ChatGPT ads moved from a US test toward wider Free and Go delivery

OpenAI is expanding its advertising strategy across the US by introducing ads to all ChatGPT Free and Go users, signaling a shift towards more commercial models. This move comes as operational costs increase due to growing demand and advanced computing needs.

• OpenAI announced a planned US test in January 2026, began showing ads to some Free and Go users on 9 February, and told Reuters on 21 March that it would extend ads to all US users of those plans in the coming weeks. That sequence is a plan followed by a phased launch, not a universal rollout completed on the reporting date.
• Free and Go are eligible, but Go is a low-cost paid plan. Plus, Pro, Business, Enterprise and Edu are excluded under the published policy, as are accounts identified or predicted as belonging to people under 18.

The dates separate planning from delivery

On 16 January, OpenAI's initial advertising announcement said it was not launching ads yet and planned to test them in the coming weeks for logged-in US adults on Free and Go. The test began on 9 February for a subset of eligible US users. Axios independently described it as a test for some US Free and Go users.

On 21 March, a spokesperson told Reuters that OpenAI would begin showing ads to all US Free and Go users in the coming weeks. The statement expanded the intended audience; it did not establish that every eligible account saw an ad that day. OpenAI continued to call the product a pilot, and delivery depended on eligibility, a relevant match and staged availability.

OpenAI later said it would move beyond the United States to pilots in Canada, Australia and New Zealand, and its May update planned further expansion to the United Kingdom, Mexico, Brazil, Japan and South Korea. A country named in a future plan is not proof that delivery has started there. The durable status is therefore a live, expanding pilot rather than a single completed global launch.

Who may see an ad

The current consumer FAQ says ads may appear on Free and Go. It says Plus, Pro, Business, Enterprise and Edu accounts remain ad-free. Calling all eligible users “free-tier” users is inaccurate because Go is a paid, lower-cost subscription.

OpenAI says it does not show ads in accounts where a user says they are under 18 or the service predicts that they are under 18. Its public guidance on logged-out sessions has changed during the pilot, so the defensible claim is the account-level exclusion—not that age screening is complete in every session. Temporary Chats are described as ad-free.

Matching uses conversation context, but advertisers receive a narrower record

Ads appear below a response, carry a sponsored label and are separated from the answer. OpenAI says the ad system runs separately from the chat model and advertisers cannot shape, rank or alter the answer. Those are product-policy claims; they should be tested through placement, labeling and response-independence monitoring rather than treated as independently audited results.

For selection, OpenAI says it uses the current chat topic plus basic context such as general location or language. If personalization is enabled, it may also use the current thread and personalized response context, ad interactions, past chats and memory. Turning personalization off does not remove contextual ads: the current thread can still be used.

OpenAI says advertisers receive aggregated, non-identifying performance information such as views or clicks, not chats, history, memories, name, email, precise location, IP address or sensitive information. If a user directly messages an advertiser through an ad, the advertiser sees those messages. Clearing ads data stops its use for selection, with deletion from servers taking up to 30 days under the FAQ.

Ad controls and an ad-free option are different

Eligible users can turn personalization off, dismiss or report an ad, see why it appeared and clear ads data. Free users can choose an ad-free experience with lower message limits and reduced access to some tools; Plus and Pro provide paid ad-free routes. Disabling personalization alone does not make a Free or Go session ad-free.

The advertiser-facing ChatGPT Ads guide describes contextual delivery, CPM and CPC auctions and aggregate campaign measurement. None of those mechanics proves that ads finance a particular amount of free access, lower infrastructure costs or improve advertiser returns. OpenAI itself describes the platform as early and evolving.

The regulatory question is control effectiveness, not the existence of a policy

OpenAI's ad policies restrict placement near sensitive or unsafe contexts and set rules for ad content and landing pages. The policy changed in April and May, including a more specific approach to some medical, legal and financial advice contexts. The existence of a written rule is not evidence that age prediction, context classification, labeling or advertiser review works without error.

The US Federal Trade Commission's 2025 study of consumer AI chatbots includes OpenAI and seeks information about children and teens, monetization, disclosures, advertising, rule enforcement and the use or sharing of conversation data. It is a Section 6(b) study, not a finding that OpenAI violated the law. It does show where regulators are asking for measurable safeguards.

What remains unproven

Reuters linked advertising to revenue diversification and rising computing costs. That explains the business motive, but it does not prove that ads will subsidise access, reduce subscription pressure, improve profitability or reshape digital advertising. Those outcomes require user reach, ad load, conversion, retention, cost and trust data over time.

Watch for region-by-region in-product notices, actual eligibility by plan, changes to under-18 and logged-out handling, the difference between personalization controls and removing ads, changes to sensitive-context policy, advertiser measurement boundaries, complaints and regulatory action. Those events—not a roadmap sentence—show whether the rollout and its safeguards are operating as described.