AI in your business email: Gmail, Outlook and faster emails

A business inbox can eat up half your day. The same questions about price, dates and availability, long threads you have to make sense of, offers written from scratch. AI in Gmail and Outlook shortens this work, but only when you know what to hand over to it and what to keep from it.

In this guide I show what I find summaries, drafts and templates useful for, how to set up privacy and how to check an email before sending it. I describe the situation as of 8 November 2026. Google and Microsoft change features and button names often, so always compare them with your own panel.

What AI does in email, and what it doesn’t

Both programs offer a similar set of capabilities. Gemini in Gmail (as part of Google Workspace) and Copilot in Outlook (as part of Microsoft 365) can:

• summarize a long thread and list what was agreed and what you are waiting for,
• write a draft reply based on a few words from you,
• improve the tone and style of a message: shorter, more polite, more specific,
• find information in your mailbox, e.g., “when did the customer give the dimensions of the terrace”.

What they don’t do: they don’t know your price list unless you paste it in or they have access to the document. They don’t know what the customer heard from you over the phone. They are not responsible for the content of the email; you are. That’s why I treat them as a fast assistant that writes the first version, not as a secretary I hand my mailbox over to.

Step 1: check what your plan includes

Not every plan has the same AI features in email. The range depends on your Google Workspace or Microsoft 365 subscription and on whether the administrator has turned the option on. I don’t give plan prices here because they change. First, see what you already have in your subscription.

1. Log in to your email and look for the sparkle icon or a Copilot or “Help me write” button.
2. If they aren’t there, ask the person who manages the company account, or look at the provider’s plans page.
3. Check whether the administrator can turn AI on for selected people. In a small company it’s worth starting with one person and testing.

If you use free personal email, the privacy rules are different from those of a business account. Use a business account for company matters.

Step 2: thread summaries

This is the simplest and safest use. You open a long thread that has come back after a week and ask for a summary. In Gmail it’s the “Summarize this email” button at the top of the thread; Outlook has a similar option in Copilot.

After the summary, ask one control question, e.g.:

List in bullet points: 1) what the customer ordered, 2) what dates and amounts were mentioned, 3) what I’m waiting for from them.

Check amounts and dates in the original messages. The summary is meant to save you reading the whole thing, but always verify the numbers.

Step 3: drafts of replies and offers

The best drafts come out when you give facts, not just a topic. Instead of “write a reply to the inquiry”, type:

Reply to the customer politely and briefly. Facts: door installation possible within 2 weeks, the measuring visit is free, the price depends on the measurements. Suggest two dates for the visit. Don’t give any amount.

This instruction has three elements: tone, facts and a prohibition (what not to write). The last element protects you from made-up prices and promises. Adding “if you don’t know something, leave a blank in square brackets” also works well.

I have collected ready-made, detailed instructions to copy in the guide Business prompts: 14 ready-made instructions for ChatGPT and Gemini. You will find templates for replies to quote requests and follow-ups in the guide on replying to a quote request.

Step 4: templates instead of AI where they are enough

Not every email needs AI. Repetitive messages (an order confirmation, a request for invoicing details, the reply “thank you, we’ll get back to you with a quote”) are better kept as a template. Gmail and Outlook have built-in template options; check their names and location in the settings of your version.

A practical split:

A template doesn’t hallucinate, doesn’t need a license and always says the same thing. For a few of the most common customer questions, that is often enough.

What not to paste into AI

This is the most important section. The rule: only what you could show to someone outside the company goes into AI, unless you are certain about the privacy settings and the agreement with the provider.

Don’t paste:

• passwords, codes, access keys and password reset links,
• card numbers, bank account numbers and PESEL numbers (Polish national ID numbers),
• sensitive customer data: health, court cases, children’s data,
• confidential contracts and offers, if you don’t have the customer’s consent,
• whole mailboxes and contact lists to “tidy up”.

When you need help with such an email, replace the data with placeholders: “Customer A”, “amount X”, “date Y”. Then insert the real data by hand. In medical practices and companies that work with health data, it’s safest not to use AI at all for content about specific people.

It’s worth writing down the rules for the whole team on one page. You will find a template in the guide AI use policy for your company.

Step 5: privacy settings

Before the team starts writing with AI, check four things in the account settings and in the provider’s documentation:

1. Whether email data is used to train models. For business plans, providers usually state that it isn’t, but read this in the official documentation for your plan.
2. Who can turn AI on. The administrator should decide who uses it.
3. History of conversations with the assistant. See whether it is saved and how to delete it.
4. Data processing agreement. When your emails contain customers’ personal data, make sure the agreement with your email provider covers the processing. If in doubt, ask the person responsible for GDPR.

This list does not replace legal advice. It only shows what to ask about so you don’t have to guess.

Step 6: checking before you send

AI writes smoothly, so mistakes are hard to spot. Before you click “Send”, go through this list; it takes 30 seconds:

• The recipient’s first and last name are correct, and the form of address fits.
• Amounts, dates and addresses match your documents.
• No promises you are not making, e.g., “we guarantee”, “always within 24 hours”.
• Recipients: check the “To” and “Cc” fields, especially when replying to all.
• The tone sounds like you. If the email is too effusive, cut it in half.
• Attachments are included and correct.

The same applies to fake emails. AI won’t always recognize a scam, and attackers use it too, so when a bank account number changes or an urgent payment comes up, call a number you know.

How to roll this out in your team

In a small company, a two-week plan is enough:

1. Choose one person and 3 types of email that take up the most time.
2. Write instructions for them with facts and prohibitions, as in step 3.
3. After a week, count how many minutes a day it saves and how many times mistakes had to be corrected.
4. Only then extend it to the team and introduce a one-page set of rules.

When there are many repetitive emails and the replies rely on data from other systems (orders, quotes, CRM), it’s worth going a step further and automating the process. Emails from a form can, for example, land in a draft with the facts filled in, and you just approve them. Also remember to configure your domain properly: I describe SPF, DKIM and DMARC in detail in the guide on business emails going to spam.

Summary

AI in email works best for summaries and first versions of replies, while templates handle repetitive messages more cheaply and safely. Don’t paste passwords or sensitive data, and always read an email before sending it. If you would like me to set up automatic replies and drafts or implement AI in your company, write to me. I reply within 24 hours.