Why it matters
ChatGPT Ads can appear after a conversation has revealed a need, constraint or trade-off. The planning advantage will come from matching a useful message to that situation—and measuring the next action without treating context hints as keywords.
01 · Start with the conversation
The targeting unit is a situation, not a search term.
OpenAI’s current advertiser documentation says ChatGPT Ads considers the context and intent of the conversation, the landing page, the ad title and copy, advertiser-provided context hints and, when ad personalisation is enabled, selected signals from a user’s broader ChatGPT experience. The ad itself appears below an eligible conversation and remains separate from the answer.
Context hints sit at ad-group level. They describe conversations, topics, keywords or needs in which an offer may be relevant, but OpenAI is explicit that they are not exact-match keywords and do not guarantee delivery in a particular conversation. Its guidance recommends descriptive phrases, focused ad groups and separate structures for meaningfully different products, audiences or use cases.
That changes the brief. A useful ad group should describe a customer situation with enough specificity to guide relevance: what someone is trying to decide, which constraints matter and when the product genuinely helps. My inference is that a list of category terms will be too thin. Conversational media planning needs a library of decision contexts, not a larger keyword export.
02 · Read the buying system
Familiar controls do not make this a familiar channel.
Ads Manager Beta now documents three campaign objectives. CPM is designed for reach, CPC for traffic and conversion-optimised CPC, or oCPC, for a selected downstream event. Advertisers pay per thousand impressions for CPM and per valid click for CPC and oCPC. OpenAI says eligible ads compete in a relevance-weighted, second-price auction intended to balance advertiser and user value.
That combination matters. A higher bid is not the whole decision, and OpenAI does not publish a mature performance benchmark for the beta. Delivery depends on objective, bid, relevance, creative, landing-page experience and the eligible conversations available. This is not evidence that conversational ads outperform paid search or social. It is evidence that the first test should isolate how context and message fit affect delivery before a team makes a channel-level efficiency claim.
Creative should follow the same logic. OpenAI recommends clear, specific, benefit-focused copy and multiple distinct variations for an offering. I would make those variations correspond to different decision situations rather than cosmetic headline changes. One ad might help a buyer compare options, another reduce setup risk and another make a time-sensitive next step easier. Each should land on a page that substantiates the exact promise made.
The conversation creates the opportunity; the ad and landing page still have to earn the next action.
03 · Build measurement first
The measurement stack is recognisable—and easy to overread.
OpenAI’s conversion documentation supports a browser Pixel, Conversions API or both. A click reference called “oppref” is appended to the landing-page URL and can be preserved through the journey and sent with server-side events. When the same conversion is sent through both methods, the same event ID is used for deduplication. Eligible advanced matching and modelled measurement may also contribute where available.
The reporting interface currently includes impressions, clicks, spend, CTR, average CPC, average CPM and attributed conversions. OpenAI also states that Ads Manager does not expose the individual ChatGPT queries or search terms that generated clicks. Platform totals can differ from an analytics or commerce system because of attribution windows, time zones, consent and storage conditions, event configuration, deduplication and modelling.
For senior marketers, the consequence is simple: agree the reconciliation method before evaluating the test. Preserve static UTM parameters and oppref, define one primary conversion event, validate Pixel and API events with shared IDs, and compare platform reporting with first-party outcomes on the same time basis. Treat modelled conversions as an estimate within the platform’s method, not as independently observed transactions.
04 · Design the first test
Learn the new behaviour before asking it to scale.
I would not begin with a broad brand campaign or a direct efficiency comparison against a mature channel. Start with one offer that has a clear decision journey, enough eligible demand to deliver and a conversion that the business already trusts. Separate a small number of customer situations into focused ad groups, then write genuinely different messages for each.
The first learning agenda should be diagnostic. Which situations deliver? Which messages earn clicks without creating weak downstream behaviour? Where does the landing page fail to continue the conversation? Which conversions appear in both first-party systems and Ads Manager, and where do the methods diverge? Those questions are more useful than declaring a platform winner from an early blended CPA.
ChatGPT Ads is still a beta, and its inventory, formats, buying tools and optimisation will continue to change. That is a reason to document the experiment, not a reason to wait for certainty. My view is that the durable capability is conversational media planning: translating real decision contexts into relevant messages, verifiable landing-page evidence and a measurement chain the team can defend.
What marketers should do next
Turn the signal into a better decision.
- Choose one offer with a clear decision journey and a trusted downstream conversion; do not use the first test as a catch-all brand campaign.
- Write a short library of customer situations, constraints and decision questions, then separate meaningfully different contexts into focused ad groups.
- Create distinct benefit-led ads for those situations and make each landing page substantiate the exact promise carried from the conversation.
- Implement Pixel and Conversions API measurement with oppref preservation, shared event IDs, consent review and a documented reconciliation window.
- Evaluate delivery, message-context fit and first-party conversion quality before comparing the beta’s CPA or CPM with mature channels.
Sources & further reading
01OpenAI Help Center — Ads in ChatGPT: The Basics02OpenAI Help Center — Create Ad Groups for ChatGPT03OpenAI Help Center — Create Campaigns for ChatGPT04OpenAI Help Center — Create Ads for ChatGPT05OpenAI Help Center — Conversion Measurement06OpenAI Help Center — Measure Results07OpenAI Help Center — ChatGPT Ads Beta FAQs08OpenAI — Ad Policies