Microsoft 365 Copilot vs Gemini for Workspace: What the $30 Seat Actually Buys

Since 1 July 2026, Microsoft and Google both charge $14 a month for their Business Standard suite. Only one of those $14 seats puts AI inside the word processor and the spreadsheet. Microsoft 365 Copilot in Word, Excel and PowerPoint still costs a further $21 to $30 per user. Gemini for Workspace ships inside the Google plan, rationed by monthly usage limits instead of a price tag.

What a Microsoft 365 Copilot seat and a Gemini seat each include

The two vendors have arrived at the same shape from opposite directions. Both now give every suite user a basic AI chat. Both reserve the in-document assistant for a paid tier. The difference is where that paid tier begins.

Microsoft gives Copilot Chat to every Entra account user with an eligible Microsoft 365 subscription at no extra cost. On its own pricing page, that free tier covers web-grounded chat, Copilot in Outlook and the Copilot app, and agents "priced on a metered basis".

The paid Microsoft 365 Copilot licence adds the parts most buyers are actually paying for. Its comparison table lists Copilot in Word, PowerPoint, Excel, OneNote and Teams as licensed-only, alongside the Researcher and Analyst agents and priority model access. Agent usage and Copilot connectors move from "Metered" to "Included".

The boundary is not as clean as the table suggests. Microsoft's December 2025 pricing announcement said Copilot Chat users would get "access to Word, Excel, and PowerPoint agents", with a footnote that availability varies by tenant. Your own tenant decides which side of the line a given feature sits on this quarter.

Google went the other way in January 2025. It folded Gemini into its Business and Enterprise plans and dropped the add-on. Business Standard plus the old Gemini Business add-on had cost $32 a user. The combined plan became $14, which Google said was only $2 more than Workspace without Gemini.

The tier still matters. On Google's current pricing page, Gemini in Docs, Sheets, Slides, Drive, Meet and Chat requires Standard or above. Business Starter gets Gemini in Gmail and basic access to the Gemini app. A head of operations planning a rollout on Starter seats will find the spreadsheet assistant missing.

The fair comparison is therefore Microsoft 365 Copilot against Google Business Standard or higher. Microsoft's $30 seat is the enterprise add-on. Google's $30 product, Gemini Enterprise, is a separate agent platform, covered below.

Microsoft 365 Copilot vs Gemini pricing, line by line

Microsoft's July 2026 price reset is what makes this comparison sharp. Its December 2025 announcement listed new list prices for every commercial suite, and justified them partly by the AI added to the base plans.

Microsoft suite prices are its July 2026 list prices for SKUs that include Teams. Google prices are from its Workspace admin help page on Business edition pricing, updated October 2026. Flexible monthly billing costs more on both sides: Google Business Standard is $16.80, Copilot Business is $25.20.

The $30 seat Google does sell

Google does have a $30 product, and buyers confuse it with Gemini in Docs. Axios reported in October 2025 that Gemini Enterprise launched at $30 per person per month, with a $21 Gemini Business tier for smaller teams. That is an agent and search platform sitting beside Workspace, not the assistant inside Sheets.

I could not open a current Google page confirming those seat prices, so treat them as launch pricing. The point stands either way. If your Google shortlist includes Gemini Enterprise, you are no longer comparing a free bundle against a $30 add-on. You are comparing two $30 seats.

Where Microsoft 365 Copilot and Gemini meter usage

A flat price per seat is only half of the cost model. Both vendors now meter something, and the meters sit in very different places. That difference decides which bill surprises you.

Copilot Studio sells capacity in packs of 25,000 Copilot Credits at $200 per pack per month. Pay-as-you-go billing needs an active Azure subscription and charges at the end of the month for credits consumed. Microsoft's pricing page says a varying number of credits is billed "whenever an action or response is completed by an agent".

For licensed Copilot users, assistive custom agents carry no extra charge. For the much larger population on free Copilot Chat, every custom agent your team builds is a metered line. If you plan to license 50 power users and give everyone else agents, the meter is where your budget goes. I explained why that design spreads in the analysis of who pays when AI is bundled or metered.

Google's meter is a set of per-user monthly limits. Its usage limits page, updated October 2026, lists 5,000 Sheets AI function uses a month on paid plans, 100 generated slides and 5 Docs style-matching runs. AI Expanded Access raises those to 25,000, 500 and 25. Business Starter gets none of the three.

The quotas cannot be shared. Google states that limits "don't roll over" and "can't be shared with other people". So one analyst who fills a pricing sheet with AI functions hits the ceiling while 40 colleagues leave theirs unused. Workspace Studio automations move from promotional access to enforced limits on 1 November 2026.

This is a quieter form of the pattern traced in the piece on AI shrinkflation and usage limits. The price holds and the allowance becomes the variable. I prefer Microsoft's model here, because a credit bill is visible to finance while a quota wall is only visible to the person who hits it.

Document task evidence in Word, Excel, Docs and Sheets

The brief for this comparison was to score real document tasks. I have not run a controlled bake-off, and nobody independent has published one covering both suites. What exists is a vendor benchmark for Excel, one large government trial of Copilot and vendor-published pilots of Gemini. Here is what each says.

Spreadsheets: one benchmark, one silence

Microsoft reported that Agent Mode in Excel scored 57.2% on SpreadsheetBench, a public set of real spreadsheet questions. The human baseline in the same chart is 71.3%. The earlier Copilot in Excel scored 20.0%, so the agent nearly tripled its predecessor.

Google has published nothing comparable for Gemini in Sheets that I could find. That silence is not evidence that Gemini is weaker. It does mean a buyer running a finance team on spreadsheets has one vendor number to test and none from the other. For how these accuracy rates translate into a real model, see the test of AI accuracy on Excel financial models.

Drafting and summarising: Copilot's government trial

The largest public evaluation is the UK Government Digital Service experiment. About 20,000 civil servants across 12 organisations used Microsoft 365 Copilot from 30 September to 31 December 2024. Participants reported saving an average of 26 minutes a day.

Drafting documents saved 24 minutes and creating presentations 19 minutes. The method matters, though. Savings were self-reported in ranges and averaged at the midpoints, and the report says it could not identify how the saved time was spent. It also found limits with "complex, nuanced, or data-heavy aspects of work".

That gauge is the strongest argument for checking usage before you license. Excel and PowerPoint, two of the surfaces that the paid seat adds, peaked at 23% and 24% of participants. If your rollout looks similar, most of your $30 seats are buying chat you could have had for nothing.

Gemini's evidence is thinner and vendor-published

The Gemini figures come from Google's customer stories and award entries. Google's Utah case study says the state deployed Gemini to 22,000 employees. High-volume pilot users, those with 40 or more interactions in 28 days, reported saving 3.5 hours a week. That is a self-selected heavy-user group from a 257-licence pilot.

Arizona's Department of Administration ran a four-week pilot in September 2024 with more than 200 users across nine agencies. Its StateScoop 50 award entry says results suggest employees "could increase their productivity by 2.5 hours per week". No method is described. Neither figure is comparable to the GDS trial, and I would not put either in a business case.

Admin controls in Copilot and Gemini compared

For a head of operations, this section is the actual decision. Answer quality changes every quarter. The controls you build your rollout on change slowly and are hard to reverse.

Turning the assistant on and off

Google's admin console lets you switch Gemini features on or off by organizational unit or configuration group, and group settings override organizational units. The default is on. Per-app control is limited to Enterprise Standard, Enterprise Plus and some Education editions, so Business customers get coarser switches.

Google's own documentation carries the caveat that matters: "Even if Gemini is off for a specific app, users can still access its data when using Gemini in other apps". Switching Gemini off in Drive does not keep Drive files out of Gemini answers elsewhere. Microsoft's equivalent problem is oversharing through permissions, covered in the Glean and Copilot permissions comparison.

Audit and retention

Microsoft logs Copilot activity in Purview automatically as part of Audit (Standard). Each CopilotInteraction record lists the files and sites Copilot accessed to answer, including their sensitivity label. Prompts and responses are stored in a hidden mailbox folder and governed by a dedicated retention location called "Microsoft Copilot experiences".

Google offers Gemini audit logs through the Admin SDK Reporting API, covering Docs, Sheets, Slides, Gmail, Drive, Meet and other apps. Admins can auto-delete Gemini conversation history after 90, 540 or 1,080 days, or never. Google's privacy hub says Workspace app prompts are kept for 90 days to indefinitely, as admins decide.

Data boundaries and training commitments

Both vendors make the commitment most procurement teams ask about first. Both say they do not train foundation models on your content. The differences sit in the detail around processing location and human review.

Microsoft's privacy documentation, updated July 2026, states that prompts, responses and Graph data "aren't used to train foundation LLMs". Copilot services have opted out of the Azure OpenAI abuse monitoring that includes human review. For EU customers, Copilot is an EU Data Boundary service.

Then comes the line every EU buyer should read twice: "Models provided by Anthropic as a subprocessor are currently excluded from the EU Data Boundary". Customers outside the EU may have queries processed in the US, the EU or other regions. Microsoft states the exception plainly, which is more than most vendors do. It is still an exception you have to configure around.

Google's privacy hub, updated October 2026, says Workspace "does not use customer data for training models without customer's prior permission or instruction". Content is not human reviewed outside your domain without permission. Client-side encryption can keep Gemini away from sensitive files, because Google has no technical means to read them.

On location, I found less. The hub states that Gemini Notebook does not follow your organisation's data region settings. I could not find a general statement of where Gemini in Docs or Sheets processes prompts. If data residency is a hard requirement, get that answer in writing before you sign. The wider governance gap between enterprise assistants is mapped in the comparison of ChatGPT, Claude and Gemini governance.

A scored Copilot vs Gemini comparison

Here are my scores, out of 5, on six criteria. They rest on the vendor documentation and published evidence above, not on hands-on task testing. Change the weights to match your organisation. The weights are where the decision really lives.

At these weights Microsoft 365 Copilot wins by 6 points. Raise the cost weight to 40 and cut each of the other five to 12, and the order flips: Gemini scores 69 and Copilot 62. That sensitivity is the honest finding. Neither suite dominates, and the answer depends on whether you are buying governance or buying reach.

My own call is simple. A regulated firm with an EU footprint and a finance team in Excel should pay for Copilot on the roles that need it. A firm under 300 people that wants every employee drafting with AI should start with Gemini on Business Standard. Do not switch suites for the assistant alone. The migration cost dwarfs the price difference, a point argued in the piece on standardising on one AI tool.

Where this comparison is weakest

This post was meant to score real document tasks side by side. It does not, and that is its largest gap. The scores rest on documentation, a vendor benchmark and self-reported trials.

The evidence is asymmetric

Microsoft has published more about Copilot than Google has about Gemini. A documentation-based score therefore rewards the vendor that documents more. Some of Copilot's lead on audit and data clarity may be a lead in disclosure, not in the product. A buyer with access to Google's contractual terms may find answers I could not.

Everything here moves quarterly

Microsoft has redrawn the free and paid boundary twice in a year. Google enforces new Workspace Studio limits from November 2026. The SpreadsheetBench result is a year old, and both vendors have shipped new models since. Treat every figure in this post as dated October 2026, and recheck before you sign a multi-year deal.

The cost case for Copilot may be stronger than it looks

If your firm already pays for E3 or E5, Microsoft's argument is that you get Copilot Chat, Purview and the suite you already use. A Google comparison then carries a migration bill that this scorecard never counts. For an established Microsoft shop, the realistic choice is Copilot on some seats or no paid AI at all, not Copilot or Gemini.

Frequently asked questions

Where to start: one test, two weeks

Pull your own usage data before you buy anything. In Microsoft 365, check how many users are already active in free Copilot Chat. In Workspace, check the Gemini audit logs for who is near the monthly limits. The GDS trial suggests the people who need the paid tier are fewer than the people who ask for it.

Then run one workbook test. Take five spreadsheet tasks your finance or operations team did last month, with known correct answers, and give them to both assistants. Score full correctness only. Whichever product gets more of your own five right is the one your scorecard should favour, whatever any benchmark says.

Related reading