TLDR;
OpenAI began testing ChatGPT ads with logged-in adults on Free and Go in the US. Ads were labelled and advertisers received aggregate reports, not conversation transcripts. Brands should run a measured test against profitable conversions and keep it separate from organic recommendation tracking.
What happened
OpenAI began testing labelled ads for logged-in adults on Free and Go in the US. Ads appeared separately from answers, with sensitive-topic exclusions and aggregate advertiser reporting; higher paid tiers were excluded from the initial test.
Why it matters
The test's tier, age and geography restrictions determine who can encounter the placements. Separate labels and aggregate reporting also shape what an advertiser can infer. Enterprise teams should assess this as a measured channel experiment, with no assumption that they receive conversation transcripts or that spend changes the assistant's organic advice.
The US test makes the earlier advertising plan operational for a defined group of users. That is an important change for advertisers, but it remains a test with eligibility limits, labelled placements and specified reporting boundaries. The enterprise question is whether the inventory produces useful customer action under those conditions. The context of an assistant conversation may make the placement relevant, yet advertisers should not assume they receive the conversation or can reconstruct the user's private intent. Nor should a sponsored placement be confused with the assistant's organic judgement. Keeping those distinctions clear protects both measurement and creative quality. A limited launch is suitable for a controlled learning budget, especially where the available reports cannot yet support the level of attribution the brand uses in mature channels.
How your brand can benefit / be affected
Confirm access, supported objectives and sensitive-topic exclusions. Set a capped test budget and use a clear offer with a landing page that answers the product question the ad raises.
Reconcile available ad reporting with site-side conversions and contribution after costs. Document unavailable journey data rather than filling the gaps with assumptions. Keep the initial US test distinct from subsequent regional expansions when comparing results or describing the channel's availability.
Choose a product and destination that can serve a clear need without requiring speculative knowledge about the user. Make the advertised promise match the page and the actual terms. Check category restrictions and audience eligibility before building the campaign, rather than treating all ChatGPT usage as reachable inventory. Set a small number of commercial questions for the test: whether the placement attracts qualified interest, whether the destination converts and whether the resulting customers fit the business. Prepare the analytics that the supported buying setup permits. Avoid making the pilot dependent on reporting or targeting capabilities that have not been documented as available.
Review results using the provider's stated definitions and your own downstream evidence. Separate ad delivery, site activity and qualified outcomes, and note where aggregate reporting leaves uncertainty. Investigate misleading creative, poor landing-page fit or low-quality enquiries before increasing spend. Keep organic citation activity in a different reporting stream so the test does not inflate claims about GEO performance. Involve customer-facing teams if the ads generate questions they have not previously handled. The useful outcome is a defensible decision to continue, revise or stop the pilot. Participation by itself does not show that a new channel deserves a permanent budget or that early performance will hold at scale.
News date: 9 February 2026. Editorial review: 16 September 2026. Analysis includes subsequent developments where stated.