Comparisons
Can AI Draft Emails in Your Own Writing Style? A 2026 Guide and Comparison
Yes, AI can write emails in your voice, but only some methods work. timelit, Copilot, Shortwave, Superhuman and Claude compared, plus a five-minute test.
In brief
Yes, AI can draft emails in your own writing style, but the method matters more than the model. A tool you describe your style to produces a competent, generic email with your sign-off attached. A tool that learns from the emails you have actually sent produces something people cannot tell you did not write. timelit drafts your emails automatically in the background, personalised to your writing tone and work style, inside the Outlook you already use. No prompt to write, no new inbox. Hosted in Europe, GDPR-compliant, no data sharing to third parties. The honest ceiling for this technology today is roughly one draft in four good enough to send untouched, and any vendor claiming much more is overselling.
Table of Contents
- 01 The short answer
- 02 The three ways AI tries to sound like you
- 03 What "sounds like me" is actually made of
- 04 How well does this work? An honest number
- 05 Capability matrix
- 06 Where each competitor is genuinely good
- 07 What timelit does differently
- 08 What timelit costs
- 09 Test any of these in five minutes
- 10 FAQ
- 11 Sources
01 The short answer
Yes, AI can draft emails in your own writing style. But an AI that you describe your style to will produce a competent, generic email with your sign-off attached. An AI that learns from the emails you have actually sent will produce something people cannot tell you did not write.
Those are two very different products, even when both are marketed as "writes in your voice". Most of the disappointment people report with AI email tools comes from buying the first and expecting the second.
timelit is the proactive execution layer for Microsoft 365. It drafts your emails automatically in the background, personalised to your individual writing tone and work style, inside the Outlook you already use. There is no prompt to write and no new inbox to adopt: the draft is already there when you open the thread. Users save roughly 9 hours per week, about 53 percent of their administrative time.
02 The three ways AI tries to sound like you
Almost every tool on the market uses one of these three mechanisms. Knowing which one you are buying explains most of the disappointment.
Instruction-based: you describe yourself
You write a set of rules. Keep it short. No "I hope this finds you well". Sign off with "Best". The model follows the rules.
This is what Microsoft 365 Copilot does in Outlook through Draft Instructions, a settings field where you type your preferences, plus the broader Copilot Personalization layer. Your explicit prompt wins, then Draft Instructions, then Personalization. Claude does the same when you describe a custom style instead of uploading a sample.
It works, up to a point. The ceiling is that you have to be able to articulate your own writing habits, and most people cannot. You know your emails when you read them. You do not know that you open 70 percent of them with the recipient's first name and no greeting word.
Sample-based: you paste examples
You give the model two or three pieces of your writing and it infers the pattern.
Claude custom styles work this way: upload a writing example, or pick from the Formal, Concise and Explanatory presets. Grammarly builds a voice or tone profile, and at Business level a company style guide and brand tones that apply across the organisation.
Better than instructions, and it captures things you would never have thought to specify. The limit is that a handful of samples describes how you write in general, not how you write this email to this person about this project.
Retrieval-based: it learns from your sent mail
The system indexes what you have already sent, finds the messages most similar to the one you are answering now, and uses them as live examples.
Shortwave's Ghostwriter is the clearest published example: embedding-based semantic search across your sent mail, combined with system prompts and fine-tuned models, so a reply to a customer question reuses the structure and phrasing from the last time you answered that kind of question. Superhuman's Auto Drafts go further and modulate per recipient, formal with an executive, loose with a colleague. Fyxer and timelit work on the same principle.
This is the approach that actually produces "I would have written that". It is also the approach that requires the tool to read your mailbox, which is where the privacy question starts.
03 What "sounds like me" is actually made of
When people say a draft does not sound like them, they are usually reacting to one of these:
| Layer | Example of what varies |
|---|---|
| Opening | "Hi Anna," vs "Anna," vs straight into the sentence |
| Sign-off | "Best", "Thanks", "LG", nothing at all |
| Sentence length | Short declaratives vs long qualified sentences |
| Formality | The same person writes differently to a client and to a supplier |
| Hedging | "I think we should" vs "we should" |
| Filler | Whether you say "I hope you are well" or find it unbearable |
| Structure | Bullets vs prose, one ask per mail vs several |
| Facts and phrasing | The way you always explain your pricing, your onboarding, your delivery times |
The last row separates real style matching from tone controls. Anyone can turn a formality dial. Reproducing the paragraph you have written forty times about how your onboarding works requires the system to have seen those forty emails.
04 How well does this work? An honest number
The most useful public benchmark comes from Superhuman. During the beta of its auto-draft feature, co-founder Rahul Vohra reported that 40 percent of auto-generated drafts were sent within one day, and 60 percent of those went out with no manual editing at all. That is roughly one in four drafts good enough to send untouched, from a well-funded team using frontier models from both Anthropic and OpenAI.
We take that number seriously rather than claiming to beat it, because it is the right shape for what this technology currently does. Style matching does not mean the draft is finished. It means the draft starts from your voice instead of from a press release, so editing is a ten-second job instead of a two-minute one. Across a day of forty emails, that difference is the whole point.
Two things follow, and any vendor telling you otherwise is selling you something:
- The first weeks are the worst weeks. Retrieval-based systems need volume and correction. Early drafts land near your average register, not your specific one, and they improve as they see more of what you actually send and what you actually change.
- Nothing sends without you. Every serious tool in this category, timelit included, prepares the draft and leaves the decision with you.
05 Capability matrix
| timelit | Microsoft 365 Copilot | Grammarly Business | Shortwave | Superhuman Mail | Fyxer | Claude | |
|---|---|---|---|---|---|---|---|
| Style method | Learns your tone and work style | Instructions you write, or a custom agent you build | Voice profile, brand tones, style guide | Ghostwriter: retrieval over sent mail | Retrieval, per-recipient modulation | Retrieval over sent mail | Uploaded sample or described style |
| Draft appears without prompting | Yes, by default | No | No | No | Yes, Auto Drafts on the Business plan | Yes | No |
| Works in Outlook and Microsoft 365 | Yes, natively | Yes | Yes, as an overlay | No, Gmail only | Yes | Yes | No, separate chat window |
| Requires a new email client | No | No | No | Yes | Yes | No | No |
| Beyond email | Meetings, summaries, scheduling | Word, Excel, PowerPoint, Teams | Any writing surface | Email, Tasklet automations | Email, CRM logging | Email, meeting notetaker | Anything, on request |
| Per-recipient tone shift | Personalised to your tone and work style | Only if you write the rule | Tone profiles, set manually | Single voice profile | Yes, strongest in category | Learns over time | Only if you say so |
| Hosting | Europe | Microsoft cloud, Anthropic-routed features excluded from the EU Data Boundary | US vendor | US vendor | US vendor | US vendor, states SOC 2 | US vendor |
| Third-party data sharing | None | Anthropic is a Microsoft subprocessor for some Copilot experiences | Per vendor terms | Uses OpenAI and Anthropic models | Uses OpenAI and Anthropic models | Per vendor terms | Not applicable |
| Indicative price | 20 to 30 euros per user per month | 18 to 30 US dollars per user per month plus a base licence | Per seat, business tiers | 18 to 45 US dollars per seat per month | 30 to 40 US dollars per user per month | 22.50 to 50 US dollars per user per month | Included in Claude plans |
Competitor details verified against public sources on 28 July 2026. Pricing and features in this category change monthly.
06 Where each competitor is genuinely good
We would rather you pick the right tool than the wrong one with our name on it.
Superhuman Mail has the best per-recipient voice matching available today, and Auto Drafts are proactive in the same sense timelit is: the reply is waiting when you open the thread. If you are willing to move into a new email client and your data can sit with a US vendor, it is excellent.
Shortwave's Ghostwriter is the most transparent implementation of retrieval-based style matching that anyone has published, and at the lower tiers it is the cheapest way to get real voice learning. The catch is absolute: Shortwave supports Gmail and Google Workspace only. No Outlook, no Exchange, no IMAP. If you are on Microsoft 365 it is not an option, whatever else it can do.
Grammarly Business is not really competing for the same job. It is the strongest product for organisational voice: style guides, brand tones and terminology enforcement across a whole team, in every app people write in. If your problem is fifty people writing inconsistently, that is Grammarly, not us. Note that Grammarly acquired Superhuman in 2025 and the parent company rebranded to Superhuman, with the generative assistant now shipping as Superhuman Go.
Fyxer is the closest competitor in shape: an assistant layer over Gmail and Outlook that drafts in your tone and takes meeting notes, without replacing your inbox. Reviewers consistently praise the tone matching and consistently complain about mislabelling in the first weeks. We compared that field in detail in the best Fyxer alternatives for Microsoft 365.
Microsoft 365 Copilot has reach nothing else has: Word, Excel, PowerPoint, Teams, and grounding in your organisation's data through Microsoft Graph. For style specifically it is the weakest of the serious options, because the personalisation is something you configure rather than something it learns. The workaround the community has settled on is telling: ask Copilot to analyse your sent items, have it write a style instruction block, then paste that block into Draft Instructions. That works, and it is a manual process you have to redo as your writing shifts. Our full comparison is in what to use instead of Copilot for Microsoft 365.
Claude produces the best prose of anything on this list, and custom styles from an uploaded sample are genuinely good. It is also not connected to your mailbox. Every email means copying context into a chat window and pasting the result back, which is fine for the five emails a week that matter and unworkable for the other two hundred.
07 What timelit does differently
The draft exists before you go looking for it. Every other approach in this guide, with the exception of Superhuman Auto Drafts and Fyxer, requires you to decide that this particular email deserves AI help, then ask. That decision costs more attention than the drafting does. timelit removes it: the reply is prepared in the background, in your tone, and it is there when you arrive. The mechanics are documented in Email drafting.
It works where you already work. No new inbox, no migration, no second app for your team to learn. timelit runs inside your existing Microsoft 365 interface. The measure of success is that you never open it.
It is not only email. Style-matched drafting is one output. timelit also schedules meetings and summarises discussions automatically, which is where the rest of the roughly 9 hours per week comes from.
Your sent mail stays in the EU. This is the part of style matching nobody advertises: to write like you, a system has to read what you have written. Every retrieval-based tool in this comparison is reading your sent folder. With most of them, that content is processed by a US vendor, often via frontier models from OpenAI or Anthropic. timelit is hosted in Europe and GDPR-compliant, with no data sharing to third parties and no training on customer data. There is no admin toggle that moves your correspondence somewhere else, because there is no other path.
That last point matters more in Microsoft 365 than people expect. Since January 2026 Anthropic has operated as a Microsoft subprocessor for Copilot experiences, and Microsoft's own documentation states those models are currently excluded from the EU Data Boundary. EU, EFTA and UK tenants have them off by default, and switching them on is a decision your Global Administrator has to make consciously.
08 What timelit costs
| timelit Pro | timelit Teams | |
|---|---|---|
| Annual billing | 20 euros per user per month | 24 euros per seat per month |
| Monthly billing | 25 euros per user | 30 euros per seat |
| Base licence required on top | No | No |
| Free trial | 7 days, no credit card | 7 days, no credit card |
Prices include VAT. There is no add-on licence to buy underneath, because timelit runs on the Microsoft 365 you already have. If you want the arithmetic for your own team size and hourly rate, the instant estimate calculator does it in about a minute.
09 Test any of these in five minutes
Before you buy anything on the strength of a demo, run this. It works on every tool in this guide, including ours.
- Pick five emails from your sent folder that you actually wrote, from the last month, covering different recipients: a client, a supplier, a colleague, a stranger, someone you know well.
- Feed the incoming message from each thread to the tool and let it draft the reply.
- Compare against what you really sent, and count only three things: the opening, the sign-off, and whether any sentence contains a phrase you would never use.
- Score each draft as send as is, quick edit, or rewrite.
- Repeat after two weeks of real use.
If the ratio of send as is to rewrite has not improved between the two runs, the tool is not learning, it is templating. That is the single most useful signal available to you, and it takes less time than reading a review.
10 FAQ
Can AI draft emails in my own writing style?
Yes. The tools that do it well learn from the emails you have already sent rather than from a description you write. timelit drafts emails personalised to your individual writing tone and work style, automatically in the background, inside Microsoft 365. Shortwave and Superhuman do the same for their own email clients, and Fyxer does it as a layer over Gmail and Outlook.
How does AI learn my writing style?
Three ways. It follows written instructions you provide, it infers a style from writing samples you upload, or it retrieves the most relevant messages from your sent mail and uses them as live examples. The third method produces the closest match, because it captures not just your tone but the specific phrasing you use for recurring situations.
Can Microsoft 365 Copilot write emails in my voice?
Partly, and only if you configure it. Outlook has a Draft Instructions setting where you describe your preferences for tone, length, greeting and sign-off, and Microsoft 365 Copilot Personalization applies more broadly. Your explicit prompt takes priority, then Draft Instructions, then Personalization. Copilot does not build a voice profile from your sent mail on its own. The common workaround is to have Copilot analyse your sent items, produce a style instruction block and paste it back into the settings.
Does Claude write emails in my writing style?
Claude can create a custom style from an uploaded writing example or from a description, alongside the Formal, Concise and Explanatory presets, and the output quality is excellent. It is not connected to your mailbox, so every email requires you to bring the context to Claude and carry the result back.
Is Grammarly good for writing emails in my voice?
Grammarly is strongest at organisational voice: style guides, brand tones and terminology consistency across a team, applied everywhere people write. It is designed to keep writing on-brand and correct rather than to reconstruct one individual's habits from their real correspondence. Grammarly acquired Superhuman in 2025 and the parent company now operates as Superhuman.
What percentage of AI drafts can I send without editing?
Superhuman reported that during its auto-draft beta, 40 percent of drafts were sent within a day and 60 percent of those went out unedited, so roughly one in four needed no changes. Treat any claim far above that with suspicion. The realistic benefit is not zero editing, it is starting from your voice instead of a blank page.
Do I have to let AI read my email to match my writing style?
For genuine style matching, yes. A system cannot reproduce phrasing it has never seen. The question worth asking is not whether it reads your mail but where that content is processed. timelit is hosted in Europe and GDPR-compliant, with no data sharing to third parties and no training on customer data.
Will timelit send emails without me?
No. timelit prepares the draft in your tone, before you ask, and you decide what goes out.
Does timelit work with Outlook?
Yes. timelit works directly within your existing Microsoft 365 interface. There is no new inbox and no migration.
Does Shortwave work with Outlook?
No. Shortwave supports Gmail and Google Workspace only, with no Outlook, Exchange or IMAP support. For Microsoft 365 users it is not an option.
How long before AI drafts sound like me?
Expect the first weeks to need correction. Retrieval-based systems improve as they see more of what you send and what you change. Run the five-email test at the start and again two weeks later: if the ratio of usable drafts has not moved, the tool is templating rather than learning.
What is the difference between tone settings and real style matching?
A tone setting shifts a dial: more formal, shorter, friendlier. Real style matching reproduces your specific habits, your openings, your sign-offs, your sentence rhythm, and the way you have explained the same thing forty times before. Only tools with access to your sent mail can do the second one.
How much does timelit cost?
timelit Pro is 20 euros per user per month on annual billing or 25 euros monthly. timelit Teams is 24 euros per seat per month on annual billing or 30 euros monthly. Prices include VAT, there is a 7-day free trial with no credit card required, and no base licence to buy on top because timelit runs on the Microsoft 365 you already have.
11 Sources
All external claims verified on 28 July 2026.
How style matching works
- Shortwave, "Introducing Ghostwriter: AI writing that learns from you": shortwave.com
- ZenML LLMOps Database, "Building an AI-powered email writing assistant with personalized style matching": zenml.io
- Shortwave docs, "The Shortwave AI Assistant": shortwave.com
Superhuman
- TechCrunch, "Superhuman's new auto-draft feature almost makes me like AI replies", 14 July 2026: techcrunch.com
- Superhuman Help Center, "Auto Reminders and Auto Drafts": help.superhuman.com
- Fastio, "Superhuman AI review 2026": fast.io
Microsoft 365 Copilot
- CIAOPS, "Outlook Draft Instructions vs Microsoft 365 Copilot Personalization": blog.ciaops.com
- Computerworld, "How to use Microsoft Copilot for writing in Microsoft 365": computerworld.com
- Microsoft Learn, "Anthropic models in Microsoft Online Services": learn.microsoft.com
Grammarly and Claude
- Grammarly, "Grammarly for Business": grammarly.com
- Anthropic, "Tailor Claude's responses to your personal style": anthropic.com
Shortwave and Fyxer
- CMDK, "Shortwave email review 2026": cmdk.email
- CMDK, "Fyxer AI review 2026": cmdk.email
- this+that, "Fyxer AI review 2026": thisandthat.chat
timelit product claims (proactive background drafting, personalisation to individual writing tone and work style, European hosting, GDPR compliance, no third-party data sharing, no training on customer data, roughly 9 hours per week or 53 percent of administrative time, timelit Pro and timelit Teams pricing) are timelit's own.
Superhuman shipped a new auto-draft implementation in July 2026 and this category moves fast. As of: July 2026.
Want to see what your own voice looks like coming back at you? Start your 7-day free trial, then run the five-email test above on your real sent folder.