The question is not which assistant is best. It is how many seats.
List price hides the real comparison. Fully loaded, Google Workspace with Gemini, Microsoft 365 with Copilot, and ChatGPT Enterprise span roughly four to one per seat. The productivity base the workforce already runs decides fit, but the seat count is the lever that actually moves the invoice: active weekly use settles at 35 to 45 percent of licensed seats after ninety days, so a rollout priced to headcount pays for a majority of seats that never use the tool.
Prepared by Redress Compliance · August 9, 2026 · GenAI advisory. Based on roughly 25 to 35 enterprise GenAI procurements run 2024 to 2025.
Executive summary
Fully loaded, the three span roughly four to one, and that spread is real money. Workspace Business Plus with Gemini bundled is about 22 dollars per user per month, M365 E5 with Copilot is 87, with M365 E3 plus Copilot at 66 and ChatGPT Enterprise standalone around 55 at a 150-seat minimum.
The productivity base decides fit: Copilot suits Microsoft 365 estates because it grounds on the customer's own Graph data with no data movement, Gemini suits Workspace and is now effectively free with the existing bill.
And ChatGPT Enterprise runs standalone with the strongest model coverage but no native productivity integration.
For a ten thousand-seat enterprise the gap between the cheapest and most expensive configuration is 7.8 million dollars a year, every year, before any consideration of which product produces the best operational outcome.
Seat count is the lever, because active weekly use settles at 35 to 45 percent after ninety days.
The seats that use an assistant cluster in marketing, sales operations, finance analysis and product management; the seats that do not are operations, customer service, manufacturing and most of HR, and the other 55 to 65 percent produce no return and are a structural overspend.
The buyer-side framework deploys to the measured active population, not to every E3, E5 or Workspace seat, and secures a contractual right to scale down at each anniversary triggered by usage telemetry.
Sizing GenAI seats to measured weekly use rather than headcount typically removes more than half of the licensed-but-idle seats, and Microsoft has accepted a scale-down right in well-prepared 2025 and 2026 renewals.
All three indemnify against IP claims, but the scope, caps and exclusions differ, and none cover trade secrets.
Microsoft's Customer Copyright Commitment, Google's Generative AI Indemnification and OpenAI's Copyright Shield all indemnify paid customers against third-party IP claims arising from outputs, subject to standard exclusions for prohibited use, modification of outputs and disabling safety features.
And none indemnify against trade-secret claims, defamation, or claims from the customer's own fine-tuning on third-party data.
Data residency also diverges: Copilot follows the M365 tenant geography with the EU Data Boundary GA, Gemini uses Workspace residency settings, and ChatGPT Enterprise defaults to the United States with limited regional options.
So the default residency boundary is rarely sufficient at enterprise scale and is a contract conversation, not a product one.
Negotiate the AI add-on inside the suite renewal, because separating them hands the vendor the lever.
The AI add-on is exactly what the publisher uses to escalate the broader productivity contract, so buyers who negotiated it separately from the Microsoft EA or Workspace renewal gave up 10 to 25 percent of available discount.
Exit posture is the hidden cost and the least visible question at procurement: Copilot is the most locked because of the deep Microsoft 365 integration, ChatGPT Enterprise the least because it is standalone and the data does not leave an existing tenant, and Gemini sits between.
Run the four decision questions in order, productivity base, then active population, then residency envelope, then exit posture, because answering them out of sequence produces the wrong commercial outcome.
Fully loaded per seat, at 2026 list
| Configuration | Productivity base | AI add-on | Loaded $/user/mo | Per user per year |
|---|---|---|---|---|
| Workspace Business Plus + Gemini bundled | $22.00 | Included | $22.00 | $264 |
| Workspace Enterprise + Gemini Enterprise | ~$23.00 | $27.00 typical | ~$50.00 | $600 |
| M365 E3 + Copilot | $36.00 | $30.00 | $66.00 | $792 |
| M365 E5 + Copilot | $57.00 | $30.00 | $87.00 | $1,044 |
| ChatGPT Enterprise standalone | None | ~$55.00 (150+ seat) | $55.00 | $660 |
The headline list price hides the structural difference: Copilot requires a Microsoft 365 base, Gemini Enterprise sits on a Workspace base or runs standalone, and ChatGPT Enterprise has no productivity base at all.
Copilot grounds on the customer's own files, emails and chats through Microsoft Graph with no data movement, which is why it feels native.
Google's early-2025 move bundled Gemini into Workspace Business Plus and Enterprise at no incremental price, so for a Workspace customer the productivity assistant is now effectively free with the existing bill and the higher-tier Gemini Enterprise SKU is a 36 dollar attach.
And ChatGPT Enterprise sells direct with the strongest model coverage in a single subscription but no embedding inside Word, Excel, Docs or Sheets, so power users move between it and their suite manually.
The full economics of the Microsoft line sit in the Copilot true cost analysis and the licensing detail in the Copilot licensing guide.
The eight contract terms to redline
- Seat scale-down right: a defined right to reduce seat count by up to thirty percent at each anniversary, triggered by usage telemetry. Microsoft has accepted it in well-prepared renewals and ChatGPT Enterprise typically accepts it in the paper directly.
- Price protection and MFN: a cap on list increases, standard ask no increase with a maximum three percent CPI-linked uplift on renewal, plus most-favored-customer language for the same effective discount as comparable customers in the same industry and region.
- Data residency and training opt-out: a defined geographic boundary for prompt processing, inference and audit logs, because the default is rarely sufficient, plus explicit contractual confirmation that customer data is not used to train future foundation models, which all three offer in standard form.
- IP indemnity scope and audit-log retention: defined exclusions, caps and claim process, since the three programs differ, and extended retention, standard ask 365 days minimum against defaults of 30 on Copilot and ChatGPT and 180 on Gemini.
- Sub-processor controls: a defined consent process for sub-processor changes and a right to terminate without penalty for a change that materially expands the data perimeter. The full clause library sits in the AI platform contract negotiation playbook.
Copilot vs Gemini vs Amazon Q
Side-by-side enterprise AI assistant economics on price, scope, and renewal risk, with the buyer-side decision framework.
Get the white paper →The four-question decision framework, in order
The procurement decision turns on four questions that run in sequence, and answering them out of order produces the wrong commercial outcome.
First, what productivity suite does the workforce already use, the grounding question: Microsoft 365 customers have a structural fit with Copilot, Workspace customers with Gemini.
And a hybrid stack should evaluate ChatGPT Enterprise standalone and accept the productivity-surface integration as a known trade-off.
Second, which user populations actually use AI assistance week to week, the seat-count question, because active weekly usage sits at 35 to 45 percent after ninety days and the other 55 to 65 percent of seats produce no return.
So deploy to the measured active population with a contractual right to scale up or down at each anniversary.
Third, what is the data residency and compliance envelope, the data-architecture question, because regulated industries, EU jurisdictions or classified data flows need the residency and audit-log posture evaluated independently and the default boundary is rarely enough.
Fourth, what is the contractual exit posture, the vendor-lock question, where Copilot is most locked through its deep Microsoft 365 integration, ChatGPT Enterprise least because it is standalone and the data does not leave an existing tenant, and Gemini sits between.
The least visible question at procurement and the most expensive one to ignore.
The comparison against Anthropic sits in the Claude versus ChatGPT comparison, the OpenAI contract terms in the OpenAI procurement playbook, and Microsoft seat right-sizing in the M365 license optimizer.
- Percentile standing for your exact deal size and industry, from real closed transactions
- Scenario simulation before the call: test alternative terms and see the financial impact of each
- A negotiation playbook, talking points, and a two page executive brief on day one
What we saw across enterprise GenAI procurements, 2024 to 2025
Across roughly 25 to 35 enterprise GenAI procurements we ran between 2024 and 2025, the deployment was sized to headcount rather than usage, and the common advice is what drove that. The common advice is to standardise on one assistant for the whole workforce and roll it out to every seat.
We disagree:
Where active weekly use settled after ninety days, leaving most licensed seats idle and producing no return on the investment.
Of available discount lost by buyers who negotiated the AI add-on separately from the suite renewal, where the publisher's escalation lever lives.
Enterprise-wide deployment paid for a majority of seats that never used the tool week to week, so the buyer-side move is to size to the measured active population, secure a contractual right to scale down, and negotiate the AI add-on inside the suite renewal where the leverage lives.
Five pitfalls recur at customers who ran the procurement as a standalone evaluation rather than a contract: deploying AI enterprise-wide before usage data exists, when year-one deployment to a defined power-user population with an anniversary scale right is the correct posture.
Treating the comparison as a feature shootout, when the differentiators that matter at scale are commercial terms, residency and indemnity, not feature deltas that close within six months; negotiating the AI contract separately from the EA or Workspace renewal.
Accepting the default residency boundary, which is a contract conversation not a product one; and skipping IP indemnity diligence, when the exclusions, caps and claim process differ and belong as first-class contract terms.
Default residency boundaries were accepted in most first drafts and then reopened as a contract issue at material cost. A full rollout priced to headcount is the most common GenAI overspend, and the broader Google Cloud context sits in the Google Cloud practice.
Your first five moves
- Confirm which productivity suite the workforce already runs, since that decides the natural fit, Copilot for Microsoft 365, Gemini for Workspace, ChatGPT Enterprise for a hybrid stack.
- Measure active weekly AI use by population before sizing seats, not after, and deploy to the measured active population rather than every knowledge worker.
- Secure a contractual right to scale seats up or down at each anniversary, triggered by usage telemetry, the term that turns a 35 to 45 percent active rate into a right rather than a loss.
- Define the residency boundary for prompts, inference and audit logs as a contract term, and diligence the IP indemnity scope, caps and claim process for the chosen vendor.
- Negotiate the AI add-on inside the broader suite renewal, never as a separate deal, and document the exit posture and data portability before signing. The GenAI practice runs the procurement with you.
Frequently asked questions
How much do Copilot, Gemini and ChatGPT Enterprise cost fully loaded?
At 2026 list, fully loaded per user per month: Workspace Business Plus with Gemini bundled is about 22 dollars, Workspace Enterprise with Gemini Enterprise around 50, M365 E3 with Copilot 66, M365 E5 with Copilot 87, and ChatGPT Enterprise standalone around 55 at a 150-seat minimum.
The spread is roughly four to one, so for a ten thousand-seat enterprise the gap between the cheapest and most expensive configuration is about 7.8 million dollars a year, before any consideration of operational outcome.
Which enterprise AI assistant is the right fit?
The one grounded in the productivity suite the workforce already runs.
Microsoft 365 estates have a structural fit with Copilot, which grounds on the customer's own Graph data with no data movement; Google Workspace estates fit Gemini, now effectively free with the existing Workspace bill.
And organizations on a hybrid stack should evaluate ChatGPT Enterprise standalone, which has the strongest model coverage but no native embedding inside Word, Excel, Docs or Sheets.
The fit decision comes before the price decision.
How many AI assistant seats should you actually buy?
As many as have measured weekly use, which is 35 to 45 percent of licensed seats after ninety days, not one per knowledge worker.
The active seats cluster in marketing, sales operations, finance analysis and product management; operations, customer service, manufacturing and most of HR barely use them.
Deploy to the measured active population and secure a contractual right to scale down at each anniversary, because sizing to weekly use rather than headcount typically removes more than half of the licensed-but-idle seats.
Do Copilot, Gemini and ChatGPT Enterprise all offer IP indemnity?
Yes, through Microsoft's Customer Copyright Commitment, Google's Generative AI Indemnification and OpenAI's Copyright Shield, all indemnifying paid customers against third-party IP claims from outputs, subject to standard exclusions for prohibited use.
Modification of outputs and disabling safety features.
None indemnify against trade-secret claims, defamation, or claims from the customer's own fine-tuning on third-party data. The scope, caps and claim process differ, so diligence the indemnity as a first-class contract term rather than a marketing claim.
Should the AI add-on be negotiated with the suite renewal?
Yes. The AI add-on is the lever the publisher uses to escalate the broader productivity contract, so negotiating it separately from the Microsoft EA or Google Workspace renewal gave up 10 to 25 percent of available discount in our procurements.
Negotiating the two together produces a materially better outcome. The AI line is also where price protection, the scale-down right, residency terms and indemnity scope should all be pinned, inside the renewal where the leverage lives.
Which AI assistant has the worst vendor lock-in?
Microsoft Copilot is the most locked, because of its deep Microsoft 365 integration; ChatGPT Enterprise is the least, because it is standalone and the data does not move out of an existing tenant; Google Gemini sits between the two.
Exit posture is the least visible question at procurement and the most expensive one to ignore, so document the exit position and data portability before signing rather than discovering the cost at the first renewal or migration.
