ChatGPT Ads South Africa: A Practical Small Business Guide

The following are hypothetical planning examples. Naming a business does not mean it advertises on ChatGPT, has approved these drafts or has achieved any stated outcome.

An Eastridge electrical business: qualify the job before the call

Consider an Eastridge business such as WWJDElectrical Services when thinking about local trades. Before developing ChatGPT Ads, confirm its current services, coverage, availability and approved claims directly. A sample customer situation is a homeowner planning electrical work and comparing what to ask a contractor.

A suitable landing page would explain the relevant service, information needed for a quote and actual areas served. An enquiry could ask for the suburb, a short description and preferred contact method. Avoid asking the visitor to upload unnecessary personal documents.

Measure suitable job enquiries, not every phone-button tap. If the service cannot cover a location or fulfil an urgent request, make that clear. Our Mitchell’s Plain business visibility guide and service-area page checks support that preparation.

A Bellville business: sell one clear service

Imagine a Bellville upholstery business offering dining-chair re-covering. ChatGPT Ads could lead to a page explaining the quote process, fabric choices, collection arrangements and expected next step. The example does not assume any particular business offers collection.

Compare two messages: refreshing worn dining chairs and coordinating chairs with a room update. Keep each message accurate. Both can lead to a relevant page, but record which version produces useful enquiries. A generic homepage listing every service makes that comparison harder.

A Cape Town tour operator: turn research into a workable itinerary

A hypothetical family is comparing December outings, travel time and activities for different ages. A tour operator considering ChatGPT Ads should explain route options, inclusions, accessibility information, booking conditions and how availability is confirmed.

Send the visitor to the relevant experience page. Ask for travel date, party size and essential requirements. Do not advertise availability that staff cannot confirm. The practical planning questions in our Cape Town December guide can inform useful customer research.

A local retailer: protect the margin after delivery

For a hypothetical Cape Town retailer, ChatGPT Ads must lead to accurate product information, stock and delivery terms. A low-value order may become unprofitable after fulfilment and advertising. Compare contribution per order before deciding that more orders would help.

These examples also suit a broader inclusion lens: women-owned, Khoi and San-owned and Cape Flats businesses need clear offers and workable economics, rather than assumptions about their customers or access to budget.

This tutorial follows public OpenAI documentation checked on 26 October 2026. It is not a recording of a live South African account walkthrough. Labels and available features may differ. Do not submit a campaign while still working out its budget or start date.

Step 1: define the first test on paper

For ChatGPT Ads, choose one offer and one customer need. Record the destination URL, enquiry definition, spend approval and review decision. For example: test enquiries for one planned service, then decide whether lead quality justifies continuing.

Name the campaign clearly, such as “Cape Town | Service quote | Initial test”. Do not mix unrelated products merely because the same business sells them. Keep a dated copy of the approved brief.

Step 2: create or join the correct advertiser account

Open the official OpenAI Ads Manager. If an account already exists, request an invitation instead of creating duplicates. Check the business details and account country, currency and time zone carefully. OpenAI says those three selections cannot be changed after creation.

Its account setup instructions describe an option to skip initial ad creation and enter Ads Manager. Use that route if you are still preparing. Review public brand and registered legal names separately.

Step 3: confirm account readiness and billing

Complete the requested verification and billing through the official interface. Give collaborators the appropriate role, rather than sharing a password. Keep ownership and billing responsibility clear.

With ChatGPT Ads, onboarding can create a campaign that becomes eligible to deliver after approval and other requirements are met. If you are not ready, confirm the campaign is paused. Account creation is not a harmless substitute for reviewing delivery settings.

Step 4: choose the objective and budget

Choose Views for impression-based awareness, Clicks for visits to your website or an eligible Conversions objective for a supported customer action. Record the chosen billing method: conversion optimisation can still charge for valid clicks or impressions, rather than completed sales. OpenAI’s campaign guide explains the available objectives; conversion optimisation does not mean you pay only for completed sales.

Review total exposure, start conditions and billing before submitting ChatGPT Ads. Write your approval in the campaign record. If a needed scheduling control is unavailable during onboarding, stop before publication rather than launch first and fix dates afterwards.

Step 5: check geography and customer fit

Inspect the supported locations in the account. A page describing Mitchell’s Plain does not itself restrict campaign delivery there. OpenAI’s targeting documentation distinguishes eligibility settings from descriptive context.

For ChatGPT Ads targeting Cape Town customers, check the actual available location controls and the offer’s service area together. If they do not fit, postpone the test. Do not spend in an unsuitable country simply to progress through onboarding.

Step 6: write useful context hints

Context hints explain the product, customer need and situation at ad-group level. For a hypothetical planned-maintenance offer, describe the work supported, actual service area and information needed for a quote.

Avoid a string of suburb names or commands telling ChatGPT to recommend you. ChatGPT Ads context hints are relevance guidance, not a placement reservation. Keep the statements consistent with the destination page.

Step 7: create distinct messages and a relevant destination

Draft variations that answer different needs. For example, one message might explain a quote process while another highlights a clearly defined service. Review any suggested text against what the business can deliver.

OpenAI’s creative guidance recommends relevant destinations and accurate, useful messages. For ChatGPT Ads, use the specific service or product page rather than automatically using the homepage. Confirm mobile readability, image relevance and accessible link labels.

Worked creative example: a planned service quote

For a hypothetical local contractor, a ChatGPT Ads draft might use the headline “Plan your electrical work with a clear quote”. Supporting copy could explain that customers can describe the job and service location to request the next available assessment. Publish it only if the contractor actually provides that process.

A corresponding context hint could describe homeowners comparing contractors for planned work, within the business’s verified service area. The destination should explain the assessment process and which work is included. Do not insert emergency availability, qualifications or a free inspection unless those claims are verified.

A second ChatGPT Ads variation could focus on understanding the scope before booking. Change the angle, not the underlying promise. Record why each message exists so the review can connect enquiry quality to customer need.

Step 8: test measurement, then review delivery

Before activating ChatGPT Ads, complete a labelled test enquiry. Confirm receipt, confirmation messaging and the correct reporting event. Review the destination with its tracking parameters. Obtain the owner’s approval of spend and settings.

After launch, check delivery and lead handling against the brief. Keep a change log. If enquiries are unsuitable, investigate the offer, geography and page before raising spend. This guide does not launch a campaign or authorise expenditure.

At the 26 October 2026 check, OpenAI’s campaign table lists an R83 minimum daily budget for ZAR-billed accounts. That is a platform minimum, not a recommended test budget or a South African cost-per-click benchmark. Verify the amount and billing currency in your account before approving spend. The figure comes from the minimum campaign budget table.

A daily budget is an average delivery target. OpenAI’s daily-budget guidance describes daily fluctuations and weekly controls. Use the account’s current rules to calculate exposure, particularly after budget changes. For a firm overall limit, examine the available campaign-total controls.

Separate ChatGPT Ads media spend from landing-page work, tracking, creative and management fees. No reliable South African CPC or lead-cost benchmark is established by this article. Avoid converting an overseas suggested bid into a local performance forecast.

Ask a supplier to state exactly what is included, who owns the account and how additional work is approved. Our proposal buying questions help you evaluate scope before committing.

Use gross profit rather than sales turnover when assessing ChatGPT Ads. The calculation below is hypothetical and excludes tax effects. It is a planning exercise, not financial advice or an expected campaign result.

1. Gross profit per completed job: R1,000.
2. Qualified enquiries becoming customers: 20%.
3. Expected gross profit per qualified enquiry: R1,000 × 20% = R200.
4. Illustrative allowance for media cost: R80 per qualified enquiry.
5. Amount remaining before other acquisition costs and profit: R120 per qualified enquiry.

R200 is not the amount this business should automatically bid for a lead. Staff time, management fees, unsuccessful quotes and other acquisition costs still matter. If the close rate falls to 10%, expected gross profit per qualified enquiry falls to R100.

For your ChatGPT Ads worksheet, replace every number with your own evidence. Record the offer, gross profit, close rate, target enquiry cost, total test allowance and review date. If these inputs are unknown, mark them as assumptions and use conservative limits.

Define “qualified” before the test: appropriate service, workable location, genuine requirement and a reachable prospect. A customer who needs something you do not sell is not a useful lead merely because the form was completed.

OpenAI’s conversion measurement documentation explains the platform’s measurement options. A practical business record should also connect the enquiry to its outcome. A form event alone cannot tell you whether the customer was suitable or bought.

For ChatGPT Ads, record campaign, enquiry date, service requested, qualification outcome and eventual sale. Keep personal information in the appropriate business system, not in URL parameters. Use consistent campaign tags that distinguish paid traffic from organic ChatGPT referrals.

Follow our AI referral measurement guide when defining reporting categories. If the same visitor appears in several reports, do not count those reports as separate customers. Check event duplication and attribution settings before announcing a result.

ChatGPT Ads reporting can lag. OpenAI also notes that review crawlers may visit tagged destinations without generating billable clicks. Investigate differences between platform clicks and analytics visits; do not assume either report directly proves fraud or missing customers.

Judge a small test as a small test. Low volume may be insufficient to choose a winning message. Report spend, period, observations and uncertainty together, including tests that did not produce suitable enquiries.

Before adding an advertising pixel or server-side events, review the information shared, your privacy notice and required consent with the responsible person. OpenAI’s measurement guidance requires appropriate disclosures and permissions. Hashing contact data does not remove the need to assess whether sharing it is permitted.