ChatGPT ads:
what a US business
can buy today

As of September 25, 2026, a US business can buy ChatGPT ads on its own, through OpenAI’s self-serve Ads Manager, which is still a beta. The United States is on OpenAI’s country list for self-serve buying. The ads run below ChatGPT’s answers, for signed-in people on the Free and Go plans and in logged-out use, and OpenAI matches them to the conversation using its context, the ad and its landing page, and plain-language context hints rather than keywords.

For most categories, buying is the simpler part. The choices that are hard to undo come at setup, and the measurement has limits a business that sells through inquiries should plan around before it spends. For how ChatGPT Ads compares with ads in Google’s AI Overviews and Microsoft Copilot, see what answer-engine advertising is today. Ads that open a conversation with a business’s own agent are covered in An AI agent in an ad still speaks for the business.

Who can buy ChatGPT ads in the US

Eligibility follows the business, not the person signing in. For self-serve buying, “the legal entity that will advertise and be billed” must be based in an available country; a ChatGPT login’s country does not override it. Three account rules catch agencies and growing businesses:

• Each advertiser needs its own account. Through the Ads Manager app inside ChatGPT, only business accounts are supported, and an agency cannot create the account for a client; the client creates it and invites the agency. An agency that submits campaigns for a client needs a written agreement with that client under OpenAI’s Advertising Terms.
• The country, billing currency and time zone chosen at setup cannot be changed later.
• Some new self-serve accounts can advertise only in their home country at first. Other countries open after identity verification and a level of spend OpenAI does not publish.

Self-serve is one route. OpenAI’s sales team works with larger advertisers, agencies and ad-tech companies buy for clients, and the Advertiser API needs an ad account and a key. Buying also runs through an Ads Manager plugin in ChatGPT, a Shopify app for US merchants and HubSpot.

OpenAI reviews each business against its ad policies, and business verification can take up to 7 business days. The same help page has a US column for regulated categories. It allows legal services, for advertisers licensed to practice law in the US, along with minimally invasive cosmetic procedures, dietary supplements and auto loans. It restricts health insurance, hospitals and urgent care, mortgages and personal loans to select approved advertisers, and OpenAI is not accepting applications in some restricted categories. Telehealth, prescription medicines, credit repair, debt relief and payday loans are not allowed. The ad policy also bars alcohol, tobacco, gambling, adult content, political ads and individual job or housing listings.

The two pages do not fully agree. The policy treats US insurance, dental and vision services and several other health and financial categories as case-by-case approvals, while the help table marks some of them Allowed and does not list dental or vision. For legal services, the policy asks for a license in the jurisdiction where the ad is shown, which is narrower than the table’s US license. Digital Traction’s reading is that the review, not the table, settles a borderline category.

Who sees the ads, and how an ad is matched

Among signed-in accounts, ads show only on the Free and Go plans. Plus, Pro, Business, Enterprise and Edu accounts see none, and neither do accounts OpenAI identifies as belonging to people under 18. Temporary Chats and the ChatGPT Atlas browser carry none, and ads stay away from topics such as personal health, mental health and politics.

The audience is wider than at launch. The February launch post described a test for “logged-in adult users”. The help center now says ads shown in logged-out experiences are “appropriate for all audiences”, with mature-content ads limited to eligible logged-in users.

There are no keywords to bid on. OpenAI’s system weighs the conversation’s context and intent, the ad’s landing page, title and copy, the advertiser’s context hints and, where ads personalization is on, select signals from the person’s wider ChatGPT use, then runs “a relevance-weighted, second-price auction”.

Context hints describe the moment an ad fits. They are “not exact-match keywords or targeting rules”, and they “cannot enforce geographic limits, schedules, or exclusions.” OpenAI asks for “a clear, natural phrase focused on one idea”, and an ad group can hold up to 2,000. For a plumbing business, one hint might read “a homeowner deciding whether to repair or replace an older water heater.” Digital Traction’s intent map holds those buyer moments, written once, so the same moment can become a context hint in ChatGPT and a keyword or search theme in Google.

Hard limits go in the campaign settings instead:

• Location. Beyond countries, a campaign can target states or regions, cities, markets and postal codes where available, though new product-feed campaigns are limited to countries. To run only in the US, include the United States explicitly; excluding other countries alone does not restrict delivery.
• Audiences. Customer lists can include, exclude or adjust bids, but inclusion and bid adjustments need at least 25,000 matched users. Smaller lists work as exclusions. OpenAI’s Ad Tools Terms require first-party lists, not data bought from brokers.

The landing page is an input too. Advertisers must allow OpenAI’s OAI-AdsBot crawler, and the future of search covers what that means for a thin service page. OpenAI now suggests copy and imagery in Ads Manager and offers opt-in text customization that adapts headlines and descriptions to the conversation. OpenAI’s help page on suggested ad drafts, last updated in August, says that feature prefills from website metadata and “does not generate new copy or imagery with AI”, and no help page yet describes the September tools. Under the Advertising Terms, the advertiser answers for AI-generated outputs (what the terms put on the business).

What is fixed at launch, and how billing works

A campaign takes one of three objectives: Views, billed per 1,000 impressions; Clicks, billed per valid click; or Conversions, which bids toward one standard conversion event and bills per valid click or per 1,000 impressions. In OpenAI’s words, “You do not pay per conversion.” Conversion bidding billed per click is in open beta. Billed per impression, it is “currently in beta, and access is rapidly expanding.”

Three choices cannot be changed in place:

• The objective and billing option. Moving a conversion campaign from click to impression billing means cloning it; any other change means a new campaign.
• The conversion event. It is set at creation, and custom events are not supported for this objective.
• The budget type. OpenAI’s billing page says it cannot be changed after creation, while its campaign page says a campaign-total budget can move to daily but not back. Treat it as fixed.

Maximize results is the default for eligible new ad groups. It favors volume and promises no cost-per-acquisition or return target; manual bid controls exist for a business that needs a limit. A daily budget is an average over seven days: one day’s spend can reach twice the daily budget, and a week’s spend stays within seven times it.

ChatGPT Ads bills after delivery. A self-serve card is charged when unpaid spend reaches the account’s assigned threshold, or at month end. Pausing is not instant: ads may keep appearing for up to 24 hours, and that spend is billable. The Advertising Terms give a business thirty days from the activity to dispute fees, and any adjustment comes only as a makegood or credit. The terms also make no promise that impressions and clicks are human or valid.

What ChatGPT Ads can and cannot measure

Measurement runs through the OpenAI Pixel on the site and the Conversions API from a server. When someone clicks, a click reference called oppref is added to the landing page URL, and the Pixel stores it in a first-party cookie. The standard events include lead_created and appointment_scheduled, but none of the 12 describes a qualified lead. Five details matter for a business that sells through inquiries:

• Bidding stops at the inquiry or the booking. Conversion bidding does not support custom events, and the Conversions API accepts only events timestamped within the last 7 days, so CRM outcomes have to be sent within a week, not in a monthly batch. Digital Traction’s reading is to bid on the inquiry or the booked appointment and judge qualification in the CRM, where the later outcome still matters.
• Consent is on by default. The Pixel initializes consent to true, so where consent is required, the site has to set it to false until the visitor agrees.
• Reported totals may include modeled conversions where available.
• Reporting windows are a choice: 7, 14 or 30 days after a click, plus an optional 1-day view-through window. Changing them affects reporting only, attribution is last-touch, and conversions can take 24 to 48 hours to appear.
• There are no benchmarks. OpenAI says ChatGPT Ads “does not yet have performance benchmarks” across advertisers, industries or campaign types.

For the business’s own analytics, static UTM parameters persist on ad clicks, and Ads Manager fills macros such as {campaign_id} and {ad_id}. GA4 files arrivals from ChatGPT under its AI Assistant channel, so Digital Traction’s reading is that paid tagging is what keeps ad visits apart from unpaid ones (where AI traffic lands in GA4).

What OpenAI has not published

As of September 25, 2026, OpenAI’s pages do not publish:

• Reserve prices, auction weights or frequency caps
• Auction insights, or any report of which conversations or hints matched an ad
• A lift, holdout or incrementality tool of OpenAI’s own
• What the exclusion_hints field listed in the Bulk API does
• The spend that opens countries beyond an account’s home country
• How a valid click is defined
• A way to send CRM outcomes older than 7 days
• Whether logged-out users can be targeted or excluded
• A help page for the AI text customization announced on September 16

What to check before a first test

A first test reads cleanly when the account, the tags and the comparison are settled before any spend:

1. Confirm the category in OpenAI’s US table and the ad policy, and let the review clear before building creative.
2. Create the account under the business’s own legal entity, based in the US, with the currency and time zone it will keep, and give any agency access from there.
3. Include the United States, or the states and markets being tested, in the campaign locations.
4. Decide the objective, billing option, conversion event and budget type once, and write them down.
5. For lead generation, bid toward lead_created or appointment_scheduled, and read qualified leads in the CRM.
6. Tag every ad URL with paid UTM parameters and ID macros, keep oppref through redirects, and set Pixel consent to false where consent is required.
7. Pick the reporting window before launch and read against it, noting view-through and modeled conversions.
8. Design any holdout outside the platform. OpenAI publishes no lift tool of its own, and its listed measurement partners’ capabilities vary, but state and market targeting allow a geographic split; campaign-level lift is not account-level lift explains why the account-level read is the one that counts.
9. Exclude current customers with a first-party list, which works below the 25,000 minimum.
10. Allow OAI-AdsBot and keep the landing page complete.
11. Reconcile the invoice with Ads Manager inside the thirty-day dispute window, including spend after any pause.

One tension to plan for: OpenAI recommends “a high volume and diverse set of ads”, while a test reads best with few variables. Digital Traction’s reading is to vary copy inside one ad group and hold everything else steady.

For ChatGPT ads, Digital Traction’s part in Advertising in AI is the research, the test design and the measurement. Any launch happens in the business’s own ad account, under the person who runs it.