Using AI as an email assistant: context, voice, and when to skip it

Executive overview

AI-written emails fail because the AI lacks context — who you're emailing, what you've discussed, and how you write. Treat AI like a brand-new EA: brief it before asking it to write anything.

Two inputs fix most problems: a briefing document (relationship and project context) and a voice and style guide (your tone, sentence length, and personality). Store both in a reusable project or agent so they're available in every chat.

The briefing document plus voice guide, loaded into an AI project, is the entire system.

Why AI gets emails wrong

• AI guesses at context it doesn't have — and guesses wrong
• Without background, it doesn't know if this is a first contact or a three-month relationship
• Even Copilot, which can access your inbox, must be explicitly told to read the relevant email thread
• Asking it to summarize "emails between me and X about project Y" forces it to think out loud before drafting

Building the briefing document and voice guide

• Brief AI like a new EA: who you're emailing, what you've been working on, what the goal is
• A briefing document covers your company, your role, key relationship or project background
• A voice and style guide captures tone (bubbly vs. formal), sentence style (punchy vs. long), and linguistic habits
• Get AI to analyze samples of your own writing — it will extract your voice for you
• The two documents can be combined or kept separate; both must be available when drafting

Setting up a reusable AI email assistant

• Create a project in Claude or ChatGPT — every new chat under it inherits your uploaded documents
• In Copilot, create an agent; in Gemini, create a gem
• A project is preferred over a custom GPT because each chat stays associated with the project history
• In Copilot, prompt it to summarize the relevant email thread first, then paste that summary as additional context into your agent

When to use AI for email — and when not to

• Skip AI for short, transactional emails ("Yes, I'll get it done by Tuesday") — the setup time outweighs the benefit
• Roughly 60% of everyday emails don't need AI
• Use AI for: emails requiring nuance, introductions, sales outreach, customer support, HR FAQs, helpdesk replies
• Best fit is rinse-and-repeat emails: load a Q&A briefing doc, feed it a question, get a polished reply
• For project update emails, pull in thread context first so AI knows the current state of the conversation