Net Age Is Testing ChatGPT Ads in South Africa: What We Know and What We're Learning

ChatGPT advertising has arrived in South Africa, and Net Age is starting the testing process with a select group of clients.

Getting involved early is part of how we work. Whenever a new paid advertising opportunity opens up, we want to understand what it can do for our clients. We look for ways to increase value, reach potential customers and improve results, and we’re prepared to do the practical work needed to find out whether an opportunity delivers.

When ChatGPT ads became available locally, we were ready to start that process.

Our initial testing covers four sectors: skincare, retail, property and home furniture. These give us different buying journeys to explore, from everyday product purchases to considered furniture decisions and property enquiries.

Other clients are welcome to join the testing process at the same time. We’ll work with each business to agree what success should look like, put campaigns in place, and measure the results.

The aim is straightforward: understand how the platform delivers and whether the visitors it brings turn into meaningful business.

We’re interested in the opportunity, but it is early days. We don’t have paid advertising results that justify promises about costs, sales or return on investment.

From the Launch Announcement to Practical Testing

OpenAI announced the expansion of ChatGPT ads to South Africa on 21 October 2026. Its official expansion announcement includes South Africa among the new markets. Availability does not mean every account has identical access or every campaign can immediately deliver.

Our earlier ChatGPT advertising launch article introduced the opportunity. This article moves the discussion towards the work ahead: checking access, building suitable campaigns, preparing websites and deciding which evidence will be useful. It is a testing update, not a report of completed campaign results.

Early adoption is valuable when it creates knowledge a business can use. It is less valuable when it produces a confident headline before the evidence exists. We are taking a hands-on paid-media approach, learning how the available controls work and checking what happens after a visitor arrives. We are not claiming to be the first agency involved.

Organic ChatGPT Referrals Give Us a Reason to Investigate

Clients have reported growth in the small amounts of organic ChatGPT referral traffic seen across sites we monitor. That is an observation from those businesses, not a claim about every South African website or the whole market. We are not attaching a growth percentage or a sales claim to it.

These visitors are separate from paid advertising. An organic referral can follow a link included in an AI answer or research response. A paid visit follows an advertisement. Growth in one does not establish the likely cost, quality or volume of the other.

The organic activity gives us a practical reason to pay attention: some people are already using ChatGPT while finding information and businesses. It does not prove that advertising will be profitable. Our discussion of SEO and AI search optimisation explains the separate work involved in organic visibility. Neither that work nor an advertising payment guarantees a recommendation in an answer.

A Different Research Journey, With Intent Still to Prove

A person can use a conversation to explain a need, ask follow-up questions and compare options. Rather than enter one short search, they might discuss size, availability, delivery or the trade-offs between products. A relevant advertisement could introduce an option during that process.

That creates a reasonable hypothesis: conversational research may help connect an offer with a useful moment in a buying journey. It is not proof of higher purchase intent. Someone comparing options could be buying today, planning for next year, helping a friend or simply learning.

We need to test what happens on the website and afterwards. Do visitors find the relevant product? Do they submit a useful enquiry? Do they return later? A detailed conversation sounds promising, but the business benefit has to be measured rather than inferred from how the platform works.

Advertising Does Not Buy the AI Answer

ChatGPT ads are clearly labelled and separate from AI answers. OpenAI says advertisers cannot shape, rank or alter ChatGPT’s responses through advertising. A sponsored placement is not an endorsement of the advertiser or its products.

For a business, this means an ad budget buys participation in the advertising system, not a place in an organic recommendation. We cannot purchase an answer that names a client as the best supplier. We also cannot promise that an ad will appear alongside a particular response.

This boundary should shape both the sales discussion and the creative. An advertisement should explain the actual offer, not imply that the assistant has independently chosen the business. Readers can check the distinction in OpenAI’s ChatGPT ads questions and answers, which also explain privacy and user controls.

Who Can See Ads, and Why That Matters

Ads may appear for eligible users on the Free and Go plans. Plus, Pro, Business, Enterprise and Edu accounts are ad-free. OpenAI also says it does not show ads to accounts identified as belonging to people under 18. Eligibility and availability remain important parts of the audience picture.

We should not turn those plan differences into assumptions about income, buying power or lead quality. A free-plan user may be a valuable customer, while a paid-plan user may never be shopping for the product. The useful evidence is what the reachable audience actually does.

It is also important not to confuse ChatGPT’s overall audience with advertising reach. A business cannot assume that every person using the service is available to its campaign. Test plans need to work with the eligible audience and the delivery options available to the account.

Relevance, Personalisation and Context Hints

The current conversation is one signal used in ad selection. The system also considers the ad, its landing page and other relevance signals. When a user has ads personalisation enabled, select signals from their broader ChatGPT experience may also be considered. Advertisers do not receive users’ private conversations.

Context hints let an advertiser describe genuine product features, customer needs or situations where an offer may be useful. For example, a furniture business could describe compact dining tables suited to smaller rooms, provided that accurately describes its products. Clear information is more useful than a disconnected list of popular words.

Hints are not exact-match keywords, targeting rules or guaranteed placements. They cannot enforce geographic restrictions or ensure delivery in a specific conversation. OpenAI’s context hints guidance makes this explicit. We will treat hints as relevance information, then assess delivery rather than assume it.

Geography and Account Access Need Checking First

Supported geographic controls should be used for delivery restrictions, not sentences placed in context hints. The locations available can vary by country and campaign type. Current guidance distinguishes country-only controls for new product-feed campaigns from supported regional or city options for other campaigns, where available.

Some new self-serve accounts can initially advertise only in their home country. Selecting another country does not override that restriction. OpenAI says wider eligibility depends on approved identity verification and the required advertising spend from campaigns run in the home country.

A South African account therefore cannot be promised delivery in the United States simply because the campaign screen offers a country selection. We will check account eligibility, supported locations and actual campaign settings before agreeing the test’s reach. The campaign creation guide explains these limits. A saved draft or approved ad is not proof that delivery is possible.

Skincare: Test the Shopping Experience, Not Health Conversations

Skincare gives us a product journey where clear information matters. A hypothetical test might focus on a cosmetic product range, pack size, texture or everyday routine preferences. These are examples of questions to explore, not evidence that a particular campaign or placement has already worked.

We will not build a plan around advertising in personal health conversations. OpenAI says ads are not eligible near sensitive topics, including personal health and mental health. Advertiser eligibility for a regulated category does not remove that sensitive-context restriction. Product claims and policy suitability need checking before anything runs.

The landing page should accurately describe the product, ingredients, use instructions and relevant limitations without unsupported treatment claims. A visitor should be able to see price, availability and delivery terms. The test asks whether an eligible, relevant advertising opportunity brings customers who can make an informed purchase, not whether private health discussions can be targeted.

Retail: Connect a Specific Need With a Useful Product Page

Retail allows us to explore more immediate shopping decisions. As a hypothetical example, someone researching a practical household item might compare dimensions, materials and available options. An ad could describe a genuine benefit and lead to a relevant product or collection rather than a broad homepage.

We would start with a manageable product range and reliable stock information. If the ad suggests an item is available, the destination should not reveal that it is out of stock. Delivery charges, payment options and returns information should be clear before the customer reaches the final checkout step.

Sales count alone is not enough. A retailer needs to consider margin, fulfilment costs, cancellations and returns. A campaign that produces orders at a cost the business cannot sustain is not a success. We will look for commercial value, not simply a busy checkout report.

Property: Qualified Enquiries Matter More Than Form Totals

Property often involves a longer, more considered journey. A hypothetical use case could involve someone comparing locations or looking for a home with particular practical features. The ad and destination should reflect real listings or a clearly described service, with accurate availability and area information.

We will not assume every person who reads about a suburb is ready to buy. A useful enquiry definition might include the relevant area, suitable property requirements and a clear request for assistance. The business must decide what makes a lead actionable before the campaign starts.

The enquiry experience also affects results. Mobile forms should be easy to complete, contact options should work, and the team should respond while the need is still current. Reporting should distinguish an initial form submission, a qualified enquiry and a later outcome. Those stages are related, but they are not interchangeable.

Home Furniture: Support Comparison and Considered Decisions

Home furniture gives us another longer journey, often involving room size, style, materials and delivery. A hypothetical test could explore an accurate offer for a dining set or a sofa suited to a smaller space. We would not claim that these examples are existing campaign findings.

The destination needs useful dimensions, clear photographs, material details and honest delivery information. If people need to visit a showroom or ask about an option, that next step should be obvious. An attractive ad cannot resolve uncertainty that the website leaves unanswered.

A furniture customer may compare products, discuss the purchase at home and return days later. An initial visit without an immediate order is not automatically worthless. Equally, a long visit is not proof of a future sale. We need to connect observed behaviour with later enquiries or purchases where measurement allows, while keeping the uncertainty visible.

The Website Is Part of the Advertising Test

Across all four sectors, the ad should lead to a page that continues the same promise. If the message is about a particular product, service area or useful feature, visitors should not have to search the whole website to find it. OpenAI’s ad creation guidance recommends clear, benefit-focused copy and a relevant, reachable destination.

We will check the experience on phones as well as larger screens. Product information should be readable, important details should not be hidden, and checkout or enquiry steps should work. Tracking should be checked alongside consent choices rather than added without considering privacy.

Testing may reveal a website problem rather than an advertising problem. If visitors cannot understand delivery costs or a form fails, we should fix that before drawing a conclusion about the channel. A better landing page benefits the wider marketing programme too, even if the new advertising test does not continue.

For the separate task of organic discovery, our guide to making your website easier for AI to find and understand covers access, clear content and technical readiness. That supports a readable, useful website; it does not buy an AI recommendation or guarantee advertising results.

Costs Are Still Unknown for These Tests

We do not yet have reliable South African results from our testing that justify a cost forecast. There is no universal rand trial budget that makes sense for every business. The right test depends on the offer, available margin, sales journey, measurement readiness and what the business can responsibly risk.

ChatGPT ads supports different buying and optimisation approaches. Views uses CPM, meaning payment per thousand impressions. Clicks uses CPC, meaning payment per valid click. Conversion optimisation aims towards a supported event but is billed through valid clicks or impressions, depending on the eligible billing option. It is not pay-per-sale.

That distinction matters when discussing risk: spend can accrue without sales. We will agree limits and review conditions with each business, not present a platform objective as a guarantee. OpenAI’s advertiser basics explains pricing and reporting. Guidance about bids is not a forecast of what a South African client will pay or earn.

Start With a Focused Question and a Success Definition

A useful early test asks a question narrow enough to answer. Can a selected product range attract relevant visitors? Can a service generate enquiries that the sales team considers worthwhile? Trying every offer, audience and creative angle at once can spread the evidence too thinly.

Before launching, we will agree the desired outcome and how to judge it. For an online store, that means considering order value and contribution margin, not revenue alone. For a lead-based business, it means defining a qualified lead and checking what happens after the form is submitted.

We also need practical review rules. What would justify continuing, changing the destination or pausing? What measurement failure would make the findings unreliable? The answers should fit the business, not a standard promise. A test can be useful even when its conclusion is that the channel is not ready for that offer.

Keep Paid and Organic Tracking Separate

We will use clear campaign links with UTM parameters where appropriate, so paid ChatGPT visits can be distinguished from organic referrals in website analytics. Naming should be consistent across campaigns and ads. A single combined “ChatGPT traffic” total would hide the difference the test is meant to investigate.

We will check that those parameters survive the journey and that relevant purchase or enquiry events are recorded correctly. Test events should not be mistaken for customer outcomes. OpenAI also warns that ad-review crawlers can visit tagged destinations: an analytics entry does not automatically represent a billable human ad click.

Platform reporting, website analytics and sales records may not agree perfectly. Consent settings, missing tags and different definitions can all affect counts. We will document those gaps rather than force matching numbers. The purpose is to understand the evidence well enough to make a decision.

Allow for Delayed Conversions and Uncertain Attribution

Some people act after their first visit. They may return directly, search the brand on Google or speak to someone before buying. A later sale can be influenced by several channels, and the final recorded click may not tell the whole story.

That does not mean every later sale should be credited to ChatGPT. We will use the available attribution evidence carefully, with agreed observation periods that reflect the buying journey. Property enquiries and furniture purchases may need a different review rhythm from everyday retail orders.

Where possible, sales-team feedback can help assess lead quality without inventing a precise path that tracking cannot show. We should be clear about observed conversions, reported attribution and unanswered questions. A channel report is a decision aid, not a complete record of what happened in someone’s mind.

Useful Reporting Is Not the Same as an Early Success Claim

Initial reporting can tell us whether ads are delivering, which messages receive clicks and whether destinations work. Impressions, click-through rate and spend are useful diagnostics. They are not enough to announce that a campaign has increased profit or outperformed another channel.

Small samples are especially easy to overread. One sale can change an early cost-per-sale figure sharply; one unsuitable lead does not prove the audience is wrong. We need enough relevant activity, time and dependable tracking to distinguish a pattern from chance.

Reports should show the period covered, spend, measured outcomes and important limitations. OpenAI notes that some reporting is delayed, including spend, so a displayed zero is not necessarily proof of no charges. We will avoid turning partial data into a fixed benchmark or using the best early day as a forecast.

Build Knowledge Alongside Google and Meta

ChatGPT ads should be considered alongside established channels, not as a reason to abandon them without evidence. Google and Meta may serve different moments in the customer journey. An early ChatGPT test does not establish that those channels are becoming unnecessary.

We will compare business outcomes where the offers, periods and measurement are reasonably comparable, while recognising that attribution and audience overlap complicate the picture. Protecting effective existing activity can be more sensible than moving a large budget into a channel whose performance is still unknown.

The aim is to find whether ChatGPT adds useful value to the paid advertising mix. That might mean additional qualified enquiries or sustainable sales, but those are outcomes to test, not promises. If the evidence supports expansion, we can discuss it. If it does not, the test should still leave the business better informed.

Join the Testing Process Now

We will share findings when there is enough meaningful data to explain what we have learned responsibly. That includes useful limitations and things that did not work, not only positive headlines. We will not invent case results, reveal client identities or generalise from a handful of outcomes.

Businesses do not need to wait for a finished case study to take part. If you want to explore ChatGPT advertising now, contact Net Age to join the testing process. We can discuss your offer, account access, website readiness and the commercial result that would make a test worthwhile.

For now, our position is simple: the opportunity is available, the practical work is starting, and the results still need to be learned. We are ready to do that work with clients who want a measured approach to a new paid advertising channel.