The short answer

There is no universal winner, but there is a fast heuristic. If your organisation already runs on Microsoft 365 — Outlook, Teams, Excel, SharePoint — Copilot is the higher-impact choice, provided you deploy it selectively rather than site-wide. If you are building a stack from scratch or watching cost per seat, Google Workspace with Gemini embedded delivers respectable AI capability at roughly a third of the price.

Everything below explains why, with the numbers to back it up.

Two vendors, opposite bets

The 2025–2026 licensing reshuffle tells you everything about strategy.

Google folded Gemini into Workspace in March 2025 and discontinued its separate $20–30 add-ons. Today, Business Starter (~$7) gets a limited assistant in Gmail’s side panel, while Business Standard ($14) and Business Plus ($22) ship full AI across Docs, Sheets, Slides and Meet.

Microsoft kept the add-on model. Copilot remains a $30-per-user monthly surcharge on qualifying enterprise plans, on an annual commitment. On 1 July 2026 the base licences underneath got pricier — Business Standard rose from $12.50 to $14, E3 from $36 to $39, E5 from $57 to $60 — so a fully loaded seat now runs to $69 on E3 and $90 on E5.

Headline comparison

Prices are list prices in USD per user per month on annual terms; regional pricing, taxes and negotiated discounts vary.

What the sticker price doesn’t tell you

Budget holders who compare “$30 add-on” against “free AI” are comparing the wrong numbers. Four hidden line items matter more.

1. The qualifying-licence trap

Copilot cannot be bought on its own — it requires an eligible Microsoft 365 plan first. For greenfield companies, that is the entire stack, not a surcharge.

2. Consumption economics

Copilot Studio sells agent capacity in credits — $200 monthly for 25,000, or $0.01 pay-as-you-go — with heavy workloads (voice, premium tools, document grounding) burning through credits five to thirty times faster. Google’s equivalent, Gemini Enterprise, starts at $21 per seat but partners estimate $15–40 in additional consumption per power user.

3. Expired discount ladders

Microsoft’s volume discounts on the enterprise add-on — 15% at 10 seats rising to 40% at 1,000 — lapsed on 30 June 2026; only a 15% three-year-commitment discount for 300+ seats survives.

4. The migration bill

Switching ecosystems to “save on AI” means migrating mail, files and — hardest of all — habits. Plan for three to six months of reduced productivity, not a line item in a quote.

Capability: depth against breadth

Copilot’s 2026 release cycle has been about embedding. Third-party models (Claude, GPT-5.2, Sonnet 5) are selectable inside Copilot Chat; Excel executes Python in-workbook via Edit with Copilot; agents are submitted to a corporate store with admin approval; AI-generated content can carry watermarks. The pitch is simple: the assistant sees a decade of your meetings, mail and spreadsheets. No rival can replicate that context.

Google’s 2026 story is generative range. NotebookLM has been rebranded Gemini Notebook while gaining agentic research and secure cloud code execution that turns source material into multi-page reports with charts and slides, and cinematic Video Overviews (Veo 3) convert documents into narrated video — a format Microsoft simply does not offer at present.

If daily work is meetings, email and spreadsheets, depth wins. If it is research, synthesis and content, breadth wins.

The evidence problem

Vendors publish impressive numbers; independent evidence is rarer and more useful.

The most cited figure — a Forrester Total Economic Impact study commissioned by Microsoft — models a 116% three-year ROI, $19.7m NPV and sub-eleven-month payback for a 25,000-seat organisation, driven by nine hours saved per user monthly, with SMB projections of 132–353%. Microsoft reports 20 million paid Copilot seats and penetration into 90% of the Fortune 500.

Now the counterweight. A three-month UK government pilot across 1,000 licences found high satisfaction (NPS 31) yet no robust evidence that time savings translated into productivity gains; Copilot-assisted Excel tasks were observed to be slower and less accurate, and 22% of participants encountered hallucinated content. A separate six-month trial at the Department for Work and Pensions measured a modest 19 minutes saved per user per day.

The honest reading: both assistants pay off reliably in language-heavy work — meeting summaries, first drafts, document search — and both underperform where numerical precision matters. Returns correlate with activation, not licences. An organisation lifting weekly active usage from 25% to 55% roughly doubles its return without buying a single additional seat.

Which company are you?

The established Microsoft estate

Staying is almost always cheaper than leaving. Deploy Copilot to the 20–40% of roles where work is genuinely language-led (leadership, sales, finance, marketing) rather than all seats. An 800-seat firm equipping 320 roles at $69 saves roughly two-thirds of the cost of a blanket rollout while keeping most of the benefit — provided SharePoint is tidied first, because an assistant trained on messy data confidently produces polished nonsense.

The cost-first SMB

On Business Standard, Microsoft’s own bundled plan ($23.50 with Copilot) undercuts buying the pieces separately. But if AI is a nice-to-have rather than a workflow, Google’s $14 tier already includes it.

The knowledge-work native

Teams living in research, learning and content — and already comfortable in Google Docs — should think twice before paying the Microsoft tax. Gemini Notebook’s agentic research and video generation have no direct equivalent in the Microsoft stack today.

Five questions to ask before signing

• What share of our seats will use this weekly? If the honest answer is under 40%, buy fewer seats.
• What is our fully loaded cost per seat, including base licences, consumption and admin overhead — not the add-on headline?
• Which data-hygiene work is required before grounding AI on our documents — and who owns it?
• How do we measure success? Define activation rate and time-to-draft metrics before rollout, not after.
• What happens to our discounts at renewal, given the 2026 changes to volume pricing?

Bottom line

Copilot is the better product for organisations whose work lives inside Microsoft’s estate — bought selectively, with clean data and honest measurement. Gemini is the better deal for everyone else, and the only sensible starting point for teams built on Google from day one. Neither pays for itself on autopilot: in 2026, the differentiator is not the model but the discipline of the organisation deploying it.