ChatGPT Ads: everything OpenAI has confirmed so far

Where ChatGPT Ads stands now

Eight months after the first test, ChatGPT Ads is a real ad platform. OpenAI says it hit a $1 billion annualized revenue run rate in under 200 days, works with tens of thousands of advertisers, and runs ads in more than 60 countries. You can buy it yourself through the Ads Manager beta in most of those markets.

It's still early. OpenAI hasn't published performance benchmarks, several features only reach "qualifying" advertisers, and new formats are still in testing. Treat it as a channel to test, not one to plan a quarter around.

Timeline

Date	What happened
Jan 16, 2026	OpenAI announces it will test ads for Free and Go users in the US, and publishes its ad principles.
Feb 9, 2026	The US test starts for logged-in adults on Free and Go. Buying is CPM only.
Mar 2026	The first version of OpenAI's ad policies is published.
Mar 26, 2026	OpenAI says pilots in Canada, Australia and New Zealand will start in the coming weeks.
May 5, 2026	The self-serve Ads Manager beta starts rolling out and CPC bidding launches. OpenAI names its agency and ad tech partners. Pixel and Conversions API measurement had launched shortly before.
May 7, 2026	OpenAI announces pilots in the UK, Mexico, Brazil, Japan and South Korea, all live by Aug 11.
By Aug 18, 2026	Conversion-optimized bidding, location targeting and custom audiences are added.
Aug 31, 2026	Self-serve opens across 31 European markets, India, the Middle East and North Africa. OpenAI reports a $1 billion run rate.
Sep 16, 2026	OpenAI starts testing Sponsored Agents with select US advertisers, rolls out an Ads Manager plugin inside ChatGPT and AI creative tools, and launches HubSpot and Shopify (US merchants first) integrations.
Sep 23, 2026	Ads roll out in Indonesia, Malaysia, the Philippines, Singapore, Thailand, Vietnam and Taiwan.
Oct 5, 2026	Negative Phrases (for qualifying advertisers) and more measurement partners arrive. A visual ad format inside image generation will be tested in the US later in October.

Who sees the ads

• Plans: Free and Go only. Plus, Pro, Business, Enterprise and Edu accounts don't get ads.
• Age: no ads for accounts identified as under 18, based on the stated age and OpenAI's age prediction.
• Logged-out visitors: can see ads, but only ones suitable for all audiences.
• Places with no ads: Temporary Chats, and (during the test) the ChatGPT Atlas browser.
• Opting out: Free users can choose an ads-free mode with fewer daily messages and no image tools. Go users have to switch to Free or upgrade.
• Personalization: users can turn off ad personalization, see why they got an ad, hide or report it, and clear their ads data. Personalized ads aren't available in the EEA or Switzerland yet.

Where ads show and what they look like

Ads appear below the end of a response, labeled as sponsored and visually separated from the answer. One unit can hold one or more items, from one advertiser or several. Each ad has your business name, favicon, a headline, description copy, a landing page and an image.

• AI text customization (launched Sep 16): an opt-in setting.

Formats still in testing:

• Visual ads during image generation: an ad shown while ChatGPT creates an image, kept separate from the image. Testing starts in the US later in October with an initial group of advertisers.
• Sponsored Agents: after seeing an ad, a user can choose to start a labeled conversation with a business's own agent. This is in testing with select US advertisers. The business only sees the messages sent to it.

How to buy

Ways in:

• The self-serve Ads Manager beta. OpenAI also offers an Ads Manager plugin that runs inside ChatGPT, with plain-language commands.
• OpenAI's Ads Solutions team, for larger advertisers.
• Agencies (dentsu, Havas, Omnicom, Publicis, WPP).
• Ad tech partners (Adobe, Criteo, Kargo, Pacvue, StackAdapt). With partners, OpenAI still makes every delivery decision.

Eligibility: the legal entity that advertises and gets billed has to be based in a country on OpenAI's availability list. Some new accounts can only advertise in their home country until identity verification clears and they reach a spend threshold there.

Bidding: you choose views (CPM), clicks (CPC) or conversions. Conversion campaigns are billed per click or per impression, never per conversion. The auction is relevance-weighted and second-price. OpenAI's suggested starting point is a $3 to $5 max CPC.

Budgets: the minimum daily budget is $25 in the US (€15, £15, CA$25, A$25, ¥2,500, ₹725). Daily budgets work as a 7-day average, so one day can spend up to twice the daily amount.

Billing: self-serve accounts pay by credit card after they spend. You're charged when unpaid spend hits your payment threshold (for example $25), and at month end for any balance. OpenAI may raise the threshold automatically after successful payments. Billing country, currency and legal entity may not be editable yourself after setup. Contact support if they're wrong.

Targeting and controls

There are no keywords. OpenAI matches ads using the conversation's context and intent, your landing page, title and copy, your context hints and targeting, and, for users who allow it, personalization signals from their wider ChatGPT use.

• Context hints: set on each ad group. OpenAI says plainly that they aren't exact-match keywords and don't guarantee delivery.
• Locations: countries, plus states, cities, markets and postal codes. Product feed campaigns are country-level only.
• Platforms: Android and iOS apps and web, and desktop web.
• Custom audiences: upload emails, phone numbers (plain or SHA-256 hashed) or Google Advertising IDs to include, exclude or adjust bids from 0.1x to 10x. Including an audience or using bid adjustments needs at least 25,000 matched users. OpenAI advises against custom audiences when you target the EEA or Switzerland.
• Negative Phrases: keep ads off conversations that conflict with your brand's own rules. Only qualifying advertisers have it so far, and OpenAI hasn't defined what qualifying means.
• Brand suitability: OpenAI says its own guardrails keep ads out of sensitive contexts. OpenAI is developing brand suitability evaluation pilots with DoubleVerify and IAS, run without access to user conversations.

Measurement

• Tracking: a pixel and a Conversions API cover purchases, leads, installs and sign-ups, with an Event Quality Score.
• Reporting: impressions, clicks, CTR, spend, average CPC and CPM, and conversions. Clicks and impressions refresh about every 15 minutes. Spend can lag 7 to 8 hours.
• Attribution windows: you choose a click-through window and can add a 1-day view-through window. These change reporting only, not optimization.
• Partners: Hightouch, Tealium and LiveRamp for conversion data. AppsFlyer, Triple Whale, Adjust, DV Rockerbox, Northbeam, Branch, Singular, Kochava, Airbridge and Tenjin for attribution. Fospha, Measured and INCRMNTAL for full-funnel measurement.
• Incrementality: OpenAI calls its geo-experiment work with Haus, Measured and WorkMagic "still in its early stages." Brand lift testing with Kantar and Cint is also early.

The results OpenAI quotes, like one advertiser's CPA coming in 15.3% below its paid search benchmark, are single-advertiser numbers reported by vendors. They're useful as a signal, not a benchmark.

What isn't allowed

• Content: no political ads.
• Categories: no adult or dating, alcohol or tobacco, gambling, recreational drugs, counterfeits, credit repair or debt settlement, or individual job and housing listings.
• Claims: no unsupported wellness claims, and no ads that imitate ChatGPT's interface.
• Regulated services: In the US only, OpenAI may approve health and financial advertisers case by case, and allows legal services from advertisers licensed where the ad shows.
• Placement: ads don't run near personal health, mental health, politics, self-harm or other sensitive conversations.
• Data: advertisers never get chats, chat history, memories, names, emails, precise location or IP addresses. OpenAI says it never sells user data to advertisers.

What I'd do first

1. Connect measurement before you spend. Install the pixel or Conversions API. If you already use one of the named attribution partners, connect it so you aren't judging the channel on pixel data alone.
2. Start with clicks. CPC at OpenAI's suggested $3 to $5 is the easiest way to learn what traffic is worth before you move to conversion bidding.
3. Write context hints like a search term report. Use the problems and comparisons people bring to ChatGPT, not your brand terms. Landing page quality matters, because the landing page is part of how ads get matched.
4. Keep the test small and separate. OpenAI hasn't published performance benchmarks, so judge it against your own paid search numbers, with its own budget and its own success metric.

Reported but not confirmed by OpenAI

Some widely shared details come from press or partner reports, not OpenAI's pages. These include a large minimum commitment for early pilot advertisers, reported by Adweek, and claims that the minimum was dropped when self-serve launched. They also include partner minimums at StackAdapt and Criteo, CPM price ranges, and specific limits on Negative Phrases. I'll add any of them here once OpenAI confirms.