A South African company choosing between ChatGPT Business, Microsoft 365 Copilot and its own AI should start with two questions: which office suite its staff work in, and how much of the work depends on its own systems.
ChatGPT Business is the quickest way to give staff a capable general assistant under a company contract, and it can now be connected to company files in SharePoint or Google Drive. Copilot suits a company on Microsoft 365 whose file permissions are in order, and Gemini plays the same part for Google Workspace.
Your own AI is worth it in four cases. The answers have to come from your ERP or CRM, the permissions go deeper than files, the assistant has to act in your systems, or the data must be stored and processed in South Africa.
All three licences pass the first test a company should set. None of them trains its models on your data by default, and each comes with a data processing agreement. The real differences are in what each one can see, who in your company controls it, and where it processes your data.
This article assumes you're choosing a company plan. If some of your staff still use personal accounts, my earlier article on whether ChatGPT is safe for your company's data covers that risk and what POPIA expects of the employer.
What's the difference between ChatGPT Business, ChatGPT Enterprise and Microsoft 365 Copilot?
ChatGPT Business and Enterprise run as a separate app, which can reach company files and other apps once they're connected to it. Copilot works inside Microsoft 365 from the start, on the emails and files each person is allowed to open.
ChatGPT Business, which OpenAI called Team until it renamed the plan, gives a company a shared workspace with admin controls. Your admins set how long conversations are kept, and they can view, export and delete what staff type. ChatGPT Enterprise adds an audit log through OpenAI's Compliance API, and it's the plan OpenAI names when it describes choosing where data is stored.
Both plans include what OpenAI calls company knowledge. Once apps such as SharePoint, Google Drive, Slack or GitHub are connected, ChatGPT can search them and show which source each answer came from. OpenAI says it respects your existing permissions, so each person sees only what they're allowed to view in those apps.
Admins can also add custom connectors to other internal systems. That feature is still in beta, and OpenAI leaves vetting each connector to the company.
Microsoft 365 Copilot is an add-on licence. Each user needs a qualifying Microsoft 365 plan first, and Copilot then draws on that person's mail, files, chats and meetings through Microsoft Graph. Gemini in Google Workspace is the Google equivalent, and since January 2025 it has been included in Workspace Business and Enterprise plans without a separate add-on.
Is ChatGPT Enterprise secure and confidential?
ChatGPT Enterprise is secure by the usual measures. According to OpenAI's enterprise privacy page, it encrypts data at rest with AES-256 and in transit with TLS 1.2 or higher. It also has a SOC 2 Type 2 audit report and supports single sign-on through SAML. OpenAI doesn't train on Business, Enterprise or API data by default, and it signs a data processing addendum for all three.
Confidential needs a closer look, because it depends on who you're keeping things from. Against OpenAI's training and against the outside world, a company plan keeps your conversations confidential. Inside your company it doesn't, by design. On Business your admins can read and export what staff type, and on Enterprise the audit log records it.
Staff should know this before they use it. A manager who drafts a disciplinary letter in the company workspace has shared that draft with whoever administers the account. That's reasonable, but it belongs in your AI use policy so staff aren't caught out later.
So the encryption matters less than what your staff put in and who can then read it.
Is Microsoft 365 Copilot safer than ChatGPT for company data?
Copilot and ChatGPT Business sit on the same contractual footing. Microsoft acts as a data processor under its data protection addendum, and its documentation says prompts, responses and the data Copilot reads through Microsoft Graph aren't used to train foundation models.
Copilot also inherits your sensitivity labels, retention policies and audit settings. Its conversations are stored with your other Microsoft 365 content, where admins can search them in Purview.
Where Copilot differs is in how much it can reach without anyone setting it up. Microsoft says Copilot "only surfaces organizational data to which individual users have at least view permissions". That sounds like a safeguard, and it is one, provided your permissions are right.
Let's say a SharePoint site holding salary reviews was shared with "everyone in the company" in 2019, and it was never changed. Until now, an employee would have had to go looking for it. With Copilot, a question about pay increases can bring it straight into the answer.
Copilot is a very good librarian working in a building where some rooms were left open to everyone years ago, and nobody remembers which.
Microsoft's own advice is to use SharePoint's permission models so that the right people have the right access. So before rolling Copilot out, have someone check which sites and folders are shared with everyone.
Copilot is therefore safer than ChatGPT only for a company whose file permissions are tidy. For a company whose permissions aren't, Copilot reaches the open rooms from its first day, while ChatGPT reaches them only once someone connects SharePoint.
Where does Gemini in Google Workspace fit?
Gemini works the same way for a company on Google Workspace. Google treats prompts as customer data under its Cloud Data Processing Addendum. Its privacy hub for Workspace says your content isn't reviewed by people or used to train models outside your domain without your permission. Admins set how long Gemini conversations are kept, anywhere from 90 days to indefinitely.
Like Copilot, Gemini only retrieves content the user has access to, so the same advice applies to Google Drive. Shared drives and links set to "anyone in the organisation" are worth checking before you switch it on. The same goes for ChatGPT once it's connected to SharePoint or Drive, because it follows the permissions in those apps however loosely they were set.
Which AI keeps your data in South Africa?
None of the three licences, going by their own pages today, can promise that a South African company's prompts are both stored and processed in South Africa.
OpenAI lets new Enterprise and Edu customers choose where some data is stored, in regions including Europe, the UK, the US, India and the UAE. South Africa isn't on the list.
Where the model actually runs (OpenAI calls it inference residency) can be fixed only to Europe, the US or the UAE. Even then, OpenAI says some supporting processing, such as authentication and routing, may still happen outside the chosen region, and anything passed to a connected app falls under that app's own terms.
Microsoft routes Copilot requests to the closest data centres but says they "can call into other regions where capacity is available during high utilization periods". It also says customers outside the EU may have their queries processed in the US, the EU or other regions.
That may change for Copilot. In November 2025 Microsoft announced in-country processing of Copilot prompts for South Africa during 2026. Its update of April 2026 named other countries for the end of 2026 and didn't mention South Africa, so check Microsoft's announcement for its current status.
Google's privacy page for Gemini in Workspace doesn't say where prompts are processed.
Where the data is processed matters under section 72 of POPIA. It allows personal information to go to a third party outside South Africa only under set conditions, such as a binding agreement giving substantially similar protection.
A company plan's data processing agreement is usually how that condition is met, and your lawyer is the right person to judge whether it's enough for your data. I'm not a lawyer, and this isn't legal advice.
How do the options compare?
What each vendor says about a company plan (checked 20 November 2026; vendors change plans often, so check their own pages before deciding)
| Trains on your data by default? | What it can see | Who in your company can see conversations | Where a South African company's data is processed | |
|---|---|---|---|---|
| ChatGPT Business | No | What staff type or upload, plus company apps connected to it (SharePoint, Google Drive and others) | Admins can view, export and delete | Outside South Africa (OpenAI offers no South African region) |
| ChatGPT Enterprise | No | What staff type or upload, plus company apps connected to it (SharePoint, Google Drive and others) | Admins, with an audit log | Outside South Africa; storage region can be chosen, model processing in Europe, the US or the UAE at best |
| Microsoft 365 Copilot | No | Mail, files, chats and meetings the user can open | Admins, through Purview | Nearest region, other regions at high load; South African processing announced, no date given |
| Gemini in Workspace | No | Workspace content the user can open | Admins control them and set how long they're kept | Not stated on Google's privacy page |
| Your own AI, built by CBOS | No; built so no outside AI model trains on it | The systems you connect, with permissions set per user | The people you decide | Stored and processed in South Africa |
When is an off-the-shelf licence the better choice?
A licence is the better choice when the work is the general reading and writing every company does. Drafting emails, summarising a long report, tidying a proposal and preparing for a meeting don't depend on anything particular to your business, and ChatGPT, Copilot and Gemini all do them well.
A licence also works from the week you buy it, and the vendor maintains and updates it for you. For a company on Microsoft 365 with tidy permissions, Copilot adds the benefit of working on the company's mail and files without connectors to set up.
If the job would look the same in any other company, buy the licence that matches your office suite. Then spend your effort on the policy and the permissions.
When is it worth building your own AI?
Building your own AI is worth it when the job is specific to how your company runs. That usually shows up in one of four ways:
- The answers have to come from your ERP, CRM or stock system, and documents alone won't do.
- Permissions go deeper than files, because an ERP controls access record by record (a branch's debtors, a client's prices) and a file permission can't express that.
- The assistant should act in your systems, with a person confirming the steps that matter.
- The data has to be stored and processed in South Africa.
Let's say a distributor wants its branch managers to ask about stock levels and overdue accounts. A licence can read the reports someone exports for it, or the files it's connected to. It can't know that the Durban manager may see Durban's debtors and not Cape Town's, because that rule lives in the ERP.
Your own AI, built on the ERP, can follow that rule. A custom connector could also reach the ERP, but someone would have to build it and carry the ERP's rules into it, and at that point the company is building its own AI anyway.
This is the kind of system CBOS builds, a private AI on a company's own data, stored and processed in South Africa, with no outside AI model training on it. It has enterprise guardrails and permissions set per user, so staff get answers only from information they're allowed to see, with the source next to each answer.
CBOS's page on a private AI assistant built on your own data explains how it decides who sees what, which answers it may give and which actions need a person's confirmation. It connects to your ERP, CRM, document stores, Microsoft 365 and older applications, and it's designed like any of the bespoke systems CBOS builds, around the company's own processes.
Isn't building your own AI slower and harder than buying a licence?
It is slower to start and costs more up front, and a vendor shipping new features often will add things your own AI won't have. Those are fair objections, and they're the reason your own AI should cover only the part a licence can't do.
A company can run both. Staff use ChatGPT, Copilot or Gemini for general writing, and your own AI handles the jobs tied to your records, your permissions and your rules about where data lives.
How should a South African company choose?
Work through these six questions in order, and the choice usually makes itself:
- Which office suite do your staff work in? Microsoft 365 points to Copilot, Google Workspace to Gemini, and neither to ChatGPT Business.
- Are your file permissions in order? If not, fix them before switching on Copilot or Gemini, or connecting ChatGPT to your files.
- Does the work need data from your own systems, beyond documents? If it does, a licence will fall short and your own AI is worth considering.
- Must the data be stored and processed in South Africa? Check each vendor's current residency page, and ask your lawyer about section 72.
- Who in your company will be able to read staff conversations, and do staff know that?
- Which one job will you start with, and how will you know it worked?
The last question is where a discovery call with CBOS fits, because the first job decides whether a licence is enough or your own AI is worth scoping.
Frequently asked questions
What's the difference between ChatGPT Business and ChatGPT Enterprise?
Both keep company data out of OpenAI's training by default, come with a data processing addendum and can connect to company apps such as SharePoint and Google Drive. Enterprise adds an audit log through OpenAI's Compliance API, and it's the plan OpenAI names for choosing where data is stored. Check OpenAI's site, because features change.
Is ChatGPT Team the same as ChatGPT Business?
Yes. OpenAI renamed ChatGPT Team to ChatGPT Business, so older articles comparing "Team" and "Enterprise" are describing the same two plans.
Is Copilot or ChatGPT better for business?
For a company on Microsoft 365 with tidy permissions, Copilot has the advantage of working on its mail and files without connectors to set up. For a company on another suite, or one that wants a standalone assistant, ChatGPT Business is simpler, and it can be connected to company files too. Neither trains on company data by default.
Can you use ChatGPT and Copilot together?
You can, but each needs its own licences, admin settings and data processing agreement. Your AI use policy should say which tool is for which kind of work.
Is ChatGPT Plus confidential?
Not for company work. Plus is a personal plan, so OpenAI can train on what you type by default, and your employer has no contract with OpenAI covering it. Company data belongs on a Business or Enterprise plan, or in your own AI.
Can ChatGPT share my data with others?
On Business and Enterprise, OpenAI doesn't train on your data by default and handles it under its data processing addendum. Inside your company, your admins can view and export conversations, so treat the workspace as shared with them. Connected apps are searched with each person's own permissions.
What is the safest AI chatbot for a business?
The safest one is on a company account with a data processing agreement, no training on your data and permissions that are in order. ChatGPT Business, Copilot and Gemini in Workspace all meet the first two. If your data must stay in South Africa, your own AI is the option that can promise it.
What are the main risks of using AI in business?
The main risks are staff putting company data into personal accounts, AI surfacing files that were shared too widely, and answers going to clients or the board without anyone checking them. A clear AI use policy and tidy permissions deal with most of them.