What Is ChatGPT Advertising, How Does It Work? An End-to-End Guide from Setup to Measurement

Most of what is written about ChatGPT advertising swings to one of two extremes: either "a revolution, everyone run" or "a bubble, stay away." Neither is useful. This guide tries a third path: a manual distilled from the channel's official documentation and from real advertiser experience in its first months, neither hyping nor dismissing. It was written the week access opened in Turkey and tested against what we saw in our own account.

What Is ChatGPT Advertising?

OpenAI's ad product is a separate unit that appears below the response in ChatGPT conversations: it carries the advertiser's name and logo, a headline, a short description, an image, and a landing page link; it is labeled "Sponsored" and is visually separated from the answer. There is one ad slot per response, and a single unit can display more than one product.

The most critical design decision is this: the ad system is separate from the model that produces the answer. OpenAI calls this "answer independence" in its official principles — buying ads will not make ChatGPT recommend you, and it will not keep you from being recommended either. That separation produces the strategic consequence we return to in the GEO section below.

Who Sees the Ads?

Ads are shown only to signed-in users aged 18 and over on the Free and Go plans; Plus, Pro, Business, and education plans are ad-free. There are no ads in temporary chats. Ads are not served near sensitive topics such as health, mental health, and politics, and political advertising is not accepted at all. Users have controls of their own as well: they can turn off personalization, hide an ad, and have their advertising data deleted.

In Turkey specifically, serving began on September 17, 2026. Given the size of the free user base, the volume of this inventory in Turkey is nothing to dismiss.

How It Works: No Keywords, Just Context

The first question everyone arriving from Google Ads asks is: "Which keyword do I bid on?" The answer is none. In ChatGPT advertising, the model does the matching. At the ad group level you write free-text context hints that describe your product or the intent you want to capture — something like "trail running shoes for rocky terrain." The system weighs those hints together with your ad copy, your landing page, and the context of the conversation at that moment, and the impression decision comes out of a relevance-weighted second-price auction. A high bid will not rescue an irrelevant ad.

What that means in practice: you hand part of the control to the model. There is no exact match, no negative keyword list, no dayparting, no device-level bid adjustment. Early advertiser experience points the same way: 5-10 hints clustered around a single intent per ad group work better than long, scattered lists.

Ad Formats

• Standard card: brand name + logo, headline, description, image, landing page. Image is a minimum of 640×640 pixels.
• Product cards: a format fed by your e-commerce catalog that can display price and rating; you upload a product feed and push price and stock updates through a delta feed.
• Carousel: a swipeable multi-product format.

Video is not supported yet. Once created, ads enter automatic review, and approval usually completes within 24 hours. In the refreshed setup flow that arrived at the end of August, you can also enter your site address and let the model suggest creative and copy — we would recommend editing that output rather than using it as is.

How Do You Open an Account?

The panel lives at ads.openai.com and is in beta. Opening an account takes five steps: registration with an OpenAI account tied to the business; identity and business verification (which can take a few business days); account name and logo; billing profile and payment card; team invitations. Legal business name, address, tax number, and website are required.

Watch out for two traps:

• Country, currency, and time zone cannot be changed afterward. Do not rush the first setup screen. TRY billing is not available for Turkey right now; accounts run in a foreign currency, typically USD.
• An agency cannot open an account on a client's behalf. There is no manager account structure like Google's MCC. The correct flow: the brand opens and verifies the account under its own legal entity, then invites the agency team with an appropriate role. This is exactly how we work with our clients — the brand owns the account, we own the management.

How Do You Build a Campaign?

The structure is familiar: Account → Campaign → Ad Group → Ad. Objective, budget, schedule, and geo and platform targeting are defined at the campaign level; bid, context hints, and product selection at the ad group level; creative and destination URL at the ad level.

There are three campaign objectives: Impressions (CPM), Clicks (CPC), and Conversions (conversion-optimized, paid per click). The minimum daily budget on dollar accounts is $25 per campaign. OpenAI suggests a maximum CPC of $3-5 for beginners; that number is an opening setting, not a performance benchmark. Third-party measurements report actual CPCs in the $3-5 range for e-commerce and in the $8-18 band for competitive verticals such as software and finance — in other words, this is not a "cheap clicks" channel.

The setup order we recommend: install conversion measurement first (see the next section), then draw the structure on a one-intent-equals-one-ad-group principle, and use exactly the same language in the creative as on the landing page. Because the model does the matching, consistency among ad copy, landing page, and context hints does more work here than it does on classic channels.

What Does Targeting Offer?

• Geography: country and, except for product feed campaigns, city and region level.
• Platform: desktop web, iOS, and Android.
• Custom audiences: SHA-256 hashed email and phone lists; at least 25,000 matched users are required before a list can be used for targeting. Small CRM lists run into this threshold.
• Negative targeting (excluding unwanted contexts) is in testing; classic remarketing is not available at launch.

The privacy line is clear: advertisers do not receive users' conversations, identities, or personal data — you only see aggregated performance data.

Conversion Measurement: From the Pixel to the GA4 Reality

There are two first-party routes to measurement. The first is the OpenAI measurement pixel added to your site; you create it under Tools → Conversions in the panel and install it with JavaScript. The second is the server-to-server Conversions API; it supports ready-made event types (lead_created, order_created, checkout_started, and so on) and deduplicates with a shared event ID so the pixel and the API do not double count. Attribution has two layers: the post-click window is configurable and the post-view window is fixed at 1 day; when the two overlap, the click takes priority.

On the GA4 side, set expectations from the start: there is no direct integration. GA4 only does passive tracking through UTM parameters you add to landing URLs (utm_source=chatgpt, for example), and conversions seen in GA4 are not fed back into OpenAI's optimization algorithm — only pixel and Conversions API data feed the algorithm. Two more practical warnings: panel reporting can run with a lag of up to 7 hours, so do not panic on day one; and early on, multiple advertisers documented serious gaps between the click counts in the panel and their own analytics. Do not treat panel data as the single source of truth, and make a UTM-based comparison part of the setup.

How Do You Get Results? Field Notes

• Write the success threshold before you spend. Have the answer to "at what cost per lead do we keep going?" settled before the campaign starts; the panel will not carry you to a decision — bring the decision to the panel.
• One intent, one ad group. Cluster your context hints around a single product or intent; groups that try to say everything fit no conversation properly.
• The landing page is part of the matching. The model reads your page when it matches; a disconnect between ad and page costs more here than it does on classic channels.
• If you are in a research-heavy category, you are ahead. Early impression distribution favors B2B software, security, comparison and review, insurance and finance, and education; impulse-purchase products fit this screen less easily.
• Do not judge it on last click. A user can discover you in ChatGPT and complete the purchase later on Google or on your site; read the channel alongside its incremental effect, such as branded search and shifts in direct traffic.
• Give it a 2-4 week read period. Low CTR is the nature of this screen, since there is one ad per response; your decision metric should be qualified cost per conversion, not click-through rate.
• Ask about credits. New advertisers may be offered periodic matched spend credits; check the offers that land in the panel when you plan your test budget.

An Honest Risk Map

When we recommend this channel, we say the following in the same breath. OpenAI has not published any official performance benchmarks yet; every CTR, CPC, and ROAS figure in circulation is third-party measurement or a single advertiser's claim. In early measurements CTRs sit far below classic search; costs are not cheap; one independent study reported that a share of placements turned out to be unrelated to the conversation topic; and cases have been published where audience accuracy stayed weak in B2B. The counting gaps between the panel and analytics, together with the uncertainty of attribution methodology, mean nobody can write the sentence "proven ROI" for now. In short: this is not a core budget line. It is a test line with boundaries drawn and measurement installed. The early-mover advantage is real — but the advantage lies in the learning, not in the budget.

Are ChatGPT Advertising and GEO the Same Thing?

They are not — and every strategy that confuses the two is born broken. GEO (generative engine optimization) is the work of getting ChatGPT, Gemini, and Perplexity to recommend your brand inside the answer: it is earned with content, evidence, technical structure, and authority, and it cannot be bought with ad budget. ChatGPT advertising is the labeled, rented space below the answer; visibility ends the day the budget stops. One is an asset, the other is rent. In competitive categories, the right setup runs both together: fast visibility and data from advertising, a lasting place from GEO. We explained how the organic side is won in how your brand shows up in ChatGPT, and you can measure your brand's current AI visibility with our free AI Visibility Scorecard.

Frequently Asked Questions About ChatGPT Ads

Conclusion

ChatGPT advertising is a new auction with rules of its own: context has taken the place of keywords, invitations the place of the MCC, and the obligation to produce your own data the place of ready-made benchmarks. Access from Turkey is open, the panel works, and we are completing setups in our own account and for the brands we manage. The story of the announcement and what we built this week is in a separate post: ChatGPT Ads Go Live in Turkey — We're In From Day One.