ChatGPT ads capabilities: the targeting, formats, and measurement available now, and what OpenAI has not built yet

ChatGPT ads, as of October 2026, provide a focused but narrower capability set than mature platforms like Google or Meta, with context-driven targeting, a small set of formats, and a modern but opinionated measurement stack; setup, eligibility, and pricing live in sibling references such as the ChatGPT ads overview and pricing detail.

Targeting capabilities in ChatGPT ads

Targeting in ChatGPT ads is built around conversational context, geography, platform, and first-party audience lists, without keyword bidding. Per OpenAI’s documentation, the core control closest to intent is the ad-group level "context hint," which guides where an ad is eligible to appear in conversations but does not function as an exact-match or biddable keyword list.

Context hints are often misunderstood as ChatGPT’s version of search keywords. That is incorrect: context hints provide matching guidance to the system, there is no keyword bidding, and there is no guarantee that a given phrase will trigger an impression in a specific conversation, per OpenAI’s basics guide. This means you can describe the types of prompts and topics you want to align with, but you cannot engineer precise query-level coverage.

Geo targeting supports countries and sub-national regions, including states or regions, cities, markets, and postal codes where available, according to the campaigns documentation. Product-feed campaigns are an exception and are limited to country-level targeting, so any sub-national strategy for retail feeds must be handled through separate campaigns per country rather than finer-grained locations.

Platform targeting allows you to choose among 5 surfaces: Android app, Android web, desktop web, iOS app, and iOS web, as documented in the campaign creation help. This gives some control over device and environment mix, which matters if your landing experience or conversion behavior differs between mobile apps, mobile web, and desktop.

Custom audiences are supported for both inclusion and exclusion, with a documented minimum of 25,000 or more matched users required for inclusion audiences. Exclusion audiences can be smaller, and audience membership can be updated without recreating the audience object, which reduces operational overhead for list refreshes. OpenAI advises avoiding custom audiences in the EEA and Switzerland while personalization is unavailable in those regions.

Per the basics documentation, ChatGPT uses ads personalization signals from a user’s broader ChatGPT experience only when that user has personalization enabled. This means the platform can draw on more than just the current prompt for targeting and optimization, but only for a subset of users who have opted into personalization, which affects how broadly those signals apply.

Ad formats and current betas

ChatGPT ads currently offer a standard ad card format and two feed-based betas for commerce and travel. The live, default format is a card that includes the advertiser name, logo, headline, body copy, landing page URL, and an image, and it appears below ChatGPT’s responses, per OpenAI’s basics article.

For retail advertisers, product feed campaigns are in beta and use a catalog provided via CSV, hosted URL, or SFTP, as documented in the product feeds help. The resulting product ads can show images, titles, star ratings, prices, sale prices, and brand information, and during this beta these feed-driven units are ads-only and do not surface in organic ChatGPT answers.

There is also a limited beta for hotel feeds, described in the OpenAI Ads developer documentation. These hotel feeds are a separate documented beta format, oriented around travel inventory, and are managed through the same Advertiser API that covers product feeds and other campaign operations.

Because feed products are ads-only during beta, the common belief that a product feed gets items into organic ChatGPT answers is incorrect. Per OpenAI’s product feeds documentation, feed items are used solely for ad delivery in this phase, so any organic presence must come from separate content or integrations, not from the ad feed itself.

Measurement and reporting capabilities

Measurement for ChatGPT ads combines a standard reporting UI with a modern conversion stack that parallels other large platforms, but with some documented differences. Reporting metrics include impressions, clicks, spend, click-through rate (CTR), average cost per click (CPC), average cost per thousand impressions (CPM), and conversions, and per OpenAI’s basics guide these metrics refresh roughly every 15 minutes while spend can lag by 7–8 hours.

The conversion measurement stack consists of a JavaScript Measurement Pixel, a server-side Conversions API, and an Image Tag, as described in the developer documentation. There are 13 standard event types plus support for custom events, and browser and server events can be deduplicated using shared event IDs, which aligns with how other major platforms handle cross-channel or cross-surface conversion tracking.

Attribution includes click-through conversions on a configurable click window and view-through conversions on a fixed 1-day post-impression window where available for the account. Per the Measurement Pixel documentation, view-through conversions are reported separately and are not included in the primary Conversions metric, and bidding and billing remain click-based, which affects how you interpret reported performance versus what the system optimizes toward.

UTM parameters persist on ad clicks, according to the basics article, so you can track ChatGPT traffic in third-party analytics tools that read URL parameters. There is also an Advertiser API that supports programmatic campaign management, bidding, targeting, product feeds, conversion tracking, and reporting, as documented in the OpenAI Ads developer docs.

This leads to a nuanced correction of a common belief that you can track ChatGPT ads with full parity to Meta. The stack is similar in structure, with a pixel, server-side API, and deduplication, but OpenAI documents a fixed metric set in the UI, a 7–8 hour potential lag on spend, and view-through conversions that sit outside the main Conversions metric, so reporting behavior is not identical to Meta’s.

Implementation specifics for the Measurement Pixel, Conversions API, and event mapping are covered in a separate reference on ChatGPT ads conversion tracking, which focuses on technical setup rather than capability scope.

Campaign controls and brand protections

Campaign and brand controls in ChatGPT ads focus on placement guidance, bidding support, and account governance, with fewer knobs than long-standing platforms. OpenAI’s basics documentation describes "Negative Phrases" for qualifying advertisers, which support narrow brand-safety placement requirements by signaling contexts where you do not want your ads to appear.

Bid-strength guidance is available in Ads Manager, per the basics article, giving you feedback on whether your bids are likely to be competitive. This is a softer control than explicit bid simulators but still informs pacing and volume expectations when you adjust bids.

Account settings cover members, permissions, API keys, billing, and change logs, according to the Ads Manager beta overview. This supports standard governance patterns, such as role-based access and audit trails for changes, which matter when multiple teams or agencies touch the same account.

Bulk campaign creation via CSV upload exists and is documented in the Ads Manager overview, but it does not support product-feed campaigns. That limitation means large-scale feed-based retail or hotel programs must be managed through the feed and campaign interfaces or via the Advertiser API rather than through CSV uploads.

Documented capability gaps and limits

OpenAI’s documentation also describes several explicit absences or limits that shape how you evaluate ChatGPT ads against other channels. First, there is no keyword bidding, and context hints are the only query-adjacent control, per the basics guide, which confirms that you cannot run traditional keyword campaigns and must instead rely on contextual guidance and the system’s understanding of conversations.

Campaign objectives cannot be changed after creation, and some budget-type changes are irreversible, according to the campaign creation help. This constraint makes upfront planning more important, since you cannot repurpose an existing campaign to a different objective or fully reconfigure certain budget structures once they are set.

The conversion stack has a specific gap for mobile app events. Per the supported events and Conversions API documentation, the Measurement Pixel does not support app_installed or app_opened events, which are available only through the Conversions API, and there is no native mobile SDK setup documented at this time, so app-heavy advertisers must plan on server-side integration for these events.

OpenAI’s basics article states that ads reach only Free and Go plan users, and that paid-tier accounts and users under 18 are excluded from ad delivery. This corrects the misconception that ChatGPT ads can reach the entire user base, since a meaningful portion of users are explicitly outside the ads inventory.

For product feeds, the product feeds documentation clarifies that feed products are ads-only during beta and do not appear in organic conversations. This directly contradicts the belief that participating in the product feed beta will seed organic exposure, and it reinforces that the feed is an advertising surface only in the current phase.

Who should test ChatGPT ads based on capabilities

Given the current capabilities, ChatGPT ads are best suited for advertisers who can work within contextual targeting, have web-based conversion paths, and can tolerate some reporting lag. The combination of conversational context hints, geo and platform targeting, and optional custom audiences supports intent-adjacent targeting, but not the precise query control of keyword bidding, so brands that depend on exact-match search strategies will find a different fit here.

The measurement stack, with its pixel, Conversions API, and 1-day view-through reporting, favors teams that already run modern server-side tracking and can interpret click-based billing alongside separate view-through metrics. Advertisers with strong web analytics, clear UTM strategies, and the ability to reconcile a 7–8 hour spend lag will get more value from early testing.

Retail and travel advertisers who can participate in the product feed and hotel feed betas have an additional reason to test, since these formats are purpose-built for catalogs and inventory, even though feed items are ads-only during beta. For a broader view of where these capabilities fit into channel mix and use cases, see the separate reference on best uses of ChatGPT ads, which focuses on strategic fit rather than feature inventory.