TL;DR
Since January 2026, Gemini (running Gemini 3) is built into Gmail and turned on by default. Three features are now free for everyone: AI Overviews (thread summaries), Help Me Write, and Suggested Replies that match your style. Two are Google AI Pro / Ultra only: asking your inbox questions in plain language (“ask your inbox”) and Proofread. The skill is letting the machine sort, summarize, and draft from your bullet points, while you keep the voice human on the replies that matter. With the prompts below, a 30-100 email-a-day inbox gets to zero in 30-40 minutes without sending anything that makes recipients squint.
What’s actually in Gmail now (June 2026)
Google rolled the “Gemini era” of Gmail out on January 8, 2026, in the US, in English first. Here is what you get, and what it costs:
| Feature | What it does | Plan needed |
|---|---|---|
| AI Overviews (summaries) | Auto-summary of a long thread at the top of the conversation | Free, all users |
| Help Me Write | Draft or polish an email from a prompt | Free, all users |
| Suggested Replies | One-click replies that mimic your writing style | Free, all users |
| Proofread | Grammar, tone, and style rewrite | Google AI Pro ($19.99/mo) and up |
| Ask your inbox | Natural-language search across your mail history (“Who quoted the bathroom job last year?”) | Google AI Pro / Ultra |
| AI Inbox | Auto-prioritizes VIPs and pulls to-dos from important mail | Rolling out, early access |
Plan prices as of June 2026: Google AI Pro is $19.99/month (Gemini 3.1 Pro, 1M-token context, 2 TB storage) and Google AI Ultra is $99.99/month. Gemini is also bundled at no extra cost in paid Google Workspace plans — Business Starter gets Gemini in Gmail; Business Standard ($14/user/month annual) and up include the full Google AI Pro feature set. For the broader picture across Docs, Sheets, and Drive, see the Gemini for Google Workspace deep workflow.
A note on privacy that the old version of this guide got wrong: Google states it does not train Gemini models on the content of your personal Gmail. The opt-out is the Smart features setting (Settings → See all settings → General). Turning it off also disables tab categorization, predictive text, and package tracking, so weigh that before flipping it.
Who this is for
Anyone receiving 30+ emails a day and feeling underwater: managers with team threads, founders juggling investors and customers, sales reps with active pipelines, consultants with multiple clients, support leads triaging escalations.
When to use it — and when not to
Good fits: daily inbox triage, end-of-day cleanup, drafting replies to threads with shared context, and finding the half-finished thread you forgot to send Friday. When the reply is long enough to belong in a Doc instead of the inbox, switch to the Gemini in Docs deep workflow.
Skip it for: cold outreach (your voice matters too much; LLM-drafted cold emails get filtered or ignored); short personal replies under three sentences; legal or contractual emails where exact phrasing carries weight; and anyone who has told you they dislike AI-drafted replies.
Before you start
- Build a
voice fileDoc in Drive: three to five emails you wrote yourself that read like you at your best. Pin it to your Drive sidebar so you can paste from it fast. - Set a morning triage window of 15-20 minutes. Longer and you re-litigate every thread; shorter and the triage stays shallow.
- Decide your recipient policy up front. Client-confidential and legal threads stay manual no matter how good the draft looks.
- Pick one prompt format and stick with it for a week. Switching prompts mid-week destroys your sense of whether the workflow actually works.
The daily workflow, step by step
- Morning triage. Open Gemini in Gmail (the side panel, or the Gemini app pointed at your account) and run:
Summarize my unread inbox from the last 24 hours. Group by: action-required, info-only, marketing, automated. Suggest a priority order within action-required.Run this once at the start of the day. If you have AI Inbox in early access, let it surface the VIP/to-do snapshot first, then run the prompt on what’s left. - Per-thread summary. For each action-required thread, open it and read the AI Overview at the top of the conversation. If it’s missing or thin, ask:
Summarize this thread in 3 bullets. Highlight any deadlines or open questions directed at me.Read the bullets, not the whole thread, on the first pass. - Draft from bullets, never from the thread alone. Give Gemini your points first, then the tone reference:
Write a reply matching the tone of @voice-file. Friendly, concise, no fluff. Points I want to make: 1) ..., 2) ..., 3) ...Letting Gemini write from the thread alone is the single biggest mistake — it guesses your stance and gets it wrong in subtle ways. - Edit every draft. Change two or three words per email so it sounds like you. The phrases to delete on sight: “I wanted to circle back,” “Hope this email finds you well,” “Just following up.” Those are the tells.
- Retrieval for long threads (Pro/Ultra). This is where the paid tier earns its keep. Ask your inbox in plain language:
Find the prior context on [topic] with [person]. What is the open question, and what was the last commitment from each side?Gemini searches your mail history and cites the threads it used. At Google I/O 2026 this expanded into voice search (“Gmail Live”), so you can ask hands-free. - End-of-day close-out. Run:
List threads I started but did not finish replying to today. List threads where I owe someone a response by tomorrow.A two-minute close-out beats a Monday-morning surprise.
Prompt templates worth saving
Keep these in a Doc called Gmail snippets and reuse them daily. The @voice-file reference is what keeps replies from reading like a press release.
| Job | Prompt |
|---|---|
| Triage | Summarize unread mail from the last 24h. Group: action-required / info-only / marketing / automated. Rank action-required. |
| Thread digest | Summarize this thread in 3 bullets. Flag deadlines and questions aimed at me. |
| Draft | Reply in the tone of @voice-file. Friendly, concise, no fluff. Points: 1) ... 2) ... 3) ... |
| Tighten (sensitive thread) | Tighten the wording of my draft without changing the meaning or my stance. |
| Scheduling | Extract the proposed meeting times from this thread and check them against my calendar. |
| Close-out | List threads I owe a reply on by tomorrow, and threads I started but didn't finish today. |
For sensitive threads, invert the flow: write the draft yourself, then ask Gemini only to tighten it. That way the stance is always yours.
Tuning it so the voice file actually works
Run a real measurement on your first normal weekday (not a quiet-Monday). Triage as above, summarize five action-required threads, and draft three replies from bullets. Then track two numbers:
- Edits per draft. Two or fewer means the voice file is doing its job. More than five means the voice file or your bullets need work — fix one variable the next day and re-measure.
- Clarifying replies. If a recipient asks a clarifying question your real email wouldn’t have prompted, the draft drifted from your stance. Read three sent emails aloud each week; if any sound like a different person, refresh the voice file. Plan to refresh it every two months regardless, because your own writing drifts.
Common mistakes
- Letting Gemini draft from the thread alone. It guesses your points and gets them wrong in subtle, embarrassing ways.
- Sending the first draft unedited. Recipients can tell, and your reputation pays for it slowly.
- Using triage every morning but never the retrieval features. Half the value lives in “ask your inbox,” not in summaries.
- Skipping the voice file. Every reply reads like a template and relationships erode quietly.
- Auto-sending. Even where it’s technically possible, don’t. The first wrong send teaches you why this rule exists.
- Leaving Smart features off without realizing it also kills categorization and predictive text — then blaming Gemini for a worse inbox.
FAQ
- Is Gemini in Gmail free, or do I need to pay?: The summaries, Help Me Write, and Suggested Replies are free for all Gmail users in the US as of 2026. Asking your inbox questions in plain language and Proofread require Google AI Pro ($19.99/month) or Ultra ($99.99/month).
- Does Google train Gemini on my emails?: Google states it does not use the content of your personal Gmail to train Gemini models. The features run on Gemini 3. You can still opt out of all of it via the Smart features setting, which also disables categorization and predictive text.
- Does Gemini learn my tone automatically?: Suggested Replies adapt to your style, but for full drafts you should still feed samples or reference a voice file. The voice file is what stops replies from sounding generic.
- Can it auto-reply for me?: Suggested Replies are one click, but a human should approve every send. The first wrong auto-send costs more than the time it saved.
- Will it search across my whole email history?: Yes, within your account, but only on Google AI Pro or Ultra. The retrieval feature is half the reason the Gmail integration is worth using.