ChatGPT ads stopped being a rumor and became a product you can buy. OpenAI opened a self-serve Ads Manager at ads.openai.com with CPC bidding and conversion tracking. Here is how to actually set up a campaign, the two rules that decide whether it works, and why the constraint is your content supply, not the ad account.
For most of 2026, "ads in ChatGPT" was a direction, not a destination — a set of hiring signals and a single sponsored unit in testing. That changed. OpenAI first announced it was testing ads in ChatGPT on February 9, 2026, and on May 5, 2026 it opened a beta self-serve Ads Manager to US advertisers, added cost-per-click bidding alongside the original CPM buying, and shipped conversion measurement including a Conversions API. In plain terms: you can now go to ads.openai.com, register a business, and run a campaign inside ChatGPT yourself, without an agency in the middle.
This guide is the practitioner version — how to actually set one up and what makes it work. If you want the strategic read on where the ad formats are heading and what an ad surface inside an AI assistant means for creators, that is covered separately in the guide on image and video ads in ChatGPT, and the original launch is in the news write-up on ads coming to ChatGPT. Here we assume you want to buy.
The unit is deliberately restrained. A ChatGPT ad is a native sponsored placement — a brand name, a small favicon, a short headline, a one-line description, an image, and a link to your landing page — that appears at the bottom of a relevant answer. OpenAI has described it as having more in common with native ads than with traditional display banners, and the character and image limits are tight, so the creative discipline is closer to a search ad than a social one: say the thing in a few words, make the image legible small.
Richer formats — image, video, native, conversational, and interactive units, including a floated "ask the ad a question" experience — have been described as in development, but the buyable product today is the native unit. Do not plan a campaign around a format that has not shipped. Build for the unit that exists and treat the richer ones as upside when they arrive.
The self-serve flow will feel familiar to anyone who has run a search or social campaign, with a few AI-native wrinkles. Four steps matter.
Access runs through ads.openai.com (the self-serve beta opened to US advertisers first) or through technology partners such as StackAdapt. You register a business advertiser account, add a payment method, and you are in. Because it is a beta, expect the surface area to keep changing — custom audiences and measurement tooling have been rolling out in waves since the May launch.
You supply the native unit: brand name, headline, description, image, and destination URL. This is the step most advertisers underestimate, because the small canvas punishes generic creative. A blurry stock image or an off-message headline reads as noise under a helpful answer. You also cannot bring one asset and win — like any auction surface, ChatGPT ads reward testing multiple creatives against the same intent, so you need a set, not a single file.
You describe the topics and keywords your ad is relevant to, set a budget and pacing, and choose your bid — CPC if you want to pay for clicks, CPM if you want reach. The critical mental adjustment: the context you provide is a hint, not a lever. OpenAI is explicit that these are not exact-match keywords and do not guarantee your ad shows in a specific conversation. The model decides relevance from the actual chat, so you are steering, not targeting.
The Ads Manager reports impressions, clicks, click-through rate, average CPC and CPM, spend, and conversions. You can attach UTM parameters, use pixel-based measurement, and send server-side events through the Conversions API to track purchases, sign-ups, add-to-carts, and lead submissions after someone clicks. This is what moved ChatGPT ads from a brand-awareness experiment toward a performance channel — you can now close the loop on what a click was worth.
Setup is the easy part. Two properties of this surface, both different from social advertising, decide whether the spend returns anything.
A feed ad interrupts someone who came to scroll. A ChatGPT ad answers someone who just described, in their own words, a problem they want solved — often a problem with a purchase at the end of it. That is a higher-intent moment than almost anything in social, which is the real reason to care about this channel. But it cuts both ways: the ad only earns the placement if it is genuinely relevant to the question, and relevance is judged by the model against the conversation, not bought with a keyword. Sharp creative pointed at a well-defined intent wins; a broad, generic buy gets filtered out of the contexts where it does not belong.
This is the part that changes the math. Before any paid unit loads, ChatGPT is already answering "what should I buy" questions organically — naming tools, products, and brands inside the response. So every relevant conversation has two slots: the paid one you bid on, and the organic mention you earn. The organic one is free, sits above the ad, and carries the model's implied endorsement. You earn it the way you earn a search ranking — by having a broad, on-brand content footprint the model can find and cite. Ignoring it to focus only on the paid slot is like buying search ads while never publishing a page. Building that footprint is generative engine optimization; the guide on AI search visibility covers how it works.
Put the two rules together and the bottleneck becomes obvious. A CPC channel with conversion tracking is a performance loop: launch several creatives, kill the losers, refresh the winners, point each at a landing page that matches the intent, and repeat. That loop runs on a steady supply of on-brand creative and tight landing content — and it runs faster when the organic answer above the ad already names you, because a warm, cited brand converts a click that a cold one wastes. The ad account is a few forms. The thing that actually determines your results is how much relevant, on-brand content you can produce and keep producing.
That is a production problem, and it is where most advertisers stall. Shooting fresh creative for every test, writing a new landing page per intent, and simultaneously publishing enough organic content to get cited is more than a small team can hand-make at the cadence a live auction rewards.
Be clear about the boundary first. Kompozy does not bid in the auction or replace the OpenAI Ads Manager — you buy, target, and measure inside ChatGPT's own platform. Kompozy is the layer underneath it: the engine that produces the content the campaign consumes. It is an AI content generation and multi-platform publishing engine, not a repurposing tool, and it maps cleanly onto the two things a ChatGPT ad campaign is starved for — creative to test and a footprint to get cited.
On the creative side, Kompozy turns one input into a whole test set. The native unit needs a strong, legible image and a sharp headline; Photo Posts, Persona Photos, Quote Graphics, and Infographic Photos give you a library of on-brand stills to try instead of stock, and as the richer image and video formats open up, Persona Shorts, Marketing Shorts, and Clipped Shorts give you branded motion to run. Every asset is governed by one Persona Brief and a face-locked persona pool, so a dozen creative variants stay recognizably the same brand rather than drifting — which is exactly what you want when you are refreshing a CPC campaign weekly without booking a shoot.
On the destination side, the click has to land somewhere that matches the intent that earned it. Kompozy generates Blog Articles and text landing content tuned to the specific questions your ad targets, so the page a user hits continues the conversation instead of dumping them on a generic homepage. And the same engine builds the organic footprint that wins the free placement above your ad: Text Posts, Carousels, Blog Articles, and Email Newsletters fanned across nine social platforms plus blog and email from one scheduling queue, with a per-post review gate so nothing off-brand ships. That breadth is what an AI model draws on when it decides which brands to name — the content that lowers your effective CPC by getting you recommended before the auction even runs.
The point is that one content operation feeds both slots. The images and video you generate become ad creative; the articles and posts you publish become the citation footprint and the landing pages; and it all holds to a single brand identity because it comes from one engine. You are not running a separate creative studio for the ads and a separate content team for organic — you are running one supply line into a channel that rewards supply.
You can act now, and the moves compound. Register at ads.openai.com and ship one native creative against a single high-intent topic to learn the surface with real money and real data — small budget, tight targeting, CPC bidding so you pay for outcomes. Point that ad at a landing page built for the exact question, not your homepage. And start building the organic footprint in parallel, because the brands that win the cheapest, highest-converting ChatGPT placements will be the ones the model already recommends. The placement is new; the work that fills it — on-brand creative and citable content, produced faster than you can shoot it by hand — is work you can start today. For the wider context of where paid social is heading, see the guide on social media advertising in 2026.
Through OpenAI's self-serve Ads Manager at ads.openai.com, opened in beta to US advertisers on May 5, 2026. You register a business account, add a payment method, upload a native ad unit (brand name, headline, short description, image, and a landing-page link), set a budget and pacing, choose a bid, and launch. You can also buy through technology partners such as StackAdapt.
There is no fixed price — it is an auction. OpenAI started the pilot on a CPM (cost per thousand impressions) basis and, with the May 2026 self-serve launch, added CPC (cost per click) bidding so you can align spend with clicks and downstream actions. OpenAI has said it plans to add more bidding models over time. You set the budget and pacing; the auction sets the clearing price.
Ads show only to logged-in users on the Free and Go tiers; Plus, Pro, Business, Enterprise, and Education stay ad-free. At launch the eligible markets were the US, Canada, Australia, and New Zealand, with the UK, Japan, South Korea, Brazil, and Mexico named as planned expansions. No ads are shown to users under 18 or predicted to be under 18. The self-serve Ads Manager opened to US advertisers first.
Primarily by contextual relevance — an ad is matched to the topic of the conversation, so it appears against questions with buying intent. You supply topic and keyword context, but OpenAI is explicit that these are hints, not exact-match search keywords, and do not guarantee placement in a specific chat. Geographic targeting is available, and personalization from a user's past chats and memory is optional. Advertisers do not receive user chats, names, emails, or IP addresses.
No — OpenAI says the ad system is separate from the chat model and ads do not influence answers. But the answer above the ad still recommends brands organically, so a citable content footprint matters as much as the ad buy. Producing both the ad creative at volume and that organic footprint is a content-supply problem; an engine like Kompozy generates on-brand images, video, landing content, and cross-platform posts from one source to feed it.
You advertise in ChatGPT through OpenAI's self-serve Ads Manager at ads.openai.com, opened in beta to US advertisers on May 5, 2026. You register, upload a native sponsored unit, set a budget, and bid by CPC or CPM. Ads show only to logged-in Free and Go users in a handful of countries, are matched contextually to the question a person just asked, and run on a system OpenAI keeps separate from the answer itself.
Get started → · ← All guides · Compare Kompozy vs other tools