The demo is genuinely impressive. Someone opens a Google Doc, clicks the little sparkle icon, types “help me write a project brief,” and eight seconds later there’s a coherent draft on screen. Everyone nods. The pilot gets approved.
Then the seats go live across a 200-person company, and by week three the complaints start landing in your inbox. The sales team says Gemini can’t find anything useful in Gmail. Finance wants to know why the Sheets summaries keep referencing the wrong tab. Someone in Legal asks a question about eDiscovery that nobody can answer confidently.
That gap between the demo and the Tuesday-afternoon reality is what this review is about. Gemini for Workspace is a real product with real utility, but the limitations are specific and they matter more the larger your team gets. Here’s what tends to bite.
Three different products are hiding under one name
Before anything else, get this straight, because it causes more confusion than any technical issue. “Gemini” refers to at least three separate things inside a Google shop:
- Gemini in Workspace (the side panel inside Gmail, Docs, Sheets and the rest), sold as a per-user add-on on top of a paid Workspace plan.
- The Gemini app at gemini.google.com, which is a consumer-style chatbot with its own data terms and its own chat history.
- Gemini API / Vertex AI, which is the developer route and has nothing to do with your staff writing emails.
Your employees will use the second one regardless of what you buy, often signed in with a personal Gmail on a work laptop. Chats in the consumer app don’t appear in your Workspace admin console, and the retention rules are different. That’s a governance problem dressed up as a training problem, and no amount of internal comms fully solves it.
The licence maths nobody puts on the slide
Workspace Business Standard runs around $14 per user per month. The Gemini add-on for that tier has been priced around $24 per user per month on annual commit, more if you go month to month. Stack them and you’re near $38 per seat, per month, for one person.
That number is fine if the tool saves each employee two hours a month. It’s less fine when you discover that half the value people actually want (drafting an email, cleaning up a paragraph) is already available free elsewhere.
Worth reading alongside this: our breakdown of what ChatGPT Plus gives you for $20 a month, because that’s the comparison your CFO will make whether you like it or not. For a wider view of how the per-seat costs stack up across vendors, the AI chatbot pricing comparison is a useful sanity check before you sign a two-year commit.
Seat assignment is clunkier than it should be
You can buy Gemini seats for just the people who need them, which is the right move financially. In practice, admin console management is fiddly. Licences get assigned to the wrong OU, someone leaves and the seat sits idle for months, and there’s no clean self-serve way for a manager to request one. Budget an hour a month for licence hygiene or you’ll quietly waste 10% of the spend.
Feature parity across apps is a myth
The marketing page shows a tidy grid with checkmarks everywhere. The reality is that quality varies enormously depending on which app you’re in and what you’re asking for.
What each app actually handles well
Gmail is the strongest of the bunch. Thread summaries are accurate and genuinely time-saving on long chains, and draft replies are a reasonable starting point. Where it falls down is search. Ask it to “find the contract discussion from the Q3 vendor negotiation” across 40,000 emails and you’ll get three plausible-looking threads that are the wrong ones.
Docs drafts competently and rewrites tone well. It struggles with structural work on long documents. A 40-page policy document gets truncated in the side panel, so the suggestions you get are based on whatever chunk it could read.
Sheets is the app most people overestimate. Simple formulas, quick table generation, and explanations of what a formula does are all fine. Multi-tab models with cross-sheet references and named ranges go wrong often enough that you’ll double-check every output, which defeats the point. And it won’t restructure a messy sheet for you. You still do that by hand.
Slides has the weakest output. Generated visuals look generic, and layout suggestions tend to break brand templates in ways that take longer to fix than building the slide yourself.
Meet is useful if you qualify. “Take notes for me” requires the right tier, works in English, needs transcription enabled, and prompts for consent at the start of each call. Recording retention settings also need a decision from Legal before you switch it on.
Drive summarises a single PDF well. Ask it a question spanning fifteen documents and you need to select them all first, at which point you’ve done the hard part yourself.
Language, region and mobile support lag behind the pitch
Rollout has been English-first. Many of the better features landed for English-language accounts months before wider language support, and quality in smaller languages is noticeably thinner. If you run teams in Lisbon, Warsaw or Manila, test with their actual documents before you assume parity.
Availability also varies by country, billing currency and reseller. Data residency options exist for EU and US regions, but not every combination of plan, add-on and region is supported. Get this confirmed in writing rather than trusting a partner’s verbal assurance.
On mobile, the side panel experience is thinner than desktop. People who work primarily from a phone will get a fraction of what the demo promised.
It reads and drafts, but it doesn’t really do
This is the limitation that frustrates people most, because the word “assistant” sets expectations the product doesn’t meet. Gemini in Workspace will draft, summarise, rewrite and explain. It won’t file your expenses, triage your inbox by rule, book the meeting, or push a summary into a project tool.
It’s also a silo. Grounding is confined to Google’s own surfaces. If your sales team lives in a CRM and your engineers live in Slack, Gemini in Workspace has no visibility into either without additional paid plumbing. Enterprises that want broad connector support end up looking at the higher tier of the offering, which is a different budget conversation.
For teams weighing that trade-off, our review of ChatGPT Enterprise and its custom pricing covers what the connector-heavy alternative looks like when you need it to reach beyond one vendor’s ecosystem.
Data controls to verify before rollout
Google’s Workspace terms are solid, but the details need checking against your own obligations:
- Vault and eDiscovery treatment. Confirm in writing how side-panel prompts and generated content are captured, retained and exported. Don’t assume it mirrors email retention.
- DLP coverage. Classification rules scan content once it exists. They don’t stop Gemini from generating something sensitive in the first place.
- Shadow usage. Switching the add-on off doesn’t stop staff pasting documents into the consumer app. Policy and training do more work here than admin toggles.
- Third-party sharing. If Gemini output ends up in a shared Drive folder, the original sensitivity label often doesn’t travel with it.
Who should actually skip it
If your company runs on Microsoft 365, this is a non-starter. Buy Copilot inside the tools your people already open every day rather than paying for a second ecosystem.
If your core need is heavy analysis across a large document corpus, or long-form reasoning where you’d rather have depth than convenience, a dedicated assistant will serve you better. Our comparison of Claude against the main alternatives walks through where each one earns its keep, which is faster than learning it the expensive way.
Where Gemini for Workspace does justify the spend: an organisation already standardised on Google, where drafting and summarising email is genuinely the bulk of the workload, and where IT can support a single vendor. In that setup, a $24 add-on is easy to defend. Roll it out to a 20-person pilot first, track how many people open the side panel in week four rather than week one, and let that number decide the wider deployment.