OpenAI began beta operations for ChatGPT ads in Korea on June 19, 2026, and moved to full operations in August. Soon after the local beta began, Elephant Team ran a proof of concept (PoC) and learned of a beta launch for “Sponsored Agents,” which let users talk with a brand’s AI after clicking an ad. ChatGPT ads connect ad creative to landing pages based on conversational context rather than a list of search results. Ads appear separately from AI answers and are shown to adult users on the Free and Go plans. Advertisers can use reach, click, and conversion objectives, as well as standard ads and product feeds, and review performance through aggregated, de-identified data. They cannot access users’ chats, histories, memories, or personal information.

Elephant Team ran separate reach, click, and conversion campaigns for decision-makers and practitioners at B2B companies seeking generative engine optimization (GEO) marketing. The reach campaign ran from August 27 to September 3, while the click and conversion campaigns ran from August 31 to September 7. Landing pages used Elephant’s blog content and a GEO report analyzing data from 548 companies. Each campaign had four ad groups combining keyword-based or prompt-based context hints with either decision-makers or practitioners. Context hints are signals used to assess relevance between a conversation and an ad; they are not commands that target only people who entered a specific sentence.

The ad images appeared, by the team’s estimate, at 150–200 pixels, so it kept each image to one message and supplied supporting evidence in the text and landing page. Decision-makers saw GEO vendor selection criteria and analysis of data from 548 companies; practitioners saw a GEO diagnostic checklist and implementation guide. The team recommended setting one objective—reach, clicks, or conversions—for each campaign; grouping audiences by the intent behind their questions rather than age or gender; and testing keyword and prompt hints separately. It also advised matching ad copy to landing-page messaging, using one message in the small ad space, and checking UTM parameters and GA4 events before launch. Elephant Team used the shared parameter utm_source=chatgpt-ads, distinguished cpm, cpc, and cpa by objective, and said operators should verify that URL parameters pass through correctly.

Responses varied by campaign goal. The reach campaign broadened exposure, while the click campaign produced relatively stronger post-visit behavior. The conversion campaign had the highest CTR and lowest CPC, but only limited consultation activity was observed, so the team treated it as an initial check of the contexts and creative that drove landing-page visits. In GA4, the conversion campaign had the largest volume of incoming traffic, while visitors from the click campaign showed relatively stronger post-visit behavior. Media-reported clicks may not match meaningful actions on a site; duplicate clicks, measurement criteria, and parameter settings can contribute to differences between media reports and GA4 user data. The PoC also involved an operational misunderstanding of automatic UTM substitution. For the next test, the team proposed checking each ad group’s parameters and conversion events in advance.

In the reach campaign, CTR was 0.55% for prompt ads and 0.37% for keyword ads; in the click campaign, it was 0.76% and 0.61%, respectively. The conversion campaign reversed that pattern: CTR was 1.13% for keyword ads and 0.94% for prompt ads, and keyword ads also had relatively lower CPC. The consultation activity that the team confirmed came from prompt ads. Elephant Team said the test did not show that either format was always better. It said prompts may be better suited to capturing problems that are not expressed as short search terms, while keywords may better capture already specific interest.

In the click campaign, CTR was similar for decision-makers and practitioners, at 0.69% and 0.68%, respectively, but CPC was lower for decision-makers. In the conversion campaign, CTR was 1.23% for decision-makers and 0.87% for practitioners. The team said practitioners’ information searches helped build exposure, while questions with specific aims—such as budgets, performance, or vendor selection—more readily led to clicks. Overall CTR was 0.53% for reach, 0.68% for clicks, and 1.01% for conversions. The team said these figures were not strikingly different from general display advertising, but ChatGPT ads appeared in a small space during conversations where users described problems and constraints, with one company’s ad shown on its own per screen.

Elephant Team said operating ChatGPT ads differs from performance advertising centered on bids, audience segmentation, and retargeting. It recommended sorting customer questions into problem awareness, alternative comparison, and purchase consideration; preparing content and landing pages that answer those questions; and using ad responses to imp

OpenAI began beta operations for ChatGPT ads in Korea on June 19, 2026, and moved to full operations in August. Soon after the local beta began, Elephant Team ran a proof of concept (PoC) and learned of a beta launch for “Sponsored Agents,” which let users talk with a brand’s AI after clicking an ad. ChatGPT ads connect ad creative to landing pages based on conversational context rather than a list of search results. Ads appear separately from AI answers and are shown to adult users on the Free and Go plans. Advertisers can use reach, click, and conversion objectives, as well as standard ads and product feeds, and review performance through aggregated, de-identified data. They cannot access users’ chats, histories, memories, or personal information.

Elephant Team ran separate reach, click, and conversion campaigns for decision-makers and practitioners at B2B companies seeking generative engine optimization (GEO) marketing. The reach campaign ran from August 27 to September 3, while the click and conversion campaigns ran from August 31 to September 7. Landing pages used Elephant’s blog content and a GEO report analyzing data from 548 companies. Each campaign had four ad groups combining keyword-based or prompt-based context hints with either decision-makers or practitioners. Context hints are signals used to assess relevance between a conversation and an ad; they are not commands that target only people who entered a specific sentence.

The ad images appeared, by the team’s estimate, at 150–200 pixels, so it kept each image to one message and supplied supporting evidence in the text and landing page. Decision-makers saw GEO vendor selection criteria and analysis of data from 548 companies; practitioners saw a GEO diagnostic checklist and implementation guide. The team recommended setting one objective—reach, clicks, or conversions—for each campaign; grouping audiences by the intent behind their questions rather than age or gender; and testing keyword and prompt hints separately. It also advised matching ad copy to landing-page messaging, using one message in the small ad space, and checking UTM parameters and GA4 events before launch. Elephant Team used the shared parameter utm_source=chatgpt-ads, distinguished cpm, cpc, and cpa by objective, and said operators should verify that URL parameters pass through correctly.

Responses varied by campaign goal. The reach campaign broadened exposure, while the click campaign produced relatively stronger post-visit behavior. The conversion campaign had the highest CTR and lowest CPC, but only limited consultation activity was observed, so the team treated it as an initial check of the contexts and creative that drove landing-page visits. In GA4, the conversion campaign had the largest volume of incoming traffic, while visitors from the click campaign showed relatively stronger post-visit behavior. Media-reported clicks may not match meaningful actions on a site; duplicate clicks, measurement criteria, and parameter settings can contribute to differences between media reports and GA4 user data. The PoC also involved an operational misunderstanding of automatic UTM substitution. For the next test, the team proposed checking each ad group’s parameters and conversion events in advance.

In the reach campaign, CTR was 0.55% for prompt ads and 0.37% for keyword ads; in the click campaign, it was 0.76% and 0.61%, respectively. The conversion campaign reversed that pattern: CTR was 1.13% for keyword ads and 0.94% for prompt ads, and keyword ads also had relatively lower CPC. The consultation activity that the team confirmed came from prompt ads. Elephant Team said the test did not show that either format was always better. It said prompts may be better suited to capturing problems that are not expressed as short search terms, while keywords may better capture already specific interest.

In the click campaign, CTR was similar for decision-makers and practitioners, at 0.69% and 0.68%, respectively, but CPC was lower for decision-makers. In the conversion campaign, CTR was 1.23% for decision-makers and 0.87% for practitioners. The team said practitioners’ information searches helped build exposure, while questions with specific aims—such as budgets, performance, or vendor selection—more readily led to clicks. Overall CTR was 0.53% for reach, 0.68% for clicks, and 1.01% for conversions. The team said these figures were not strikingly different from general display advertising, but ChatGPT ads appeared in a small space during conversations where users described problems and constraints, with one company’s ad shown on its own per screen.

Elephant Team said operating ChatGPT ads differs from performance advertising centered on bids, audience segmentation, and retargeting. It recommended sorting customer questions into problem awareness, alternative comparison, and purchase consideration; preparing content and landing pages that answer those questions; and using ad responses to imp