective statistical measures. The score is published with the profile, together with how it was calculated and where it is less reliable. The measures are weaker on informal channels such as chat than on email, and Voiceprint says so rather than hiding it. Do we need Microsoft 365 Copilot to use Voiceprint? Not for the analysis or the voice profile, which work on their own. A Copilot licence is required only if you want to install the Voiceprint agent inside Microsoft 365 Copilot, because that agent runs under your Copilot entitlement. Where does the data go? It stays inside Microsoft. Voiceprint reads Microsoft 365 through Microsoft Graph with read-only permissions, and runs its analysis on Azure infrastructure in Australia. Communications are not sent to a third-party AI provider. ## Hear what your own voice profile sounds like The quickest way to judge Voiceprint is on your own writing. We will walk you through a real profile and the report behind it, then you can decide whether the drafts sound like you. [Ask for a Voiceprint walkthrough](https://lucidlabs.com.au/#contact)