Training Claude to Draft Emails That Actually Sound Like You
How to build a Claude scheduled task that learns your email voice and reply templates, then auto-drafts responses in your inbox.
What is an AI email voice profile?
An email voice profile is a written summary of how you communicate in writing: your greetings, sign-offs, tone, sentence length, and the phrasing you default to in recurring situations. Instead of asking an AI model to guess your style each time, you extract it once from your own sent mail and turn it into a reference document the model reads before drafting anything. Paired with a library of common reply templates (refund replies, invoice chases, meeting requests, polite declines), this profile lets a tool like Claude write drafts that sound like you wrote them, not like a generic assistant.
TL;DR
- A voice profile is built by having Claude read 60 to 90 days of your sent email and extract rules about tone, greetings, sign-offs, and phrasing patterns.
- Alongside the voice profile, Claude can build a template library for recurring email types, things like refund responses, invoice chases, proposal follow-ups, and interview invites.
- These assets get stored outside your local machine, in a tool such as Notion, because cloud-based scheduled tasks can’t reach files sitting on your desktop.
- Once set up, a scheduled task checks your inbox on a recurring basis and drafts replies using the voice profile and matching template as reference.
- The setup process is driven by a done criteria: a clear finish condition (for example, “every email in the last 90 days has been reviewed and summarized into rules”) that tells Claude when the extraction job is complete.
- This approach was demonstrated as one of several scheduled task workflows for small business owners, alongside invoice reconciliation and subscription audits, all built on the same underlying pattern.
- The setup takes roughly 10 to 15 minutes once you understand the connector and storage workaround, and it runs continuously afterward without manual intervention.
How do you build a voice profile from your own inbox?
The process starts by connecting Claude to your email account through a Gmail connector, then giving it a clear instruction: read through your sent mail over a defined window (90 days works well for most people) and extract patterns. You’re not asking it to summarize content. You’re asking it to notice form: how you open emails, how you close them, whether you use first names, how formal or casual your phrasing gets depending on the recipient.
The output is a set of rules, essentially a style guide for your own writing, written in plain language. This becomes a reference document that any future drafting task can read before it writes anything on your behalf.
The second half of the job is spotting repetition in content, not just style. Most people who send a high volume of email are actually sending a small number of email types over and over. Refund and cancellation responses, invoice chases, proposal follow-ups, support answers, interview invitations, and polite declines all tend to repeat with minor variation. Claude can scan for these recurring themes and turn each one into a named template (things like “the polite no” or “the follow-up chase”) with your typical wording built in.
Why can’t the AI just draft emails without this step?
Without a voice profile, an AI drafting tool has to guess your tone from scratch every time, and it defaults to generic, slightly stiff phrasing that reads as obviously AI-written. Feeding it your actual sent mail as source material removes the guesswork. Instead of the model inventing a tone, it’s pattern-matching against real examples of how you’ve handled similar situations before.
This matters most for recurring, low-stakes replies where speed counts more than originality; refund confirmations, scheduling replies, or invoice reminders don’t need a fresh creative approach each time. A template built from your own past emails gets these out the door faster while still sounding consistent with everything else you’ve sent.
Where should the voice profile and templates be stored?
This is the part that trips people up. Cloud-based scheduled tasks run on Anthropic’s infrastructure, not on your laptop, which means they cannot read files sitting in a local folder on your desktop. If your voice profile and templates live only in a folder on your machine, a scheduled task has no way to reach them.
The workaround is to move that reference material into a connected app that the scheduled task can access remotely. Notion is a practical option here: you create a shared space for your voice profile, your templates, and any other business context (contact lists, brand voice guidelines), and grant the scheduled task access to that space through a connector. When the task runs, it pulls the voice profile and matching template from Notion rather than from a local file. This also makes the setup easier to share across a team, since permissions can be managed at the folder level rather than tied to one person’s laptop.
Built like a system. Not vibe-coded.
Remy manages the project — every layer architected, not stitched together at the last second.
How does the scheduled task actually use these assets?
Once the voice profile and template library exist somewhere reachable, the scheduled task itself is straightforward to define. You set a recurring check (hourly, twice a day, or however often makes sense for your inbox volume) with instructions to scan incoming email, identify which ones match a known template category, and draft a reply using both the matching template and the voice profile as reference.
The done criteria for this kind of task is simple: every relevant email in the inbox either gets a drafted reply based on an existing template, or gets flagged because it doesn’t match any known pattern and needs a human response. That second category matters. The goal isn’t to auto-send everything; it’s to remove the drafting effort for repetitive replies while leaving judgment calls to you.
Is this worth setting up compared to just writing replies manually?
For anyone sending a high volume of repetitive email, the time saved compounds quickly. The setup itself takes about 10 to 15 minutes once the connector and storage workaround are in place, and after that the task runs on its own schedule without further input. The alternative, drafting the same five or six email types by hand every day, adds up to real hours over a month even if each individual reply only takes a minute or two.
The tradeoff is that this only pays off if your email volume actually has repeatable patterns. If most of your email is genuinely unique and context-heavy, a template library won’t help much, and you’re better off keeping AI assistance limited to tone-matching on a case-by-case basis rather than building a full template system.
Frequently Asked Questions
What is the difference between a voice profile and an email template?
A voice profile describes how you write in general (tone, greetings, sign-offs, sentence style). A template is a specific pre-written response for a recurring situation, like a refund confirmation or invoice chase. The voice profile shapes tone across everything; templates handle recurring content.
How much sent email history do you need to build a good voice profile?
Around 90 days of sent mail gives enough volume to spot real patterns without the profile becoming outdated. Shorter windows may miss less frequent but still recurring email types.
Can this system send emails automatically without review?
The setup described here drafts replies rather than sending them outright. Emails that don’t match a known template get flagged for manual review rather than answered automatically, which keeps a human check in place for anything unusual.
Why use Notion instead of a local folder for storing templates?
Cloud-based scheduled tasks cannot access files stored locally on your computer. Storing the voice profile and templates in a connected app like Notion means the scheduled task can retrieve them remotely, and it also makes sharing access across a team easier.
Does this require coding or API access?
No coding is required. The setup relies on connectors (integrations that let the AI tool access services like Gmail or Notion) rather than raw API keys, which keeps the process accessible to non-technical users.