Everything brands need to know before running ChatGPT ads

ChatGPT Ads turned the world's best-known assistant into a media channel in under a year. If you are deciding whether it deserves a line in your budget, the facts that matter are who sees the ads, how you buy them, and which categories OpenAI will actually approve.

This guide sets out the platform as it stands, with every figure attributed to its source. If you would rather work from a ready-made playbook than assemble one, our AI Ads Blueprint walks through campaign structure for answer-engine advertising end to end.

What follows is the structural picture: placement, audience, buying routes, bidding, measurement and policy. Those are the parts that decide whether ChatGPT advertising can work for your business at all, before any question of budget.

Key Takeaways

• Ads appear only to Free and Go users; Plus, Pro and Enterprise stay ad-free, per PPC Land.
• ChatGPT Ads now runs in more than 60 countries and regions, with 63 markets listed in OpenAI's help centre.
• There are three buying routes: the self-serve Ads Manager, OpenAI's Ads Solutions team, and named agency partners.
• CPC and CPM run in parallel; OpenAI recommends starting bids of $3–$5 per click, according to Search Engine Journal.
• Conversions API and pixel measurement track purchases, leads and sign-ups, per Marketing Dive.
• OpenAI's help centre says advertisers never receive chats, history, memories, name, email, precise location or IP address.
• Approval is per campaign setup, not per industry — and the rules changed four times in five months.

What ChatGPT Ads are and where they appear

ChatGPT Ads are paid placements that run inside the ChatGPT interface. They are shown as separate, clearly labelled units rather than woven into the assistant's answer, according to PPC Land's report on the platform's expansion into Asia. The distinction matters commercially as much as ethically: you are buying a placement beside an answer, not a mention inside one.

OpenAI has been explicit that keeping those two things apart is a design principle. David Dugan, Head of Global Solutions at OpenAI, said: "We're creating a new ads model – one that supports businesses and broader access to AI while staying grounded in clear principles around answer independence, privacy, and user control," as quoted by Search Engine Journal.

The commercial trajectory is steep. According to Marketing Dive, the ads pilot began in February 2026 and OpenAI expects to generate $2.5 billion in ad revenue in 2026, having reportedly eyed $100 billion by 2030. Early pilot participants named by Marketing Dive included Target, Albertsons and Williams-Sonoma — large retail advertisers rather than small local businesses.

For anyone already working on being cited by assistants rather than advertised beside them, the two efforts are complementary rather than competing.

Who actually sees ChatGPT ads

This is the single most important eligibility fact, and it is easy to miss. Ads in ChatGPT are shown only to users on the Free and Go plans. PPC Land reports that Plus, Pro and Enterprise subscriptions remain ad-free.

The practical consequence is a reach ceiling shaped by willingness to pay. The audience you can buy is, by definition, the audience that has not subscribed to a higher tier. For consumer brands with broad appeal that is a large and perfectly good market. For products sold to senior technical buyers or enterprises — the people most likely to be on a paid plan already — the addressable audience inside ChatGPT is structurally narrower than the platform's headline user numbers suggest.

Geography is the second filter. ChatGPT Ads is available in more than 60 countries and regions, with PPC Land noting 63 markets listed in OpenAI's help centre and Search Engine Roundtable reporting 60 countries and regions. The most recent step was on 23 September 2026, when ChatGPT Ads expanded to Indonesia, Malaysia, the Philippines, Singapore, Thailand, Vietnam and Taiwan, with all buying routes open on day one, per PPC Land. Earlier expansion covered the UK, Mexico, Japan, Brazil and South Korea, according to Marketing Dive.

The three ways to buy ChatGPT advertising

There is no single front door. PPC Land reports three distinct routes to buying ChatGPT Ads, and which one you use largely determines how much control and how much hand-holding you get.

The first is self-service through Ads Manager. OpenAI launched the beta self-serve Ads Manager on 5 May 2026, according to Search Engine Journal, which also reports that verified US advertisers can register at ads.openai.com, add payment information, set budgets, upload ads and manage campaigns directly. Marketing Dive describes the same portal as letting advertisers set budget, bid and pacing preferences, upload and launch campaigns, and access performance data in a dedicated portal.

The second is OpenAI's own Ads Solutions team, the managed route for advertisers who want the platform to help build the campaign. The third is agency partners, which PPC Land names as dentsu, Havas Media, Omnicom Media, Publicis Groupe and WPP. The presence of the five big holding companies tells you who OpenAI expects the large budgets to arrive through.

One detail from the September expansion is a useful signal about how OpenAI now launches markets: PPC Land notes that unlike India's earlier launch, which opened without self-serve access, the seven new Asian markets opened all three buying routes simultaneously. If you are planning for a market that has not launched yet, expect self-serve from the start rather than a managed-only phase.

How bidding and budgets work: CPC and CPM

ChatGPT Ads supports two bidding models side by side. CPM came first, introduced during the initial pilot so brands could gauge demand and early performance, according to Marketing Dive. Cost-per-click arrived with the Ads Manager beta on 5 May 2026, per Search Engine Journal, and aligns spend with the clicks that follow an impression rather than the impression itself.

On price, there is one published guide figure and it is a recommendation, not a rate card: OpenAI recommends starting bids of $3–$5 per click, as reported by Search Engine Journal, which adds that CPM is still supported in parallel. Treat that range as a planning anchor for a first test, not as the cost you will pay — a bid sets what you are willing to pay, and competition decides the rest.

Because CPC and CPM answer different questions, the choice should follow your objective rather than your habit. CPC suits direct-response campaigns where a click has a traceable value. CPM suits reach and awareness work where the goal is to be seen by a defined audience. Daily minimums also apply and vary by market, and we do not estimate them here — check the current minimum for your own market in Ads Manager before you set a budget.

Measuring ChatGPT advertising: Conversions API and pixel

Measurement was the pilot's weakest point and is now its fastest-moving area. Marketing Dive reports that OpenAI launched a Conversions API and pixel-based measurement tools that let advertisers track post-engagement actions including purchases, leads and sign-ups. That is the standard performance-marketing toolkit, and its arrival is what makes CPC bidding meaningful rather than decorative.

Reporting has since deepened. As of 30 September 2026, Search Engine Roundtable reports that the Insights API can retrieve conversion events beyond a campaign's primary goal, so several metrics can be tracked at once instead of optimising blind to everything but one number.

What you will not get is conversation-level insight. Aggregated reporting is the model, consistent with the privacy limits set out below. If your internal case for the channel depends on knowing what people asked before they clicked, that case cannot be built on this platform's data — a point worth settling before you commit budget. Brands weighing up how they already appear across assistants often start with a free AI visibility audit to get a baseline before they spend.

Which categories are allowed — and which are not

Eligibility is where most ChatGPT advertising plans succeed or stop, and the rules are narrower than on mature ad platforms. OpenAI's help centre states that it currently focuses on household and consumer goods, local services, travel and entertainment, and digital products and education.

The prohibited list is explicit. The same help centre article says OpenAI currently disallows adult content, alcohol and tobacco, counterfeit goods, gambling, political content, recreational drugs, scams and fraud, weapons, and other unsafe or sensitive categories.

Between those two lists sit the regulated categories, admitted case by case. Search Engine Journal reports that eligible health ad types include health insurance, dental services, supplements, vision products, medical testing, and hospitals and urgent care; that finance ads are approved by manual review, one advertiser at a time; and that housing and job listings were added on 10 August 2026.

Two further constraints shape where an approved ad can run. Search Engine Journal reports that ads cannot appear near political content or sensitive user contexts involving mental and personal health discussions. And OpenAI's help centre notes that audience and contextual targeting cannot exploit sensitive traits or discriminate.

The most important procedural point is that approval is not a badge your brand keeps. Per OpenAI's help centre, it reviews the advertiser, ad copy, image, targeting, destination and placement context, and approval is specific to the submitted setup rather than a blanket endorsement of a brand or industry. Change the creative or the landing page and you are back in review.

Plan for the rules to move, too. Search Engine Journal reports that OpenAI updated its ChatGPT ad rules four times in five months between March and August 2026, so eligibility can change between planning a campaign and launching it.

What advertisers never receive about users

The privacy boundary is unusually clearly drawn, and it is worth knowing precisely because it caps what targeting and reporting can ever do. According to OpenAI's help centre article on the basics of ads in ChatGPT, advertisers never receive chats, chat history, memories, name, email, precise location, IP address, or sensitive information such as health, mental health or political topics.

Within those limits, the platform still offers real audience tooling: OpenAI's help centre documents support for custom audiences and for campaigns built from product feeds. Product-feed buying in particular has matured — Search Engine Roundtable reports that as of 30 September 2026 advertisers can create multiple product feed campaigns and ad groups from a spreadsheet, with ad templates generated automatically.

If your brand is also working on being found inside assistant answers, the same audience is reachable two ways at once — paid placement beside the answer, and earned presence inside it.

What changed most recently in Ads Manager

The platform is still being built, which is an argument for testing early and for not over-engineering a long-term plan around today's feature set. Search Engine Roundtable reports four changes as of 30 September 2026.

Bulk product feed campaign creation lets advertisers build multiple campaigns and ad groups from a spreadsheet, with ad templates generated automatically. Product review status tracking in the Products tab lets advertisers follow review progress and see rejection reasons. The Insights API can now retrieve conversion events beyond a campaign's primary goal. And account settings now support separate fields for internal account name, public brand name and registered legal name for billing.

None of these is headline news on its own. Together they describe an ad platform moving from pilot to operational tooling — the sort of plumbing that only gets built when a channel is expected to carry real volume.

Conclusion

ChatGPT advertising is no longer an experiment: it runs in more than 60 countries and regions, it has three buying routes, CPC and CPM bidding, a Conversions API and a maturing Ads Manager. The constraints are equally concrete — a Free and Go audience only, a narrow set of focus categories, per-setup approval, aggregated reporting, and rules that changed four times in five months.

Those facts, not a forecast, are what should decide your answer. If your customers sit in the eligible categories and on the unpaid tiers, a measured test is reasonable now. If they are enterprise buyers on paid plans, or you need conversation-level attribution to justify spend, the platform cannot yet do what your case requires. Either way, the earned side of the same surface is worth working in parallel, which is what our generative engine optimisation work is for.