AI Email Assistant FAQs: Privacy, Cost and Whether AI Can Send Without You
Google and Microsoft both say your email is not used to train their AI models — but the honest answer to most email-assistant questions depends on your account.
Google and Microsoft both say your email is not used to train their AI models — but the honest answer to most email-assistant questions depends on your account.
It is a fair moment of hesitation. The assistant offers to summarise a thread or draft a reply, and a small voice asks: what exactly can it see, who else sees it, and could it ever send something in my name? Both companies answer most of that in their privacy pages. This FAQ works through the six questions readers ask most, using only official statements. New to these tools? Start with our pillar guide to AI email assistants explained.
Key points
- Google says Gemini in Gmail does not use your content to train its AI models, and does not store your prompt or its output without your permission; Microsoft says Copilot prompts, responses and Graph data are not used to train its foundation models.
- Neither assistant sends email on its own: both produce drafts, and sending stays yours.
- Access follows your existing permissions, work accounts sit under your organisation’s admin, and both tools can be turned off.
Does the AI read all of my email?
It uses content to answer your prompts — within the access you already have.
Not in the way the worry imagines. Google says Gemini in Gmail “uses your content in Google Workspace to provide more useful responses to your prompts” — the thread you ask about, the messages you ask it to search or summarise. That is processing to answer you, on relevant content.
Microsoft draws the same boundary. For work accounts, Copilot “only surfaces organizational data to which individual users have at least view permissions”. At home it is narrower: Copilot “only makes use of the content in the file that you’re working in or content in another file that you ask it to look at”.
Is my mail used to train the AI models?
Both companies say no for the in-email assistants — with one Google caveat.
Google’s commitment is direct: Gemini in Gmail “doesn’t use your content to train or improve Gemini or other generative AI models”, and Google does not use Workspace data to train the underlying models outside Workspace without permission. Your content is also not used for advertising targeting. Microsoft’s wording matches: “Prompts, responses, and data accessed through Microsoft Graph aren’t used to train foundation LLMs.”
The caveat matters, though. Google warns that if you share data with the separate Gemini Apps or Search services for Personal Intelligence, or through screen actions such as screenshots, that data “will be processed according to their respective terms and policies, and may be used for model training and improvement.” The assistant inside Gmail and the separate Gemini app are not the same product.
Can Gemini or Copilot send email without me pressing send?
No — both stop at a draft, and the sending step stays yours.
Plain answer first: in Gmail, “Help me write” generates or refines a draft in your compose window; you edit it and you send it. Google applies the same confirmation principle across Gemini in Gmail: your emails are “never deleted, archived, labeled, or marked read or unread until you explicitly confirm”.
Outlook works the same way: Microsoft describes Copilot as giving you drafts and summaries to review rather than fully automating the work, and the documented flow ends with you keeping the draft and pressing Send yourself — see our guide to using Copilot in Outlook step by step.
Who can see the AI’s summaries of my email?
A summary is made for you, inside your permissions — but a work account is never only yours.
A summary inherits the access of whoever asked for it. Copilot can only build one from data you may already view, so it cannot summarise a colleague’s private thread into your inbox. Google adds that Gemini does not store your prompt or its output without your permission.
The honest work-account nuance sits with your employer, not strangers. Microsoft says Copilot interaction data is stored under your organisation’s Microsoft 365 commitments, and administrators can view and manage it with Content Search or Microsoft Purview, and set retention policies. Normal workplace governance, not a leak.
What does it cost — who gets it free, and who needs a plan?
Access follows your plan and account type; neither tool is simply free for everyone.
We are not quoting prices here, because prices change and only the official plan pages are a safe source. What they state is the shape of access. Google says Gemini in Gmail requires an eligible plan: work and school accounts through an eligible Google Workspace plan, personal accounts through a Google AI plan — Google AI Plus, Pro or Ultra.
Microsoft ties Copilot in Outlook to account type: work or school accounts, and Microsoft accounts using outlook.com, hotmail.com, live.com or msn.com addresses. Its home page covers use under Microsoft 365 Personal, Family, Premium or Pro plans. The check is not “is it free?” but “does my plan include it?”
Can I turn the AI off?
Yes — in Gmail through smart-feature settings; at work, your admin holds some switches.
In Gmail, go to Settings, then “See all settings”, open the “General” tab, scroll to “Google Workspace smart features” and choose “Manage Workspace smart feature settings”. Turn off “Smart features in Google Workspace” — the setting covering Gemini summaries and drafts — and save; with it off, those experiences stay unavailable. A separate setting covers Smart Compose and summary cards. Our Gemini in Gmail guide shows the features this setting controls.
For Copilot, Microsoft’s home page gives personal users a linked switch: Copilot in Outlook depends on the “Experiences that analyze content” and “All connected experiences” privacy settings — turning off either disables it. At work, your organisation’s settings decide.
What this means
The promises are stronger than most people expect — and narrower than the headlines suggest.
Read the official pages with fresh eyes and a surprising picture emerges. The thing most people fear — “my emails are being fed into the model” — is what both companies rule out most clearly, in writing, for the assistants inside Gmail and Outlook. The thing fewer people think about — which product, which account, which screen you pasted a thread into — is where the real differences hide. Privacy here is not a yes-or-no property of “AI email”. It is a property of the exact door you walked through.
That is why the personal-versus-work question deserves more attention than the training question. A work account gives you the strongest written protections and an employer whose administrators can see and retain interaction data under company policy. A personal account gives you direct control of the switches, but its protections depend on which experience you are actually in — and Google itself flags that sharing data with the separate Gemini Apps can move it under terms where it may be used for training and improvement. Neither setup is “the private one”. They are private in different directions, and knowing which direction matters to you is the real decision.
In our view, the sending question is the one to relax about, and the reading question is the one to stay awake for. Both assistants stop at drafts; nothing leaves your outbox without your hand on it, and the genuine risk in drafting was never a rogue send — it is you, tired at 6 p.m., sending an unreviewed draft that states something you never meant. The quieter risk is that summaries feel so convenient you forget they are generated from your mail, inside systems whose settings you have never opened.
So do the small, boring things, because they are the ones that work. Open the smart-feature settings once and choose deliberately, instead of inheriting defaults. Keep sensitive threads in the account they belong to, and never paste work mail into a personal AI chat for convenience. Read every AI draft once, out loud if it matters. Do that, and these assistants become what they should have been all along: a faster pair of hands on your inbox, with your judgement in charge.
Bottom line
Use the assistants, check the account you are in, and keep your hand on send.
On the official wording, your mail is not used to train Gemini or Copilot, summaries stay inside your permissions, and neither tool can send without you. Your homework is smaller: know which account you are on, set the switches as you want them, and review every draft.
Related guides
Read next:
- AI Email Assistants Explained: Gmail and Outlook
- How to Use Gemini in Gmail Step by Step
- How to Use Copilot in Outlook Step by Step
Sources
