Quick answer: in Google Workspace, check the Gemini features already included in your edition. In Microsoft, distinguish Copilot Chat from a Microsoft 365 Copilot licence. For work across applications, test ChatGPT Business and Claude Team on the same task with the same permissions. Choose Enterprise when you can name the control you need.
Editorial review: 9 October 2026. We checked the providers’ public documentation. We have not run a comparative model benchmark or audited their contracts. The recommendations are BRTHLS purchasing criteria, not experimental results.
Problem
A useful answer in a personal chat does not establish that a tool is suitable for an entire company. Buying it changes who can consult documents, how access is revoked, what happens when usage runs out and who reviews an action before it executes.
Model names alone do not answer those questions. They also blur distinct products: a chat subscription, an assistant inside a productivity suite and an API for building an application.
Thesis
Choose the simplest plan that passes a test using your work and constraints. The outcome that matters is an accepted task at a known cost, within the right permissions.
There is no universal winner. If a tool needs fewer corrections but requires moving documents into another system, include that work in the comparison. For a custom integration, budget separately for the API, development and operation.
Framework
What are you actually buying?
The figures below are public USD rates displayed on the linked pages at review time, not a euro quote or a tax-inclusive budget. Confirm currency, applicable taxes, commitment, extra usage and contractual terms when purchasing.
| Option | Verified purchasing model | What to check |
|---|---|---|
| ChatGPT Business / Enterprise | Business Standard: USD 20 per seat/month billed annually; USD 25 billed monthly. Enterprise: a sales quote. | Business lists Microsoft 365 and Google Workspace connections. Check the data and actions each connection permits in your account. |
| Claude Team / Enterprise | Team Standard: USD 20 per seat/month billed annually; USD 25 billed monthly. Enterprise lists a seat charge plus usage at API rates. | Team and Enterprise include connectors; audit logs and custom retention are listed for Enterprise. |
| Gemini in Google Workspace | AI capabilities are included according to edition. Starter differs from Standard, Plus and Enterprise. | Check the applications in your edition and the administrator’s settings before assuming a task is covered. |
| Microsoft 365 Copilot | Copilot Chat is available at no additional cost with eligible plans. A Microsoft 365 Copilot licence adds capabilities. | Having E3 or E5 does not, by itself, mean having the additional Microsoft 365 Copilot licence. |
Data residency, training use, retention and permissions are separate questions. Request contractual evidence for each requirement your organisation has. “Data in Europe” is not an answer to all of them.
Where to start
| Your starting point | First evaluation | Reason to consider another option |
|---|---|---|
| Documents and collaboration in Google Workspace | Gemini in your current edition | It fails the task or control you defined. |
| Work concentrated in Microsoft 365 | The actual scope of Copilot Chat and licensed Copilot | Your test needs a capability the integration does not deliver. |
| Work spread across applications | ChatGPT Business and Claude Team | One handles the complete workflow better, including review, permissions and cost. |
| Specific administration or contractual requirements | A plan that documents those requirements | The provider cannot demonstrate the control or its scope. |
These are evaluation recommendations. They do not claim that a model hallucinates less, writes better Spanish or is safer for legal work than another. Those conclusions need appropriate evidence from your use case.
A test you can repeat
Download the pilot evaluation CSV. It contains proposed cases and acceptance criteria, with blank results. Open it in a spreadsheet or give it to an agent. The column names are in English for reuse.
| Proposed case | Evidence to retain | Condition for proceeding |
|---|---|---|
| Answer using authorised documentation | Question, answer and supporting source passage | The answer is verifiable and acknowledges missing information. |
| Ask about a restricted test document | Identity used and observed result | It does not expose content that identity must not access. |
| Prepare an action with external effects | Proposal, approval and execution record in a test environment | The workflow respects required approval and lets you verify what happened. |
Use synthetic or authorised data. Record the plan, date, configuration, review time and errors. A favourable demonstration is not proof of reliability across every case.
Why it matters now
The pages reviewed no longer support the old division between “assistant without integrations” and “integrated suite”. Nor do they justify presenting a Business price as an Enterprise rate. A maintained comparison must identify the product and plan behind each claim.
Anti-example
Choosing Claude because it is “better for contracts” without testing its citations, or rejecting ChatGPT because it “only connects through an API”, replaces evaluation with a label. Buying seats for everyone before checking permissions extends that mistake.
Protocol (3 steps)
- Define the work. Specify the input, acceptable output, permitted data and person who decides whether the result is useful.
- Compare under the same conditions. Run the template cases, retain evidence and record failures and human review.
- Record a decision. Choose, retest or reject. Before expanding usage, confirm the contract, access, cost, owner and exit route.
The cost of an accepted task includes the licence, additional usage, integration, operation and review. Do not claim savings without measuring the work needed to correct the output.
Related
- Free AI and operational tools: what is free and what needs a budget.
- AI Tool Sprawl: too many tools, too little decision quality.
- Spanish version of this guide.
Sources consulted
- OpenAI: business plans and pricing.
- Anthropic: Claude plans and administration controls.
- Google: Workspace edition comparison.
- Microsoft: Copilot Chat and Microsoft 365 Copilot.
Correction in this edition: we removed an unsupported public minimum for ChatGPT Enterprise, references to old plans and models, and unevidenced claims of superiority. The previous comparison is replaced by the plan table and evaluation protocol. The original publication date is retained.
Next step
Complete the template for a process that currently needs human review. If permitted data, ownership or acceptance evidence remain unclear, an AI audit can clarify those decisions before you buy more licences.