ChatGPT at Work: Business, Enterprise, Privacy and Use
In most companies ChatGPT is already in use, just not officially. People draft quotes, summarise inspection reports or translate supplier emails with their own accounts. So the real question is rarely whether to allow ChatGPT, but which plan to use, how to configure it and what to use it for. This guide covers the plans, administration, data handling and the main features, based on OpenAI's own pages as of 31 October 2026.
What does ChatGPT at work actually mean?
Using ChatGPT at work means running a dedicated, centrally managed workspace where staff sign in with company identities instead of personal logins. OpenAI sells two plans for this, ChatGPT Business and ChatGPT Enterprise. Companies use it mostly for writing, research and data analysis, and increasingly for repeatable tasks connected to internal systems.
Typical everyday jobs in a mid-sized company:
- Drafting and editing: quote cover letters, complaint replies, job adverts, internal announcements.
- Summarising: long standards, technical specifications, meeting minutes or audit findings.
- Analysing: grouping a spreadsheet of returns by root cause, or checking a CSV export from the ERP system for anomalies.
- Researching: pulling together market and competitor information from the web, with citations.
- Translating: supplier correspondence, operating manuals, product data sheets.
The difference from a personal account is not the interface. It is the contract and the controls: who has access, how long data is kept and whether content can be used for training. If you are still deciding whether ChatGPT is the right tool at all, or whether another assistant, an automation platform or a self-hosted model fits better, start with choosing AI tools for your company. This article goes one level deeper. It assumes ChatGPT is the chosen assistant and explains how to set it up properly.
How is a company workspace put together?
A company workspace has four parts: the models, working features such as projects and GPTs, apps that connect to internal systems, and the admin console. The model line-up is largely the same on both plans. What differs is context size, administration, security features and support.
Models. According to OpenAI's pricing page, Business and Enterprise include GPT-6.1 Sol, GPT-6 Astra and GPT-5.6, among others, plus legacy models. The line-up changes often, so plan around tasks rather than a specific model name.
Context window. This is where the plans diverge noticeably. For the fast Instant models, OpenAI lists a total context window of 54,000 tokens on Business (roughly 40 pages of input) and 128,000 tokens on Enterprise (roughly 250 pages). Reasoning models get 256,000 tokens on both, around 320 pages. If your teams routinely feed in whole tender packs or contract bundles, these numbers matter.
Working features. Both plans include file uploads, data analysis, projects and shared projects, deep research, scheduled tasks, image generation, and creating and sharing GPTs inside the workspace.
Apps and workspace agents. Apps connect ChatGPT to internal sources; OpenAI names Google Workspace, Slack, GitHub and Microsoft 365 for Business. Workspace agents handle custom workflows.
Admin console. This is where admins manage members, switch features on or off, approve or block apps and set retention periods.
Business or Enterprise: which plan fits?
ChatGPT Business is the self-serve plan for teams from two people upwards, aimed by OpenAI at organisations with 2 to 200 staff. ChatGPT Enterprise is negotiated with sales and adds identity management, finer permissions, compliance logging, data residency and dedicated support. OpenAI describes Free, Go and Plus as plans for individuals.
If you are weighing Anthropic's assistant instead, Claude for business covers its plans and data rules. The overall pattern of a team plan plus a negotiated enterprise contract is similar; the differences lie in admin features and contract details.
| Criterion | ChatGPT Business | ChatGPT Enterprise |
|---|---|---|
| List price (31 Oct 2026) | $20 per standard seat per month billed annually, $25 billed monthly | custom, via sales |
| No training on business data | by default | by default |
| SAML SSO, domain verification | yes | yes |
| SCIM, role-based access, IP allowlisting | no | yes |
| Data residency | no | yes, ten regions |
| Compliance API logs | no | yes |
The figures come from OpenAI's pricing page, where a premium seat is listed at $100 per month billed annually or $125 billed monthly. Local currency prices may differ, so check at checkout. OpenAI describes the premium seat as five times the usage of a standard seat without the five-hour limit, and both seat types can be mixed in one workspace. That is useful in practice: heavy users get premium seats, everyone else standard.
Business is enough when you have up to a few dozen users, an identity provider for single sign-on, and you are happy to add and remove users by hand. For many companies that is a sensible starting point.
Enterprise becomes necessary when HR or IT expects automatic provisioning through SCIM, your privacy assessment requires data to stay in a specific region such as the EU, UK or US, auditors want compliance logs, or you need the ISO 27001, 27017, 27018 and 27701 certifications that OpenAI's comparison table marks for Enterprise only. The connector directory is also listed as Enterprise-only.
ChatGPT or the API: the API is not a ChatGPT plan. It is how developers build OpenAI models into their own applications, billed by usage and governed by its own data retention rules, covered below.
How does OpenAI handle your data?
OpenAI commits not to train on inputs and outputs from ChatGPT Business, Enterprise or the API unless you explicitly opt in. Data is encrypted, and a data processing addendum is available for all three. That means ChatGPT can be operated in line with the GDPR, but responsibility for how it is used stays with you.
Key points from OpenAI's enterprise privacy page, last updated on 8 January 2026:
- Training: data from Business, Enterprise and the API (after 1 March 2023) is not used for training by default. The same page states that data from services for individuals is used.
- Encryption: AES-256 at rest, TLS 1.2 or higher in transit.
- Retention: on Business and Enterprise, admins decide how long data is kept. Deleted conversations are removed within 30 days unless OpenAI is legally required to keep them; for Business, OpenAI also reserves longer retention where reasonably necessary to protect its services or others from harm.
- Audits: OpenAI reports a completed SOC 2 Type 2 audit for both Business and Enterprise.
- Data processing addendum: available for Business, Enterprise and the API. Under Article 28 of the GDPR, a controller may only use processors that provide sufficient guarantees of appropriate technical and organisational measures, so if you process personal data of EU residents you need it signed.
- Data residency: OpenAI's comparison table lists ten regions, including the US, EU, UK, Japan, Canada and Australia, for Enterprise only.
One detail is easy to miss. On Business, workspace admins can view, export and delete members' conversations, according to OpenAI. That helps with accountability, but it also counts as monitoring in many jurisdictions. In Germany, for example, section 87 of the Works Constitution Act gives works councils a say over technical systems designed to monitor employee behaviour or performance. Settle this before rollout. This section is general orientation, not legal advice.
GPTs, apps and agents: turning chat into a tool
Inside a company workspace you can package repeat tasks as custom GPTs, connect company systems through apps, and repeat routines with scheduled tasks and workspace agents. That turns ChatGPT from a writing aid into a tool for specific processes, though not into a fully automated pipeline.
GPTs. A GPT is a saved set of instructions, files and tools. Picture a "complaint reply" GPT that knows your tone rules, holds your warranty terms as a PDF and has three good example replies. OpenAI says the workspace privacy commitments also apply to GPTs used inside it. If admins allow public sharing, externally published GPTs may be reviewed further, so most companies keep that switched off.
Apps. Apps let ChatGPT read from connected sources such as Microsoft 365, Google Workspace or Slack. OpenAI states that existing permissions are respected and each user has to authenticate with the connected application first. Data retrieved through apps is not used for training by default, and admins choose which apps are enabled.
Projects. A project keeps chats and files for one piece of work together, for example a bid. Shared projects let a team work from the same material.
Scheduled tasks and workspace agents. Scheduled tasks can trigger, say, a weekly competitor round-up; workspace agents run custom workflows. ChatGPT is not built for multi-step processes that write into ERP or CRM systems with nobody watching. For that you need your own application on the API or a workflow platform such as n8n, covered in n8n for business; the page on AI agents shows what such processes look like for quotation and technical service.
Where the API differs. Per OpenAI's data documentation, API content is kept for up to 30 days for abuse monitoring, and the Responses API stores response data for at least 30 days by default unless store is set to false. Zero Data Retention can be requested for eligible endpoints.
What do you need in place first?
You need more organisation than technology: a named owner, a short acceptable use policy, a clear privacy basis and an identity provider for sign-in. Your data does not have to be cleaned up before you start, but tidy permissions in shared drives matter as soon as you connect apps.
Organisation:
- one person who runs the workspace and collects questions
- an acceptable use policy of one or two pages: which data may go in, which may not, and how outputs are checked
- sign-off from your data protection lead and, where relevant, employee representatives
- a short training session on what ChatGPT does well and where it gets things wrong
Classify your data. A simple traffic-light scheme works: public and internal text is fine; personal data only with a processing agreement and a clear purpose; design files, HR records and contracts with confidentiality clauses only with explicit approval, or not at all.
Technology:
- an identity provider for SAML single sign-on, usually the one your staff already use
- a verified company domain so people cannot bypass the workspace
- a decision on which apps are enabled and which stay blocked
How should you roll it out?
Roll ChatGPT out in five steps: set the framework, run a pilot, collect use cases, enable GPTs and apps selectively, then extend access. Each step has a concrete output, so after a few weeks you decide based on your own experience rather than vendor claims.
- Set the framework. Choose the plan, sign the processing addendum, write the policy, involve employee representatives. Output: signed documents and a configured workspace.
- Run a pilot. Ten to twenty people from different departments, including a few sceptics. Keep apps disabled for now.
- Collect use cases. After two to four weeks, ask what people used it for, what cost time rather than saving it, and where it went wrong. Gather examples in one shared list.
- Enable GPTs and apps. Build one GPT each for the three most frequent tasks, with examples and clear rules. Enable apps only where source permissions are correct.
- Extend access. Seats for everyone who needs one, premium only for heavy users. Revise the policy based on what the pilot taught you.
In our own operations, a pilot that is too short has proven misleading: the first days are driven by curiosity, and the genuinely useful patterns only show up later.
What are the limits and risks?
ChatGPT produces plausible answers that are not always correct, and it only knows what you give it. The biggest risks are not technical but behavioural: unchecked outputs, confidential data in the wrong plan, and apps with overly broad permissions.
- Wrong facts and figures. Always check standard numbers, legal references, prices and calculations. Spot-check results from data analysis too.
- Shadow use. As long as personal accounts are tolerated, company data ends up where OpenAI explicitly allows training. A company workspace alone does not fix that; you also need a clear rule that personal accounts are not for work.
- Over-broad permissions. Apps respect source permissions. If a folder of salary data is accidentally shared company-wide, ChatGPT will find it too. Clean up permissions before connecting anything.
- Data residency. If data must stay in a specific region, you need Enterprise; OpenAI's comparison table does not list residency for Business.
- Lock-in. GPTs and projects do not move easily to another vendor. Keep instructions and example documents outside ChatGPT as well.
- Not everything belongs in the cloud. For data that must never leave your premises, a self-hosted model is the alternative; the page on sovereign AI describes that route.
How do you get started?
Start with ChatGPT Business for a pilot group, a two-page policy and a signed data processing addendum. After a few weeks, check whether requirements such as SCIM, regional data residency or compliance logs justify moving to Enterprise. Plan larger automation separately, on the API or an agent platform.
A short checklist for week one:
- choose the plan and create a workspace with a verified domain
- configure single sign-on and invite pilot users
- sign OpenAI's data processing addendum
- set the retention period in the admin console deliberately
- keep apps disabled for now and switch off public GPT sharing
- share the policy and run a one-hour introduction
If ChatGPT is meant to be one building block of a wider AI strategy, it pays to plan plan selection, policy and follow-on projects together. The page on a fractional head of AI explains what that kind of support looks like.
Frequently asked questions
Can staff use their personal ChatGPT accounts for work?
It is a bad idea. OpenAI says it uses data from its services for individuals to improve its models, while Business and Enterprise data is excluded from training by default. Personal accounts also leave the company with no admin control and no data processing agreement.
When does Enterprise make sense instead of Business?
OpenAI positions Business for teams of 2 to 200 people. In practice the switch is usually driven by requirements such as automated provisioning via SCIM, role-based access, compliance logs or regional data residency rather than headcount.
Can managers read employees' chats?
On ChatGPT Business, OpenAI says workspace admins can view, export and delete members' conversations; Enterprise provides access through the Compliance API. Agree in writing how this is used before rollout, and involve employee representatives where they exist.
Is ChatGPT a substitute for building on the API?
No. ChatGPT is a tool for people at their desks; the API is a building block for processes that run inside your own systems without a chat window.
Sources
- OpenAI: Enterprise privacy at OpenAI (updated 8 January 2026) openai.com
- OpenAI: Business and Enterprise pricing openai.com
- OpenAI Platform Docs: Your data platform.openai.com
- Regulation (EU) 2016/679 (GDPR), Article 28: Processor eur-lex.europa.eu
- German Works Constitution Act (BetrVG), section 87 gesetze-im-internet.de