Article · paid media
Ads in ChatGPT:
is it worth investing now?
ChatGPT knows why your customer clicked. Do you?
Get the free checklistOn 1 September, a marketing executive in Portugal launched one of the first campaigns on ChatGPT Ads. She got 29 clicks, no sales and no answer to the most important question: why did those people click? The platform knew. She did not.
Since August, ChatGPT ads have also been running in Brazil, one of the markets I work in. The question I hear most often is: is it worth getting in now?
My short answer: it is worth testing, not scaling. And it is not worth starting without a plan. In this article, I show what the first tests reveal, where the platform's blind spot is and how I assess a new channel before putting money into it.
Prefer to go straight to the free checklist? Click here.
What changed: from keyword to intent.
Since 4 August 2026, ChatGPT has shown ads in Brazil to adult users on the Free and Go plans. Subscribers to the higher paid plans do not see ads.
The big news is not the ad space itself, but the moment the ad appears:
- On Google, the ad responds to the word the person typed.
- On Instagram and Facebook, it tries to interrupt the scroll based on profile and behaviour.
- On ChatGPT, it appears in the middle of a conversation, while the person is explaining a problem, comparing options or about to decide.
To choose the ad, the platform combines the context of the question being asked at that moment, the parameters set by the advertiser and the user's conversation history. In practice, the ad does not appear because someone typed “personalised chocolates”. It appears because someone wrote: “I'm organising the company's end-of-year celebration and need a gift for 40 employees, up to €20 each”.
An important detail if you work in health, wellbeing or beauty: OpenAI states that ads do not appear near sensitive or regulated topics such as health, mental health and politics. Depending on how your business presents itself, your campaign may reach far fewer people than expected. Check this before you build your plan.
What the first tests show.
The first results vary so much that there is no “market average” for ChatGPT Ads yet. Compare two accounts published in September:
| Lídia Amorim's test | Agency reports (Gustavo Abreu) | |
|---|---|---|
| Click-through rate (CTR) | 3.07% | 0.6% (traditional search: around 2%) |
| Cost per click | €0.77 | US$7 |
| Sales | None (945 impressions, 29 clicks) | No conversions |
| What went wrong | A billing error paused the campaign within a few hours; support replied days later | 57 charged clicks produced fewer than 20 visits in Google Analytics |
Links to the full accounts are at the end of this article.
The two cases do not contradict each other. They show a platform in beta, tested with small samples, in different sectors. Concluding that it “works” or “doesn't work” from this would be a misreading. Lídia herself acknowledges that her sample is too small to judge the channel.
What the two accounts have in common is more revealing than the numbers: in both, the advertiser finished the test unable to explain the result.
The blind spot: the platform knows, the advertiser does not.
The biggest problem with ChatGPT Ads today is not the ad. It is measurement.
The platform knows what kind of conversation the ad appeared in, which need triggered it and why that person was likely to be interested. The advertiser receives impressions, clicks and cost. Protecting the privacy of conversations is essential, and nobody wants to read each user's question. But there is a lot of room between exposing a conversation and offering aggregated data on the contexts that generate the best results.
Add to that the mismatch in numbers. When the platform charges for 57 clicks and Analytics records fewer than 20 visits, someone is counting wrong. And the advertiser pays the difference.
That is why my rule is simple: on a new channel, your own tracking decides, not the platform's report. Without a pixel installed and without links tagged with UTM parameters, you only know how many people saw and clicked. You do not know who bought, who gave up, or how much each visit really cost.
Pressing “Create” is easy. Knowing what you are testing is strategy.
Launching a campaign on ChatGPT Ads takes minutes. That is exactly the risk: the ease of the button makes it feel as if the strategy is already in place.
In 25 years in marketing, I have learned that the campaign is rarely the problem. The problem is almost always before it, in a poorly defined goal, or after it, when nobody reads the right numbers. That is why my work does not start in the ads manager. It starts with three questions:
- What is this channel's role in your plan? ChatGPT Ads does not replace Google or Meta. The question is where it fits in your customer's journey, and whether your customers are actually talking to AI before they buy.
- What is the goal of this test? At this stage, the honest goal is to learn, not to sell. That changes everything: the budget, the timeline and, above all, what counts as success. A test that produces a clear lesson is a good test, even without a single sale.
- Which number will decide the next step? A high CTR does not pay the bills. Before investing, I define which metric means scale, adjust or pause, and I check that it can actually be measured, outside the platform's dashboard.
A good campaign manager gets the campaign running. A good strategy makes sure every euro invested answers a business question.
The click is only the beginning: look at the whole journey.
In Brazil, a large share of sales that start with an ad end in a WhatsApp conversation. And that conversation has become more expensive: according to Gustavo Abreu's newsletter, since 1 October Meta charges for reply messages sent through the official WhatsApp API.
In practice, a lead from a paid ad can cost you twice: at the click and in the conversation. This is the journey that needs to be measured:
- The person sees the ad in a conversation with ChatGPT.
- They click and land on your page.
- They start a WhatsApp conversation or fill in a form.
- They are helped and enter your CRM.
- They buy, or they do not.
If you only measure the first step, you are optimising the cheapest half of the funnel. The right question is not “how much did the click cost?”, but “how much did each customer cost, adding up ads, follow-up and your team's time?”.
This is also why it pays to build channels you own, such as a qualified email list. Platforms change their rules. A list of people who asked to hear from you stays yours.
Before you test, use my checklist
ChatGPT Ads Checklist:
test with strategy, not luck.
I have gathered the 34 checks I run before, during and after testing a new channel like ChatGPT Ads into a free resource. It covers the plan, tracking, ad contexts, reading the metrics and the criteria for deciding whether to continue. It also includes a metrics table to fill in during the test and a decision matrix (scale, adjust or pause).
Fill in the form to get the checklist.
And if you would like an outside view of your media plan, your goals and the numbers you track today, that is exactly what I do. When strategy comes before the button, every test becomes a lesson, and every lesson becomes a result.
Sources (in Portuguese, except OpenAI): Lídia Amorim's test · Gustavo Abreu's newsletter · launch in Brazil · ad rules (OpenAI)