№ 06Learn With Darin · Field Guide

Gemini: a practitioner's field guide.

Google's flagship AI in three uneasy roles at once: a standalone chat app, a layer baked into Workspace, and (on most newer Android phones) the system assistant. Knowing which Gemini you're talking to is most of the battle.

Updated May 2026 ~25 min read Covers Free, AI Pro, AI Ultra, Workspace add-ons

Part 01

What Gemini actually is in 2026

"Gemini" is the brand Google attaches to almost everything AI-flavored it ships. That breadth is the first thing to internalize, because the same word covers products with very different ergonomics, permissions, and audiences. As of May 2026, the Geminis you'll meet are:

gemini.google.com, the standalone web app. The closest analog to chat.openai.com or claude.ai: a chat surface with file uploads, image generation, Canvas, Gems, and Deep Research.
The Gemini app on iOS, Android, Mac, and Windows. The mobile apps are full-featured chat clients; the desktop apps are thinner wrappers that arrived in 2025.
Gemini in Workspace, accessed via the side panel inside Docs, Sheets, Slides, Gmail, Drive, Meet, and Calendar. Same model family, different surface, different data scope.
Gemini as the Android assistant. On most phones running Android 14+ that have opted in, Gemini has replaced Google Assistant as the system-level voice assistant. This is the surface Apple's ecosystem has no equivalent of.
Gemini for Google Cloud / Vertex AI, the developer-facing surface. Out of scope for this guide.

The model lineup, also as of May 2026:

Gemini 2.5 Flash: the fast, cheap default. Good enough for most chat, summarization, and short-form generation. Free users get this.
Gemini 2.5 Pro: the flagship reasoning model. Available on AI Pro, AI Ultra, and most Workspace add-on tiers. The model that does well on hard problems.
Deep Think: an extended-reasoning mode for 2.5 Pro, available on AI Ultra. The closest thing Google has to OpenAI's o-series or Anthropic's extended thinking.
Imagen 4 for image generation, accessed through the chat surface and through the Workspace apps.
Veo 3 for video generation, available on AI Ultra and inside the Vids app.

Note The model picker in the chat app does not always show what's actually being used. On Free, requests can quietly downshift to Flash even when you've nominally chosen Pro, especially under load. If a Pro answer feels suspiciously thin, it probably was.

What makes Gemini structurally different from Claude or ChatGPT is the distribution. Google does not have to convince you to install an app or visit a URL. Gemini already lives inside the documents you write, the email you read, and the phone in your pocket. That changes which workflows make sense, and which ones don't.

Part 02

Web vs Mobile vs Workspace

The standalone Gemini and the Workspace-embedded Gemini are not the same product. They share a model family and a brand, and almost nothing else. The mobile app sits between them. Here's where each one sits:

Standalone (web + apps)

Surface: gemini.google.com and the Gemini apps for iOS, Android, Mac, Windows.
Sees: files you upload, things you paste, results from Google Search if grounding is on, Workspace data only when you explicitly enable the connectors.
Has: Canvas (a side-panel editor like ChatGPT's), Gems (custom personas like Custom GPTs), Deep Research, image and video generation, voice mode, scheduled actions on Pro and above.
Best for: open-ended chat, research, drafting from scratch, anything that doesn't need to land back in a specific Google file.

In Workspace

Surface: the side panel inside Docs, Sheets, Slides, Gmail, Drive, Meet, Calendar; plus inline "Help me write" buttons.
Sees: the open document, optionally other files in the user's Drive, optionally recent emails, all subject to Workspace sharing permissions.
Has: tighter integration (insert directly into the doc, summarize the open thread), but a smaller feature set. No Gems, no Canvas, no image generation in some apps.
Best for: editing what's already in front of you. Summarizing a Doc you're reading. Drafting a reply to the email you have open.

Mobile + assistant

Surface: the Gemini app, plus (on Android) the system-level assistant invoked by long-pressing power or saying "Hey Google."
Sees: what you give it through Share or paste, and (with permission) on-screen content via "Ask about screen."
Has: voice mode, Live mode for continuous conversation, camera-based queries, integration with phone actions (set timers, send messages, open apps).
Best for: voice-first interactions, anything you'd do with a smart assistant, summarizing what's on the screen in front of you.

The piece people miss most often: the Workspace side panel does not have access to the same features as the standalone app. If you're trying to use Gems or Deep Research from inside Gmail, you can't. Switch to gemini.google.com and bring the context with you.

Tip A useful rule. If you want Gemini to edit something, use the Workspace side panel. If you want Gemini to think, use the standalone app. The side panel is a fast path for "rewrite this paragraph" or "summarize this thread." It is not the right place for hard problems.

Part 03

Capability matrix

Gemini's plan structure is messier than its competitors because consumer plans (AI Pro, AI Ultra) and Workspace add-ons (Business, Enterprise) overlap in confusing ways. Here is the simplified picture for May 2026:

Capability	Free	AI Pro $19.99/mo	AI Ultra $249.99/mo	Workspace add-on per-seat
2.5 Flash	✓	✓	✓	✓
2.5 Pro	limited	✓	✓	✓
Deep Think mode	–	–	✓	–
Deep Research	limited	✓	✓ (higher cap)	varies
Canvas	✓	✓	✓	✓
Gems (custom)	✓	✓	✓	admin-controlled
Imagen 4 image gen	limited	✓	✓ (higher cap)	✓
Veo 3 video gen	–	limited	✓	varies
Long context (1M+ tokens)	partial	✓	✓ (2M)	✓
Workspace data grounding	–	personal Drive	personal Drive	org Drive + Gmail + Calendar
Gemini in Docs / Sheets / Slides	–	personal accounts	personal accounts	✓
Gemini in Gmail	–	personal accounts	personal accounts	✓
NotebookLM Plus	–	✓	✓	✓ (Enterprise)
2 TB / 30 TB Drive storage	15 GB	2 TB	30 TB	per Workspace plan
Android assistant role	✓	✓	✓	✓

"Personal accounts" in the Workspace rows means: AI Pro and AI Ultra include Gemini in Docs and Gmail only if your account is a personal @gmail.com. If you're signed into a Workspace account (yours or your employer's), the side panel is gated by the Workspace add-on, not by your consumer subscription. This is the single most confusing part of Gemini's pricing, and it catches almost everyone the first time.

Warn If your employer uses Workspace and has not purchased the Gemini add-on, your personal AI Pro subscription does not unlock Gemini inside your work Gmail. The plans don't follow you across account contexts. This is the part of Google's licensing that feels punitive, and there is no workaround other than asking IT to enable it or doing the work in a personal account.

One more wrinkle: AI Pro and AI Ultra are bundled with Google One. The 2 TB / 30 TB storage isn't a side benefit; it's part of why those plans cost what they do. If you already pay for Google One storage, the marginal cost of AI Pro is smaller than the sticker.

Part 04

Workspace integration: where Gemini genuinely competes

If there's one place Gemini is plainly the best tool available, it's inside Workspace. Not because the model is unambiguously better than Claude or GPT-5 (it isn't), but because Gemini is the only assistant that can read your Drive, Gmail, and Calendar with proper organizational permissions. Claude and ChatGPT both ship Workspace connectors, and they work, but they introduce a third party into the data flow. Gemini doesn't.

What "grounding in Workspace data" actually means

When you ask Gemini in Gmail "what did Priya say about the Q3 launch," Gemini queries your mail index using the same permissions your account has. It doesn't open a browser session and scrape; it's reading from inside the Google security boundary. Three practical implications:

It respects sharing. If a Doc is shared with you read-only, Gemini sees what you can see. If you weren't on the email thread, Gemini won't pretend you were.
It's fast. Indexed access to Drive and Gmail is much faster than the file-by-file ingestion that connector-based tools have to do.
The audit trail lives in Workspace's existing logs, not in a separate AI vendor's dashboard. Admins who already understand Workspace audit don't have to learn another surface.

Per-app behavior, briefly

Docs: side-panel chat plus inline "Help me write." The inline tool is best for short paragraph rewrites; the side panel is best for "summarize this doc," "rewrite this section in a more formal tone," "extract the open questions." Canvas is not available in Docs; if you want a Canvas-style scratchpad, do it in the standalone app and paste back.
Sheets: "Help me organize" generates table structures; the side panel can write formulas, explain formulas, and analyze data. Formula generation is reliable; data analysis is correct most of the time but worth spot-checking against what you'd compute manually for anything load-bearing.
Slides: the inline tool generates draft slides from a prompt and can illustrate them with Imagen 4. Quality is meaningfully better than the equivalents in Word or PowerPoint, but still recognizably AI-deck. Use it as a starting point, not a finished artifact.
Gmail: "Help me write" and "Help me reply" inline; side panel for "summarize this thread" and "find emails from X about Y." The summarization is the killer feature, especially on long threads. The drafting is fine, sometimes too long, frequently too formal.
Drive: side panel for "summarize this PDF" and "find files about X." The cross-file search is the part that no third-party tool can match, because it queries the actual Drive index.
Meet: "Take notes for me" produces a meeting summary and action items. Reasonably accurate when audio quality is good; degrades fast on bad mics or heavy accents. Always review action items before sending.

The Workspace admin layer

For organizations, the Gemini add-on sits inside the Workspace admin console alongside every other Google service. Three controls matter most:

Per-OU enablement. Admins can turn Gemini on for some org units and off for others. Useful when you want to pilot it on the marketing team before rolling out to engineering.
Data residency and processing controls. Workspace's existing data-region controls extend to Gemini prompts and responses on the Enterprise add-on tier. Without it, processing happens in the default region.
Gem governance. Admins can restrict who can create, share, or use custom Gems org-wide. This matters more than it sounds: Gems can be configured with system prompts that affect output across the org.

In practice The Workspace admin console is where Gemini deployments succeed or fail. The license is the easy part. The harder part is figuring out which OUs get access, what data they're allowed to ground against, and whether Gem creation is a free-for-all or a curated set. Treat the rollout as an information-governance project, not a software install.

Part 05

The Android assistant role

This section has no equivalent in any other guide on this site, because no other AI tool plays this role. On most Android phones, Gemini has replaced Google Assistant as the system-level voice assistant. Saying "Hey Google" or long-pressing the power button no longer talks to Assistant. It talks to Gemini.

The transition started in 2024 and was largely complete by mid-2025 on Pixel and recent Samsung devices. As of May 2026, the picture is:

Pixel 8 and newer: Gemini is the default assistant, Assistant is no longer offered.
Samsung Galaxy S24+ and newer with One UI 7+: Gemini is the default; Bixby still exists as a separate option.
Older Android devices: Gemini is opt-in, available via the Gemini app's "Set as default assistant" prompt.
iOS: Gemini cannot be set as the system assistant. Siri remains the only option Apple permits.

What Gemini-as-assistant can do that the chat app can't

Phone-level actions: set timers, send messages, make calls, control smart-home devices, open apps, change settings. These work because the assistant role gets system-level intents the chat app doesn't.
"Ask about screen": long-press power, ask "what is this," and Gemini reads the current screen. Useful for translating menus, summarizing articles in apps that don't expose share, identifying what's in a notification.
Lock-screen voice: voice queries work without unlocking, with a configurable boundary on what can be answered without auth.
Live mode: continuous conversation with interruption handling, the way you'd talk to a person. The closest thing to Project Astra in shipping form.

Where it falls down

Gemini is a noticeably worse assistant than Google Assistant was for the original Assistant use cases. Setting a timer is slower. Sending a quick text takes more turns. Asking for the weather sometimes triggers a long-form answer. Google has been closing this gap for two years and the gap is still real.

The reason is structural. Assistant was a deterministic intent router: "set timer five minutes" mapped to a fixed action. Gemini is a language model that has to decide whether a query is an intent or a conversation. That decision is mostly right and occasionally wrong, in a way Assistant never was.

Warn On Pixel devices, Gemini-as-assistant has been observed to silently change Search behavior. Queries you'd expect to open a browser sometimes get answered conversationally instead, with a citation panel below. Useful, except when you specifically wanted the search results page. The setting that controls this is in the Gemini app under "Search behavior."

Privacy notes specific to the assistant role

"Ask about screen" sends a screenshot to Gemini. By default it's not retained, but check the privacy settings in your Gemini app.
Lock-screen voice respects whatever you've set under "Personal results when locked." If you've enabled it, anyone holding your phone can ask Gemini to read your last text aloud.
App actions (read messages, send messages) require per-app permissions, granted the first time and remembered after. Audit them periodically.

Part 06

Practical workflows

Gemini is at its best in the workflows that take advantage of where it lives. These six are the ones that hold up across months of daily use.

Weekly inbox triage in Gmail.

Open the Gemini side panel and ask: "summarize unread emails from this week, grouped by sender, with action items called out." It's faster than reading each thread and surprisingly good at flagging the ones that actually need a reply versus the ones that are FYI. Pair it with starring the action-item threads as you go.

ii.

Cross-document search in Drive.

"Find all docs in my Drive that mention [project name] and summarize what each one says." This is the workflow no third-party tool can replicate cleanly, because it requires real Drive index access. Particularly useful when you've inherited someone else's Drive folder structure and need to map the territory.

iii.

Deep Research before a meeting.

In the standalone app, kick off Deep Research on a company, person, or topic the night before. It runs for several minutes and produces a structured brief with citations. Skim it on the way to the meeting. Don't trust the brief uncritically: verify any specific number or quote you'd lean on, the way you'd verify a Wikipedia article you were citing.

iv.

Sheets formula explanation, then formula generation.

When you inherit a Sheet with a formula you don't understand, paste the formula into the side panel and ask "explain this." Once you understand it, ask for the variation you actually want. This pattern is faster than reading the documentation and produces fewer broken formulas than asking for novel ones cold.

"Ask about screen" for any app that doesn't expose Share.

Banking apps, government portals, in-game menus, anything that doesn't let you copy or share text. Long-press power, ask Gemini what's on the screen, and you get a summary or translation in seconds. This is the single best use of the Android assistant role.

vi.

A scoped Gem for recurring writing styles.

If you write a particular kind of artifact often (release notes, customer emails, retros), make a Gem with the format and tone baked into the system prompt. Use it from the standalone app. Don't try to reproduce the same effect in the Workspace side panel; Gems aren't available there. Paste the output back into the doc you're working in.

Part 07

Limits and pitfalls

The places Gemini is most likely to disappoint, in the order you'll meet them.

"My subscription doesn't work in my work account"

Already covered above, but it's the single most common complaint, so worth restating. AI Pro and AI Ultra unlock Gemini inside your personal Gmail and Docs. They do not unlock the Workspace side panel inside an employer's Workspace. Only the Workspace admin can do that, by purchasing the Gemini add-on. There is no consumer workaround.

"Gemini answered confidently and the answer was wrong"

True of every model, but Gemini's specific tell is overconfidence on numerical and recent-events questions. The grounding-on-search panel sometimes shows sources that don't actually support the claim above them. Read the citations; they're shown for a reason. If the claim is load-bearing, cross-check against another model. The cross-tool verification pattern from the practices guide applies here especially.

"I asked Gemini to read a Doc and it said it can't access it"

Three causes, in order of likelihood:

You're signed into a different Google account than the one with access. Check the avatar in the corner.
The Doc is shared with you via "Anyone with the link" but not actually in your Drive. Add it to your Drive (or have it shared explicitly to your address) and try again.
Workspace admin policy restricts cross-OU access. If you're external to the doc's OU, Gemini may be blocked even though you can read the doc in a browser. There is no client-side fix.

"The model picker says Pro but the answer feels like Flash"

It probably is Flash. Free accounts get rate-limited Pro access that silently degrades to Flash under load. AI Pro accounts hit the same fallback at much higher volumes, but it still happens. If you need Pro reliably, run the same prompt twice and compare; if both feel thin, retry later or upgrade.

"Image generation refuses my prompt"

Imagen 4's safety filter is more conservative than Imagen 3 was. It will refuse anything involving real people by name (politicians especially), anything that looks like brand infringement, and a wide swath of "edgy" prompts that would have worked a year ago. The refusals are rarely well-explained. Rephrase generically, or use a different tool for that specific prompt.

"Features I read about aren't available in my country"

Gemini's regional rollout is uneven. Veo 3, Live mode, Deep Think, and several Workspace features land in the US first and reach the EU and UK weeks or months later. Some features (notably anything involving real-time grounding in a user's Gmail) have stricter EU posture and arrive in modified form. If something is documented but missing for you, check the Google blog for regional availability before assuming it's broken.

"Gemini-as-assistant won't take an action it used to take"

Most often a permission issue. The first time Gemini is asked to read texts or send a message, Android prompts for the corresponding app permission. If you denied it, future requests fail silently or with a vague error. Walk through Settings → Apps → Gemini → Permissions and re-enable what you actually want it to have. Re-grant per app, not globally.

"The Workspace side panel doesn't show recent changes to my doc"

Indexing lag. The side panel reads from a cache that updates every few minutes, not in real time. If you've just made edits and asked for a summary, you may get the version from before the edits. Wait a minute, refresh the side panel, or close and reopen the doc.

Part 08

When to use Gemini vs another tool

The honest answer: not always. Gemini is excellent at a few things, fine at most things, and second-best at some things where the alternatives are clearly stronger. The rule that's served me well:

Use Gemini when the work touches your Google data. Use something else when the work is just thinking. — TWD

Concretely:

Reach for Gemini when: you're inside a Google Doc, Sheet, or Gmail thread; you need cross-document Drive search; you're using an Android phone in a hands-free context; the task involves long context (Gemini's 1M+ token windows are real and useful); you want NotebookLM-style synthesis across many sources.
Reach for Claude when: the task is structured writing, code, or anything that benefits from careful reasoning over careful retrieval. Claude's outputs are typically more thoughtful, less filler.
Reach for ChatGPT when: you want the largest ecosystem of third-party plugins, the most mature voice mode, or the most aggressive image-generation defaults. Also when you want a second opinion against Gemini, since they have different blind spots.
Reach for Copilot when: the work lives in Office, Teams, or Windows 11. Copilot's Microsoft integration is to Office what Gemini's is to Workspace.

The cross-check pattern from the best-practices guide works especially well with Gemini, because Gemini's strengths and weaknesses are different from Claude's and ChatGPT's, not just better or worse. If a Gemini answer feels confident but you can't tell if it's right, paste the same prompt into a second tool. The places they diverge are the places to dig in.

One closing observation

Gemini's biggest disadvantage versus its competitors is also the source of its biggest advantage. The same machinery that makes it the only tool with native Workspace and Android integration also makes it inseparable from Google's broader product strategy. Features arrive when they help Workspace adoption or compete with iOS, not always when they'd be most useful to you. If you're already deep in the Google ecosystem, that alignment works in your favor. If you're not, Gemini is a perfectly fine chat app that you'd probably skip in favor of one of the others.

And if any of this is out of date by the time you read it: blog.google/products/gemini is where Google posts changes; workspace.google.com/blog covers the in-Workspace updates. Both lag the actual rollouts by a few days, and regional availability is best confirmed by trying the feature in your own account.