ChatGPT for NZ Small Business: A Practical Starting Guide

Start with one low-risk task whose output you can judge. A good first trial is drafting a routine message from approved facts. The goal is a useful finished result after review, not the fastest first draft.

A first experiment: an appointment reminder #

This example uses fictional information. It shows a bounded drafting task; it is not a claim about a client outcome. The person using it remains responsible for checking and sending the message.

Use a prompt that makes uncertainty visible #

Write a friendly appointment reminder of no more than 90 words using only these facts: a consultation is booked for Thursday, 22 October 2026 at 10 am by video call. Ask the customer to reply if the time no longer suits. The meeting link and staff name are not supplied: use [meeting link] and [staff name] placeholders. Do not add a fee, guarantee or policy. Return the draft, then a list of details a person must check.

An acceptable draft might say: “Just a reminder that your consultation is booked for Thursday, 22 October at 10 am by video call. Join here: [meeting link]. If the time no longer suits, please reply so we can arrange another time. Thanks, [staff name].”

Review the date, time zone, link, recipient and wording against the booking record. Resolve placeholders before sending. If the assistant adds a cancellation rule, remove it and record that the prompt needs improvement. A confident sentence is not evidence that a business policy exists.

Keep information and permissions under control #

Agree which tools and accounts the business approves, what information can be entered, who reviews the output and how records are retained. Begin with synthetic or already-public material. Do not use client records, credentials or confidential documents in an informal experiment.

OpenAI explains the available ChatGPT data controls. Check the current plan and terms for your account; turning off training is not the same as establishing that every proposed business use is appropriate. Model names, prices and available features change.

Measure the complete task #

On these invented inputs the saving is three minutes, not the time the model took to generate text. Repeat the trial on several representative tasks, record errors and include unsuccessful attempts. Continue if quality meets the same standard and the overall effort or outcome improves. Stop or change the task if review costs outweigh the benefit.

Download the AI trial and review log (CSV). Open in a spreadsheet or text editor; the example rows are illustrative.

Know when a different fix is better #

If the real work is copying a booking between systems, a template or ordinary integration may help more than AI. If the input is unclear, first improve the process. A useful Tech Clarity session can place a small experiment within the wider system; it should not require adding AI to every workflow.