ChatGPT Ads Are Live in South Africa. That's Not the Real Story.

Every new advertising platform arrives with the same promise: get in early before everyone else does.

ChatGPT Ads landed in South Africa carrying exactly that expectation.

Within days, LinkedIn was full of agencies announcing themselves as ChatGPT Ads experts. Clients started asking whether they should be moving budget. For transparency, many marketers were worried about being late to the party.

The question is not whether ChatGPT Ads are worth testing. The bigger question is whether we are witnessing the start of a fundamental shift in how people discover, evaluate and choose brands. If consumers are moving from searching to asking, the implications extend far beyond a new media channel.

What changed?

On 21 October, ChatGPT Ads opened to South African advertisers.

Brands can now buy ads through OpenAI's self-serve Ads Manager. Ads appear separately from ChatGPT's responses, are clearly labelled, and currently target users on the Free and Go plans.

The entry cost is relatively low. The minimum spend is $25 per day, roughly R410, which puts a month's testing budget at around R12,400 before bank fees.

For most brands, that's an affordable experiment. Which is exactly why it deserves a disciplined approach rather than a rushed one.

What the early results tell us

As with most emerging platforms, the results depend heavily on who's reporting them.

Independent studies have produced mixed outcomes. One retail advertiser recorded a click-through rate of 0.91%, compared with 6.4% on Google Search within the same category. Another side-by-side test reported lower click costs on ChatGPT, $3.14 versus $6.76 on Google, but generated no qualified leads, while Google produced two.

OpenAI's published partner case studies paint a more optimistic picture.

WeightWatchers reported a cost per acquisition 15.3% lower than paid search, while supplement brand Dose reported that 67% of incremental purchases came from new customers.

That last metric is the one that stands out.

If ChatGPT's value lies in influencing people during consideration, especially consumers who haven't previously engaged with your brand, then traditional performance metrics may not tell the whole story.

Which also makes measurement more important than ever.

Earlier this year I saw a dashboard showing conversions up more than 100 times. It looked like a breakthrough until we realised the majority were app logins rather than genuine acquisition events.

A new platform with no historical benchmarks is exactly where misleading success metrics thrive.

Why South Africa may be different

Most published testing has taken place in North America and Europe.

South Africa presents a different picture. Consumer adoption of AI tools is growing rapidly, but search behaviour remains deeply entrenched. Mobile dominates digital usage, data costs still influence online behaviour, and many brands are only beginning to understand how AI assistants fit into the customer journey.

That matters because ChatGPT Ads shouldn't simply be evaluated against media benchmarks.

They should be evaluated against a broader question: What role will AI assistants play in product discovery, comparison and decision-making over the next three to five years?

The answer may vary by category, audience and market maturity. But it is becoming increasingly difficult to ignore.

The challenge for regulated industries

This was the first thing I checked for a client in financial services.

OpenAI currently restricts financial services advertising to approved advertisers and reviews applications on a case-by-case basis. Several analyses of the platform's policies suggest access remains limited outside the United States.

For banks, insurers and lenders, the first question is not whether the platform works. It is whether you are permitted to advertise there in the first place.

How I would approach a test

1. Confirm eligibility

Make sure your category, business and market are eligible, and that South African targeting is available.

2. Get measurement right first

Implement the pixel, configure server-side tracking where possible, and agree on a clear definition of conversion before spending a cent.

A login is not a lead.

A lead is not a customer.

All three should be measured differently.

3. Ringfence a learning budget

Spend an amount you're comfortable losing in exchange for insight.

Treat the first phase as a research exercise, not a growth channel.

4. Benchmark against your strongest search activity

Compare performance against real outcomes such as qualified leads, sales or customer acquisition cost.

Don't judge success purely on clicks.

5. Define success before launch

Agree on the thresholds that determine whether you scale, optimise or stop.

Most failed tests don't fail because of results. They fail because nobody agreed on success upfront.

6. Separate new and existing customers

If the early evidence proves true, incremental customer acquisition could become one of the platform's strongest advantages.

Track it independently.

One personal view: I would be cautious about launching a first test during peak retail periods such as Black Friday. Noise makes learning difficult. A cleaner testing environment often produces better decisions.

So, should brands be investing?

My advice to clients is not to be first, it is to be prepared.

There is no prize for arriving first in a channel you cannot measure.

There is value in understanding where consumer behaviour is heading before your competitors do.

Whether ChatGPT Ads become a major media platform remains uncertain.

What is becoming increasingly clear, however, is that consumers are growing more comfortable asking AI platforms for recommendations, comparisons and advice that they once sought through traditional search.

That is the shift worth paying attention to. The real opportunity may not be buying ads inside AI ecosystems. It may be ensuring your brand is visible, trusted and discoverable when those AI-driven conversations happen.

That is the test I would be preparing for now.