ChatGPT training for NZ teams.

What ChatGPT training covers

ChatGPT training for business teams is hands-on practice in your own ChatGPT Business workspace, on your own files, with the settings agreed before anyone starts. OpenAI’s Business overview lists GPTs, Projects, apps, company knowledge, deep research, ChatGPT Agent and Codex on a Standard seat. Someone who only pastes text into a chat box uses a small part of that. Training closes the gap without loosening your rules.

This page covers ChatGPT. Our wider AI training for business teams covers Copilot, Claude and Gemini.

Who it is for

• Teams moving off personal accounts. Staff were using Free or Plus for work, and you have bought a Business workspace to fix it.
• Pilot workspaces. A small group on the two-seat minimum or more, testing before you buy seats for everyone.
• Teams with seats but shallow use. Nobody uses projects, apps or data analysis yet.
• Workspace owners and admins who need to understand the settings they now control.

Still choosing? Read our ChatGPT guide, then ChatGPT versus Claude or, if you run on Microsoft 365, Copilot versus ChatGPT.

Session options, starting in the admin console

Pick the options that fit. Nobody opens a chat until the workspace settings are agreed.

Workspace settings record

Training, retention, apps, actions and GPT sharing, as set and agreed.

Seat plan

Who needs Standard or Premium, and which purchased seats sit unassigned.

Exercise pack

Role tasks from your own exports, documents and enquiries, with personal details removed.

Shared GPTs

Two or three tested GPTs for repeat tasks, built from our prompt library template.

Adoption baseline

Timed target tasks and current usage, for comparison later.

The workspace settings we check first

From OpenAI’s documentation as checked on 19 October 2026:

• Training: Business content is not used to train OpenAI’s models by default, including data reached through apps, unless you opt in, for example through feedback.
• Retention: admins decide how long workspace data is retained. OpenAI removes deleted or unsaved conversations within 30 days unless longer retention is legally required.
• Who can read chats: members see their own conversations; admins can view, access, export and delete conversations in the workspace.
• GPT sharing: members can build and share GPTs inside the workspace. If admins allow public sharing, published GPTs can leave it.
• Apps and actions: Business workspaces start with apps switched on, and admins control availability for the whole workspace. Read and write actions are enabled separately, and Google Drive actions for Docs, Sheets and Slides are on by default. Some Outlook, Teams and SharePoint actions need consent from your Microsoft Entra administrator.
• Custom MCP apps: Business can add them, but OpenAI does not verify them.
• Seats: OpenAI bills every purchased seat each cycle, assigned or not.

A sample exercise: checked analysis of a monthly export

This suits finance and operations staff. It uses data analysis inside a Project and builds the habit that matters most with numbers: a control total you check yourself.

1. Export a real file. Last month’s jobs or sales export from Xero, ServiceM8, Cin7 or your CRM, as CSV. Delete columns the task does not need, such as phone numbers, emails and addresses.
2. Create a Project. Call it “Monthly numbers” in your Business workspace and upload the file.
3. Run the prompt below. Do not accept the tables yet.
4. Check the control total. Sum the amount column yourself in Excel. If ChatGPT’s total differs by a cent, find out why before reading anything else.
5. Spot-check rows. Trace five rows from the overdue list and two from the variance list back to the source system.
6. Save or rewrite. If it passes, save the prompt as the Project’s instructions and rerun it next month.

I've uploaded jobs-export-september.csv. Each row is one job.

Before analysing, tell me the number of rows, the columns you will use, and any rows you will exclude (blank amounts, duplicates, test jobs) with the reason for each.

Then give me:
1. Total invoiced value by job type, as a table.
2. The ten customers with the most overdue jobs.
3. Every job where the invoiced amount differs from the quoted amount by more than 20%.

Show the control total of the amount column before and after exclusions. Do not round until the final tables. Treat amounts as NZD excluding GST unless a column says otherwise.

Our lesson on analysing a spreadsheet with AI goes further with the same habit.

What changes after training, and how we measure it

Afterwards people should use projects and shared GPTs for repeat work, check numbers and sources before anything leaves the building, and stop using personal accounts for work. Business includes usage analytics, which show who is active. Activity is not value, so time two or three target tasks before training and again at 30, 60 and 90 days, as set out in our lesson on measuring adoption after training. Use the result to resize seats before renewal.

Where ChatGPT falls short

• Long inputs on fast models. OpenAI’s plan comparison lists a GPT Instant input maximum of about 40 pages of text on Business, against about 320 pages for the reasoning models. Long tender packs need the right model or splitting.
• Plausible analysis with wrong assumptions. It may treat GST-inclusive figures as exclusive or quietly drop rows. Control totals catch this.
• New Zealand rules. It can answer employment, tenancy or tax questions with confidence and Australian or US law. Give it the source.
• A moving product. GPT-6 replaced GPT-5.6 in ChatGPT on 7 October 2026. Rerun your saved test tasks after each change.
• Write access. Apps with write actions can change files, so keep them off until the team is trained. Our lesson on AI tool permissions explains the trade-off.

How this differs from OpenAI’s free training

OpenAI’s own OpenAI Academy is worth using. When we checked on 20 October 2026 it offered an Apply AI at Work track (prompting and output review, reusable workflows, handing tasks to agents), a Lead AI Adoption track for leaders, live events and a small business community, and the page listed no fees. We recommend it as pre-reading.

Academy courses cannot open your workspace. They will not set your retention, decide which apps may write to your Drive, or test prompts on last month’s export. Ours do, and then measure the result. If you are weighing the two, read workshops versus self-paced courses.

New Zealand notes: Privacy Act, data location and seat prices

• Privacy Act 2020. In a Business workspace with training off, OpenAI processes data on your behalf, so under section 11 you remain responsible, including security under IPP 5. On personal plans with training on, the Privacy Commissioner names use as AI training data as one way a provider uses information for its own purposes. That can make sharing a disclosure under IPP 11, and an offshore one under IPP 12. Our lesson on protecting customer information turns this into daily rules.
• Data location. No ChatGPT plan processes prompts in New Zealand. Storage at rest in Australia is available to Enterprise and Edu, not self-serve Business, and processing in-region is offered only for Europe, the US and the UAE.
• Prices and GST. As listed on OpenAI’s pricing page on 19 October 2026, a Standard seat is NZ$33 a month billed annually or NZ$41 monthly. The page does not say whether GST is included, so check your first invoice. A two-seat pilot billed monthly for three months costs 2 × NZ$41 × 3 = NZ$246.

Written by the Levy Street team in Wellington. Last updated 19 October 2026.