Build a Personal Email Style Guide Your AI Can Follow

How to write an email style guide for AI to follow is a question that comes up the moment AI drafting starts saving time but the output feels wrong — too formal, too hedged, or simply not you. The AI is not broken. It is working with defaults designed for no particular person, which produces competent generic text and the wrong result for a message meant to carry your voice and authority.

The fix is a one-time document: a personal style guide that tells the AI exactly how you write, so every draft applies your rules rather than the defaults. This guide covers what to put in that document, how to write it in under an hour, and how to make different AI tools actually follow it.

What goes into a personal email style guide for AI?#

A personal email style guide is a plain-text document — no special formatting required — that answers six questions about how you write. Most people complete a working draft in 45 minutes.

• How do you greet people? First name only? "Hi" vs. "Hello" vs. no salutation at all?
• How does your register shift by relationship? Cold contact vs. warm lead vs. close colleague vs. long-term client
• What is your sentence-length ceiling? Short declarative sentences? No paragraph longer than three sentences?
• What words and phrases do you never use? "Hope this email finds you well," "circle back," "per my last email"
• How do you signal urgency without sounding panicked?
• How do you decline or push back without damaging the relationship?

That is the whole document. A two-page style guide is more actionable than a ten-page one, because the AI follows specific, concrete rules more reliably than vague descriptions of your writing personality.

Before you write a single rule#

Pull five to ten emails you wrote recently that you were satisfied with — messages you would be happy for a client or senior colleague to read as representative of your best work. Read them back to back. You are looking for the patterns your instincts apply without thinking.

Notice how long your sentences run. Do you use contractions? Do you jump straight to the point or acknowledge context first? How do you close — a firm next step, an open offer, or a simple sign-off? Do you use bullet points or prose? These observations are the raw material for your rules.

You are also looking for anti-patterns: the habits AI assistants bring in that are not yours. If every AI draft opens with "I hope you are doing well" and you never write that, it belongs on your banned list. Your style guide is partly a list of permissions and partly a list of prohibitions.

How to build your email style guide, step by step#

Work through these six steps in order. Each produces one section of your finished document.

1. 1

   Write your greeting and sign-off rules

   Name exactly how you address people in each relationship tier: a cold contact you are reaching for the first time, a warm lead you have already spoken with, a close colleague, a long-term client. Write the actual words — "Hi [first name]," or "Dear [first name]," or no salutation at all — not a description of them. Do the same for sign-offs: "Best," "Thanks," "Looking forward to it." These are the first and last words of every message; getting them right is free goodwill.

2. 2

   Define register by relationship

   Register is how formal or casual your vocabulary and sentence structure are. Write one sentence describing the register for each tier. For example: cold contact = professional and direct, no contractions, no humor; close colleague = casual, contractions fine, humor allowed; client = warm but clear, no jargon. When the AI has a labeled register for each tier, it stops averaging across all of them and producing mid-formal text that fits no one in particular.

3. 3

   Set a sentence-length ceiling

   Name the maximum number of clauses in a single sentence and the maximum number of sentences in a paragraph that represent your style. If you write short declarative sentences — average ten words, two-sentence paragraphs — say that explicitly. If you allow longer, more structured prose for context-heavy messages, describe the condition that triggers it. AI assistants default to mid-length compound sentences; without a ceiling, every draft drifts toward them.

4. 4

   List your banned words and phrases

   Write out every phrase you would delete from a draft on sight. Common offenders: "Hope this email finds you well," "please do not hesitate to reach out," "as per our conversation," "circle back," "synergy," "leverage," "touch base." Add your own — the phrases that sound like a language model performing professionalism. The banned list is the most read section of any style guide and the highest-impact change you can make.

5. 5

   Write your escalation and refusal scripts

   Escalation language is how you signal that something is urgent, overdue, or needs a decision now. Refusal language is how you say no without burning the relationship. Both are highly personal and hard for AI to get right by default. Write your actual phrasings: "I want to flag this before the window closes," or "I am going to pass on this one, but here is why I thought of you for it." These are the lines the AI will otherwise write in generic corporate-speak.

6. 6

   Add a structural do-not list

   Separate from the banned-words list, this targets behavioral patterns: do not add context the recipient already has; do not pad a three-sentence reply into six sentences to seem thorough; do not close with a paragraph that re-summarizes what the email just said; do not open with three lines of preamble before the point. Every AI assistant has its own padding habits. The do-not list is where you name yours.

Where to apply your style guide on different AI platforms#

Once you have a style guide, the question is where to put it so the AI actually reads it. The answer varies by platform and by surface within a platform.

One pattern worth noting in the table: every general-purpose AI tool requires you to re-paste your style guide into each new session, or into a configuration surface the AI reads before each response. Per-client variations — say, formal register and no contractions for one contact, casual for another — are not supported out of the box. You get one system-level guide and have to override it manually per message when relationships differ.

What to do when the AI still drifts from your style#

Even with a style guide in place, AI tools occasionally drift — using a phrase you banned, running sentences past your ceiling, or defaulting to a formal register when the message should be casual. There are four causes, each with a different fix.

• The guide is too abstract. A paragraph describing your "warm but direct" register is harder to apply than three concrete example opening lines. Replace descriptions with examples.
• The instruction is buried. If your style guide sits at the end of a long system prompt, the AI may weight it below earlier content. Move the style guide to the top of any context block you use.
• The conversation drifted. In long threads, early instructions lose salience. Re-paste the key rules, or start a fresh session for a new message.
• The platform has a quirk. Gemini in Gmail does not read Gems settings. ChatGPT Custom Instructions do not carry into API calls. Know which surface you are on before you debug.

The most reliable fix across all platforms is the rule of specificity: replace every vague instruction with a concrete example. "Write like I write" gives the AI almost nothing. "Open every cold email with a single sentence that names the benefit to the recipient, not the ask — for example: This is the fastest way to cut your time-to-response on enterprise leads" gives it a pattern it can follow.

A faster way to keep AI writing in your voice#

Re-pasting a style guide into every session is the manual version of this workflow. It works, but it means you are one forgotten paste away from a draft that reverts to defaults — and it gives you no way to handle per-client register variations without manual overrides.

We build AI Emaily, an AI-native email client whose Personal Context brain and per-client profiles are designed exactly for this. You write your style rules once — your register, your banned phrases, your sentence-length ceiling, your greeting and sign-off preferences — and AI Emaily reads them before every draft without re-pasting. You can also create a separate profile for each contact or relationship tier, so the register shifts automatically when the recipient changes. The client drafts; you review; nothing leaves your outbox until you approve.

AI Emaily offers a 7-day free trial — see AI Emaily pricing for plan details. The style guide you built in the steps above is exactly what the Context brain is designed to receive.