The answer is still free. The space under it is not

Short answer. OpenAI has opened self-serve advertising in ChatGPT to the Middle East. The ads sit under the answer, carry a sponsored label and, the company says, do not affect the answer itself. You still have to earn a place in the answer. What you can buy is the space beside it, and that decision comes after you count how many people reach you from those answers today.

The answer stayed free. The space under it did not.

What opened, and where?

Self-serve advertising in ChatGPT. Since 31 August it has been available in the Middle East too.

The timeline is short. In January OpenAI said it planned to test ads. On 9 February the test started in the US, for logged-in adults on the Free and Go tiers only. On 5 May a beta Ads Manager arrived along with cost-per-click bidding, whereas the first phase had sold impressions only. In August ads reached the UK, Mexico, Brazil, Japan and South Korea, then 31 European markets. On 31 August self-serve opened across India, Europe, the Middle East and North Africa.

Ads now run in more than 40 countries. In under 200 days they reached a billion dollars in annualised revenue run rate. That is a run rate rather than a billion earned, and the difference matters when you read the headlines.

Who sees ads: people on the Free and Go tiers. Who does not: people on Plus, Pro, Business, Enterprise and Education. Accounts identified as under 18 see none. On the free tier you can turn ads off in exchange for fewer messages a day.

What changed inside the answer?

Nothing. The ad sits under the answer rather than inside it.

OpenAI's help centre says ads are labelled as sponsored and visually separated from the response. The company states directly that ads do not influence what ChatGPT tells you, and that answers are optimised for what is most helpful.

That is a statement from the platform, and it does not become a verified fact by being published. Disputing it in advance would rest on nothing: no independent measurement exists, ours included. The sensible move is the one you would make with any platform promise. Write the wording down and check it later against what you see on screen.

One restriction is worth knowing before you plan anything. Ads are not eligible to run near sensitive and regulated subjects, including health, mental health and politics.

The health row matters more than it seems. For a whole class of businesses, this update simply does not apply. Clinics, therapists, family lawyers and everyone working in a regulated field stay exactly where they were, relying on search, on the answer itself and on recommendations, never on a paid slot.

Is it worth buying?

Count how many people already arrive from AI answers first. Without that number you are buying blind.

The traffic is visible in your analytics, but you have to go looking. Referrals from ChatGPT come from chatgpt.com, from Perplexity they come from perplexity.ai. In a source report they sit as separate rows and usually get lost among everything else. Look at the last three months: how many sessions, and how many of those became enquiries.

The arithmetic after that is simple. If ten people a month reach you from AI answers, a paid slot adds to those ten. It does not create a channel out of nothing. If the number is zero, the useful work is finding out why. Usually one of two things is true: the system cannot find you, or it finds you and does not treat your site as a clear enough answer. The first is a technical fix, the second is a content fix, and both cost less than advertising.

On price, the honest answer is that nobody has performance data. The channel is seven months old globally and two weeks old in the Middle East. Impressions and clicks are sold, but what a click costs in your niche and what it turns into are unknown so far. That is no reason to skip a trial. It is a reason to keep the trial out of your sales plan.

What to do before you spend a dollar

Check whether an AI can cite you at all. The check takes an evening and needs no tools.

1. Ask about your service the way a customer would. Not by company name but by the job: "who builds AI agents in Dubai". Compare the answers from ChatGPT, Perplexity and Gemini.
2. Look at who got named instead of you. Open those pages and read them. They usually carry what yours does not: specific numbers, stated terms, answers written in the customer's words.
3. Check that your site can be read at all. If key pages assemble themselves in the browser, some systems will never see them.
4. Find where other people write about you. A model leans on what is confirmed outside your own site.
5. Measure what arrives from answers today. One number, written down before any spending. A month after launching ads you will compare your numbers against it, and that comparison is the only honest way to tell whether the money worked.

Markup, page structure and plain answers to common questions are quick to fix. Mentions on other people's sites and domain reputation take months. We do the first as part of search visibility, and the material for the second comes out of the content factory.

One honest closing thought. A paid slot under the answer does not replace a missing answer. If the system never names you, advertising puts you beside somebody else's answer: that works, it costs more, and it ends the day the budget runs out. An earned mention lasts longer.