OpenAI launched the official ChatGPT advertising service in August, following a beta test in Korea on June 19, 2026. This marks the beginning of an era where brands can display ads within the context of a user's conversation , rather than in search result lists
The Elifunt team immediately began testing ChatGPT ads right after the domestic beta launch. While conducting the tests and organizing the lessons learned, we heard the news of the beta launch for 'Sponsored Agents,' which allow users to continue a conversation with a brand AI after clicking an ad. ChatGPT ads are now rapidly evolving beyond simple banner exposure into a conversational advertising ecosystem that connects product comparison, lead acquisition, and purchase review.
A New AI-Based Experience for ChatGPT Ads – Sponsor Agent
The marketing environment is changing even as marketers hesitate while exploring new products. In times of high uncertainty, it is crucial to build your own unique lessons by trying things out yourself, rather than waiting for a perfect plan.
In this article, the Elifunt team shares the concerns and honest lessons learned while executing the ChatGPT ad PoC, based on their experience in the field of organic, SEO, and GEO marketing
How are OpenAI ChatGPT ads different from search ads?
OpenAI ChatGPT ads match relevant ad creatives and landing pages based on the context and intent of the conversations users share. Ads are displayed in an area separate from AI responses and are shown to adult users on the Free and Go plans.
Currently, you can run campaigns aimed at reach, clicks, and conversions. It supports standard ad and product feed formats, and advertisers can view aggregated, non-identifiable performance data such as views and clicks. However, advertisers cannot access users' chat content, history, memory, or personal information.
You must design the context of the 'conversation' rather than the 'keywords' that customers might search for
Search ads display relevant advertisements based on the search terms entered by the user. In contrast, ChatGPT ads analyze the entire context of the conversation, including the user's problem, goal, and conditions, to match appropriate advertisements.
| division | ChatGPT ads | Search ads |
|---|---|---|
| User behavior | Explain the situation and concerns through conversation | Enter the necessary information or products as search terms |
| Key Intention Signal | Context of the question, goal, conditions, flow of conversation | Search terms, keywords, match types |
| Key exposure point | The process of defining a problem and exploring and comparing solutions | The point at which the product or service being sought becomes relatively clear |
| Core of operations | Context hints, relevance between material and landing | Keywords, bidding, audience settings |
| Expected role | Discovering search demand that has not yet become a search term | Capturing already established demand and purchase intent |
You can increase the likelihood of exposure to customers through context hints
You can enter keyword-based words or sentence-based prompts as context hints on the ChatGPT ad settings screen. However, this is not a command to “show the ad only to people who entered this sentence.” It is a signal that the ad system refers to when determining the relevance between the conversation and the creative.
Therefore, in ChatGPT ads, rather than narrowing the target audience by demographics, you must design specifically to understand what problems customers ask about and how they phrase them.
I tested ChatGPT ads like this
1. Test Objective: “Can we access the conversations of desired B2B customers?”
The target audience for this PoC was decision-makers and practitioners at B2B companies requiring GEO marketing. Campaign objectives were divided into three categories—reach, clicks, and conversions—and context hints and target roles were differentiated within each objective.
- Reach Campaign: Exposing the brand and message to as many potential customers as possible.
- Click Campaigns: Encourage users who viewed ads to visit blog or report landing pages.
- Conversion Campaign: Inducing specific actions such as checking GEO reports and requesting consultations
The reach campaign ran from August 27 to September 3, and the click-and-conversion campaign ran from August 31 to September 7. For the landing page, we utilized Elifunt blog content and a GEO report analyzing data from 548 companies.
ChatGPT Ads Test Structure and Actual Ad Impression Examples
2. Ad Group: Keywords · Prompts × Decision Makers · Practitioners
Each campaign has four ad groups configured: Keyword × Decision Makers, Keyword × Practitioners, Prompt × Decision Makers, Prompt × Practitioners.
4 ad groups divided based on context hint types and target roles
📍 Examples of Keyword-based Context Hints — Decision Makers: GEO company, GEO consulting, SEO agency, GEO performance measurement / Practitioners: Meaning of SEO, How to do GEO, Blog SEO, Website SEO
📍 Examples of Prompt-based Context Hints — Decision Maker: “In the era of AI search, where should we invest first—SEO, PR, or Content—to ensure our brand appears in search results?” / Practitioner: “What needs to be modified to rewrite an existing blog post from a GEO perspective?”
3. Ad Material: Clearly presented the 'next step' even on a small page
In ChatGPT, ad images appeared small, roughly 150 to 200 pixels. Unlike traditional display ads, including too much information within an image made it difficult to read. Therefore, we focused on a single core topic, similar to a blog thumbnail, and supplemented the evidence in the text creative and landing page.
For decision-makers, we presented content necessary for comparison and decision-making, such as “Criteria for Selecting GEO Consulting Firms” and “Industry-Specific GEO Secret Notes Analyzed with Data from 548 Companies.” For practitioners, we linked actionable resources, such as the “Essential GEO Diagnostic Checklist for Marketers” and the “Practical GEO Guide: From Website Improvement to External Channel Design.”.
Curious about more detailed ChatGPT advertising insights from Elifunt? Check it out at the 2027 Marketing Seminar: Brand Content for Surviving the AI Era , hosted by Samsung on October 21st
If you are running a campaign for the first time, please refer to the 6 steps of ChatGPT ad setup
1. First, determine the role of one of Reach, Click, or Conversion
We do not set multiple goals for a single campaign
If you expect both awareness gain and consultation conversions simultaneously, it is difficult to interpret which context and creatives were effective. Separate reach, clicks, and conversions, and first define the success criteria for each campaign.
ChatGPT Ads campaign dashboard separating reach, clicks, and conversion goals
2. Differentiating target personas by 'question intent' rather than gender or age
Distinguish between information acquisition questions and problem-solving questions
Practitioners often look for concepts, methods, and checklists. Decision-makers ask about budget, partner selection, performance, and implementation risks. Even with the same GEO topic, the subject matter and landing page must be different because the purpose of the questions differs.
3. Trying to operate keyword-based and prompt-based contexts separately
Do not mix words and sentences in one ad group
Keywords are more likely to capture already materialized interest, while prompts are more likely to capture problem situations that have not yet been compressed into search terms. Separating these two types allows you to compare impressions, clicks, and conversion results and incorporate them into future experiments.
4. Maintain consistency in ad copy and landing page messages
I suggest the 'next step' to the question
For example, a vendor selection checklist is provided for the conversation “How should I choose a GEO vendor?”, and diagnostic items are presented for the conversation “Why isn’t our company showing up in AI?”. Content that naturally suggests what the user should do after the answer is more suitable than hard sales pitches.
5. Simplifying materials based on small ad space
The image presents only one message
Long copy and complex graphs are barely visible on small screens. Design your content to present only the topic and a single key number in the image, while distributing the persuasive power among the headline, sub-copy, and landing page.
Ad creatives that separated text and images by decision-maker and working-level user
6. Verify UTM and GA4 events before execution
We look at media reports and actual site behavior together
The Elifunt team used utm_source=chatgpt-adsas a common setting and CPM, CPC, and CPA. In this case, do not assume that the media will automatically replace the parameter values; you must directly test whether the values are correctly entered into the actual landing URL.
You should not judge performance based solely on media clicks. In GA4, check user count, sessions, dwell time, CTA clicks, and consultation requests as well
ChatGPT Ad Performance: Response patterns varied depending on the goal
| campaign | CTR | CPC trends | Key Interpretation |
|---|---|---|---|
| arrival | 0.53% | Standard level | Secure the widest exposure range |
| Click | 0.68% | Similar to a reach campaign | The quality of behavior after the visit is relatively excellent |
| conversion | 1.01% | Lowest of the three campaigns | Landing inflow efficiency was high, but additional data is needed for conversion determination |
Reach campaigns served to expand exposure, while click campaigns showed a relatively good flow in post-site visit behavior. Conversion campaigns stood out for their landing inflow efficiency, exhibiting the highest CTR and lowest CPC. We confirmed that the optimal response varies depending on the campaign objective.
However, high CTR and low CPC should not be immediately interpreted as business conversion performance. Since actual consultation behavior was confirmed to a limited extent, these results should be viewed as a preliminary verification to identify which contexts and materials drive landing visits.
Media clicks and GA4 behavioral data must be interpreted together
In GA4, conversion campaigns had the largest inflow volume, while click campaigns had relatively high quality of behavior after the visit. This means that campaigns that generated a lot of clicks on a medium may not necessarily coincide with campaigns that created meaningful behavior within the site.
Discrepancies were also found between the clicks in the media report and the GA4 user data. Duplicate clicks, dimensions, and parameter settings can have an impact. In particular, this PoC involved an operational issue stemming from a misunderstanding of the automatic UTM parameter substitution method. For the next test, manual parameter verification per ad group and confirmation of conversion event consistency must be prioritized.
4 Lessons Learned by the Elifunt Team
Lesson 1. Prompts were strong in search demand, while keywords were strong in specific demand
In reach campaigns, prompt ad groups recorded a CTR of 0.55%, while keyword groups recorded 0.37%. In click campaigns as well, the prompt group CTR was 0.76%, which was higher than the keyword group's 0.61%.
Conversely, the click efficiency of conversion campaigns was better for the keyword type. The keyword type had a CTR of 1.13%, which was higher than the prompt type's 0.94%, and the CPC was also relatively lower. However, since confirmed consultation behaviors occurred in the prompt type, click efficiency and conversion potential should not be interpreted as the same thing.
We cannot conclude that one type is always superior based solely on these results. However, we were able to confirm the hypothesis that prompts are relatively advantageous for capturing issues that customers have not yet expressed in short search terms, while keywords are better for capturing already concretized interests, such as GEO companies and SEO consulting
Lesson Learn 2. 'Problem-solving questions' targeting decision-makers led to clicks
In the click campaign, the CTRs for decision-makers and practitioners were similar at 0.69% and 0.68%, respectively. However, the CPC for the decision-maker target was lower than that of the practitioner. In the conversion campaign as well, the CTR for decision-makers was 1.23%, which was higher than that of practitioners at 0.87%.
The demand for information search from working-level employees was favorable for generating exposure volume. On the other hand, questions from decision-makers with clear objectives, such as budget, performance, and vendor selection, led better to ad clicks. In ChatGPT advertising as well, the 'specificity of the problem being solved' is more important than the job title itself.
Lesson Run 3. The CTR was similar to display advertising, but the context of the print space was different
The CTR for this test was reach 0.53%, clicks 0.68%, and conversions 1.01%. These figures are not particularly unfamiliar when compared to the typical CTR range of display ads.
However, the strength of ChatGPT ads was confirmed not so much by the increase in CTR itself, but by the conversational context in which the ads were displayed. Although the ad size was smaller than Meta or Google Display ads, they were displayed at the moment customers specifically explained their problems and conditions. Another difference was that a single corporate ad appeared independently on a screen, rather than multiple ads competing against each other.
In other words, the value of ChatGPT ads lies in learning 'in what conversation context the click occurred' rather than cheap clicks.
It is difficult to run ChatGPT ads using only the syntax of performance advertising
OpenAI ChatGPT ads operate differently from performance ads, which improve efficiency solely through bids, audience segmentation, and retargeting. For an ad to trigger the next action, there must first be a conversational context where the user describes a problem, followed by creatives and a landing page that connect to that context.
Therefore, the starting point of operations is not the Ads Manager screen. You must analyze the language customers use to search and ask questions, classify questions at the problem recognition, alternative comparison, and purchase review stages, and prepare content to answer each question. The responses observed in the ads must then be reflected in improvements to FAQs, blogs, service pages, and sales messages. This is why ChatGPT ads must be designed alongside an organic marketing strategy, rather than being a short-term execution separate from SEO and GEO.
Elifunt has been addressing search data, customer questions, content structure, and landing conversions through SEO and GEO projects. In this PoC, we categorized keywords and prompts by customer roles and funnels, and connected ad creatives with blog and report landings to verify which questions generate responses.
If you are considering ChatGPT ads, check the following three things before deciding whether to run them.
- Have our customers defined the problem situations they will explain in ChatGPT?
- Are the advertising message and grounds for credibility prepared to follow up on that question?
- Does the landing page provide the answer to the question and lead to conversion after the click?
If you want to integrate ChatGPT advertising with your brand content strategy,
Elifunt handles search data-driven question discovery, context hint design, ad creative and landing page planning, and performance analysis. Consult on OpenAI ChatGPT advertising strategy
Frequently Asked Questions
Q. To whom are ChatGPT ads shown?
It is displayed to adult users using the Free and Go plans.
Q. Can advertisers view users' conversations?
No. Advertisers cannot access users' chats, chat history, memory, or personal information. They can only view aggregated, non-identifiable ad performance information.
Q. Which landing page is suitable?
A page that immediately addresses the next step of the question expressed by the user in the conversation is suitable. It is effective to provide immediately usable content, such as comparison criteria, checklists, and reports, and to naturally connect it to consultation and inquiry CTAs.
