Chapter 14: Gemini From Beginner to Advanced User

September 2, 2026

This entry is in the series Get on board with Artificial Intelligence

A quick note before you begin

Snapshot taken in the third quarter of 2026. Google frequently renames, rebundles, and reprices its AI products. The structure below will outlast the figures. Verify the specifics at gemini.google/subscriptions and, if you are using a work account, at workspace.google.com.

Three different things wearing one name

Untangle the word first, because most confusion about this product starts here.

Gemini the model family. The underlying systems Google builds. When a headline says Gemini beat something on a benchmark, this is what it means.

Gemini the app. The assistant at gemini.google.com and on your phone. This is the direct competitor to ChatGPT and Claude, and it is what the rest of this chapter mostly means.

Gemini inside Google’s other products. The side panel in Gmail, Docs, Sheets, Slides, Drive, and Meet; the AI answers above search results; the assistant on Android handsets. This is where most people actually meet it, frequently without having chosen it.

The practical consequence catches people out constantly, so it is worth stating plainly: paying for a personal Google AI subscription does not put Gemini into your work Gmail. Consumer plans and Workspace plans are separate purchases under separate terms. If your organization runs on Workspace, what you have depends on what your administrator enabled, not on what you pay for at home.

What it is, and how it differs in character

Gemini’s distinguishing argument is not that it writes better than the alternatives. It is proximity. It sits next to your mail, your files, your calendar, and your photographs, and it can read them.

That produces a different kind of usefulness. Asking a general assistant to help you plan a trip means pasting in your itinerary. Asking Gemini means letting it look at the booking confirmations already in your inbox. The gain is real, and so is the exposure: an assistant that can read your mail is an assistant that has read your mail.

Beyond integration, Gemini has been consistently strong on three things: very long inputs, multimodal work across images, audio, and video, and research tasks where breadth matters. Its free tier is also the most generous of the major assistants, which makes it the cheapest way to keep a genuine second opinion available.

The plans

Free. A working product rather than a demonstration: everyday chat, a daily allowance of the stronger model, file uploads, image generation, deep research in limited quantity, Canvas, Gems, and voice conversation. For a large number of readers this is sufficient and always will be.

The entry paid tier. A modest monthly fee for higher limits without the full feature set. Worth checking before you reach past it, because most people who upgrade are buying capacity, not capability.

The main paid consumer tier. Full access to the strongest general model, higher limits, longer context, video generation, and a monthly allowance of credits for the more expensive generative features. This is the tier most individual subscribers want.

The top consumer tier. Highest limits, the most compute-intensive reasoning modes, and large storage. Priced for people running heavy workloads all day. Its price fell sharply during 2026, which is a useful reminder that today’s premium tier is next year’s ordinary one.

Workspace. Google took a different path from Microsoft here, and it matters for cost. Rather than selling AI as a separate per-seat add-on, Google folded Gemini into Workspace subscriptions and raised the base price. An organization already paying for Workspace at the standard tier or above generally already has Gemini in Gmail, Docs, Sheets, Slides, Drive, and Meet, with the entry-level Workspace tier receiving considerably less. Higher-usage add-ons exist on top.

Two things follow. Many businesses are paying for Gemini right now and not using it. And the correct first question for anyone considering a personal subscription is whether they already have one through work, a phone purchase, or a storage plan.

Level two: Configuration

Gems. Saved, configured versions of the assistant with fixed instructions and their own reference material: the counterpart to Custom GPTs and Projects. You can attach knowledge sources, including files from Drive. The distinction worth internalizing is that custom instructions apply to every conversation, while a Gem applies only when you open that Gem. Build one when you do the same specialized task repeatedly; a saved prompt is enough for anything occasional.

Personal context and saved information. Standing details carried between conversations. Convenient, and a privacy surface. Know how to view it, edit it, and switch it off before you rely on it.

Connected apps. Permissions granting the assistant access to Gmail, Drive, Calendar, Photos, Maps, YouTube, and more. This is the feature that makes Gemini different from its competitors, and it is also the single most consequential setting in the product. Turn on what you will actually use. Review the list quarterly.

Which account you are signed into. Trivial to state, easy to get wrong, and the source of a great many accidental disclosures. Personal and work accounts have different capabilities, different data terms, and different administrators. Check the avatar in the corner before you paste anything sensitive.

Level three: Delegation

Deep research. Browses a large number of sources over several minutes and returns a structured, cited report. A long report with many footnotes creates an impression of thoroughness that is not itself evidence. Treat it as a faster way to map the territory, then read the sources that carry the argument.

Canvas. A working surface beside the conversation for drafting documents, building small applications, and iterating without regenerating everything each time. If you have used Artifacts, the idea will be familiar.

Scheduled actions. Recurring prompts that run without you present. Useful for a standing summary or a weekly check. Review them periodically, because a task configured in March and forgotten is still running in August.

Gemini Live. Voice conversation, including camera input, so you can show it something rather than describe it. Worth trying even if it sounds like a novelty; for some people it becomes the main way they use the tool.

NotebookLM. A separate Google product that deserves more attention than it gets, and which many readers of this book will find more valuable than the assistant itself. You give it sources, and it answers only from those sources, with citations back to the passage it used. That constraint is the point: grounding the model in material you selected removes the largest category of failure described in Chapter 8. It also generates audio overviews, study guides, and briefing documents from the same material. If you work with a defined body of documents, start here.

Gemini inside Google Workspace

For anyone whose working day happens in Google’s tools, this section is the chapter.

Gmail. Summarize a long thread, draft a reply from a few bullet points, adjust tone, and search your mail conversationally. Thread summarization on a forty-message chain is the feature people keep.

Docs. Draft, rewrite, and refine in the side panel, with suggested edits visible only to you until you accept them. More recent versions can assemble a first draft that pulls from your own files and mail rather than from nothing, and can match the style and format of documents you already have.

Sheets. Generate a tracking table from a description, get help with formulas, produce charts, and use AI-assisted fill to extend patterns across a column. Check the arithmetic, and prefer a formula you can read over a number you cannot trace.

Slides. Generate slides, rewrite the text on them, and create images without leaving the deck.

Meet. Automatic note-taking and transcription, so nobody has to volunteer. Recordings and notes are business records; treat them as such.

Drive. Ask questions across a folder rather than opening files one at a time. Comparing several proposals, or asking what the deadlines are across a project folder, is genuinely faster than doing it yourself.

The important asymmetry: Workspace data terms are materially stronger than consumer ones. Google states that Workspace content is not used to train models outside your domain without permission and is not reviewed by humans in the ordinary course. That is a different contract from the one covering personal use, and it is the reason work material belongs in the work account.

Where it shines

Held loosely, and true as of writing:

• Integration with mail, files, calendar, and photos, if your life already lives there

• Very long inputs, and a large amount of material held at once

• Multimodal work: images, audio, video, and live camera input

• Research and breadth, including a strong deep research implementation

• At the time this chapter was checked, one of the more generous free tiers among the major assistants

• Cost, for organizations whose eligible Workspace plans already include the features, where the incremental subscription cost may be little or nothing

Current limitations

Naming and product churn. Google’s product names and bundles change frequently. Features arrive under one name, move into another product, or disappear. That churn is exactly why Chapter 17’s LEARN method exists.

Consumer privacy posture. On consumer plans Google treats your chats as consumer data: a subset is reviewed by people and used to improve its models, and the principal control is turning off Gemini Apps Activity, which also stops your history being saved. Paying for a higher consumer tier buys capability, not different terms. This is worth repeating because the mistake is so common: on consumer plans, upgrading buys capability, not confidentiality.

Uneven integration quality. Summarizing a Drive folder works well; refining that summary through several follow-ups is less reliable, and the assistant sometimes loses the thread of what you asked. The side panels are helpful rather than transformative, and they are better at first drafts than at fine work.

The familiar failure modes apply in full. Fabricated specifics, confident errors, silent assumptions. Proximity to your data does not make the model more truthful about it; it makes the errors more plausible, because they arrive wearing your own material.

The subscription decision

Ask the same three subscription questions: am I hitting limits, do I need a specific paid capability, and what is an hour of my time worth?

Then add a fourth that is specific to Google: do I already have this? Between Workspace plans, phone bundles, storage subscriptions, and student offers, a surprising number of people are entitled to a paid tier they are not using, or are paying twice. Check before you subscribe, and check again at renewal.

Privacy settings to check

Gemini Apps Activity. Whether your conversations are retained and used for improvement, and for how long. This is the setting.

Connected apps. Every connection is a standing grant of access to a category of your life. Review the list; disconnect what you no longer use.

Account separation. Confirm which account is active. Establish a rule, now, about what may go into the personal one.

Complete these three checks before finishing the chapter. Chapter 25 builds the full framework.

Exercise 14.1: Ask it about your own material

Connect one service you are comfortable connecting — mail or files, not both — and ask a question you would otherwise have answered by searching manually. Something like: what did we agree about the delivery date, or which of these documents mentions the warranty.

Notice how much faster it is, and notice how much you would have to verify before acting on it. Both halves of that observation matter.

Exercise 14.2: Build one Gem

Pick a task you repeat: a weekly report, a type of email, a kind of analysis. Create a Gem with instructions covering context, standards, and format, and attach any reference material that should always be in scope.

Run the task through it, then run it through an ordinary conversation. The difference is the value of configuration, measured on your own work.

Exercise 14.3: Find your account boundary

Sign in to your personal account and your work account and ask each the same question about a document. Note what each can see, what each refuses, and what the interface tells you about which one you are in.