ChatGPT Ads Targeting: How Context Hints and Campaign Controls Work
How ChatGPT Ads Targeting Differs From Keywords
Search advertising trains marketers to buy keywords; ChatGPT Ads targeting works from the opposite direction. OpenAI's help documentation states that the ads system selects and delivers ads based on expected relevance and outcomes, weighing the context and intent of the current conversation, the ad's landing page, its title and copy, the advertiser-provided context hints, the campaign's targeting selections, and - when the user has ads personalization enabled - select signals from their broader ChatGPT experience. Ranking combines relevance with advertiser bids. No place in the flow accepts a keyword list, so the quality of your descriptive inputs replaces the quality of your keyword research.
The audience side is equally specific. Ads appear only for users on the Free and Go plans - Plus, Pro, Business, Enterprise, and Edu accounts see none - and never for accounts identified as belonging to users under 18. The test started in the US on February 9, 2026, and availability continues to expand by market. Ads render below the end of a response, clearly labeled as sponsored, and during this test they do not appear in the ChatGPT Atlas browser at all.
What Context Hints Do - and What They Do Not
Context hints live on the ad group and describe what your product offers, who it helps, and when it is useful. OpenAI's guidance is explicit about their limits: hints are not exact-match controls, not audience targeting rules, and not instructions to show your ad only for specific conversations. Being specific in a hint does not mean users need to type those words - hints help the system understand broader needs and natural variations in how people describe what they want. An advertiser who writes hints like keyword match types will read the results wrong; the hint feeds understanding, not a trigger condition.
Writing Context Hints That Give the System Something to Work With
OpenAI illustrates the difference with a running-shoes example: the broad hint "running shoes" gives the system little, while "cushioned everyday running shoes for beginners training for their first 5K" connects features to a real customer situation. Good hints, per the documentation, add information beyond your ad copy and landing page, describe needs or situations rather than broad audience labels, use clear natural phrases instead of disconnected terms, and include only genuine use cases for the product. Structure follows the same logic - each ad group should stay focused on a single product category, theme, or intent area, and hints that would need different ad messaging or landing pages belong in a separate ad group. The practical checklist mirrors the conversion tracking setup habit: every layer of the stack works better when each one carries distinct, honest information.
Geography: Countries, States, DMAs, and ZIP Codes
Every campaign targets countries, and inside the US the controls go down to state, designated market area (DMA), and ZIP code through the location picker. OpenAI publishes the full location catalog as a CSV download, which is the reliable way to confirm a location is targetable before building the campaign around it. A campaign created without location targeting delivers to all available locations - a default worth remembering before the first launch, because broad delivery with no geography rarely matches anyone's budget intent.
Platforms and Surfaces: iOS, Android, and Web
The platform control selects which ChatGPT surfaces can serve the campaign: the iOS app, the Android app, or Web - where Web covers both desktop and mobile web. Reporting does not yet split those surfaces the way the targeting does: insights group devices into Mobile and Desktop, and mobile web currently counts under Mobile. Combined with the placement rules from the ads FAQ - ads below the end of a response, none in the Atlas browser - the platform selection defines the practical inventory your campaign competes for.
Audience Controls: Custom Audiences Include and Exclude
Custom audiences apply at the campaign level as include or exclude selections, letting a campaign address - or deliberately avoid - people who match your uploaded lists. The mechanics of list upload and matching sit outside this article; what matters for targeting is that the include and exclude selections bound which matched users can enter the auction at all, and that the match rate of a list decides how much of it the platform can actually use. On the user side, ads personalization is a control: when disabled, the system loses the broader ChatGPT-experience signals and falls back to conversation context, creative, hints, and campaign targeting.
What Targeting Cannot Do Today
The documented limits line up neatly. Context hints cannot force an ad into a specific conversation, and OpenAI says so verbatim. Geography, platforms, and audiences constrain where ads may serve but do not promise that any particular query or dialog will match. Ads never influence the answers themselves - OpenAI runs ads on separate systems, and advertisers have no ability to shape, rank, or alter responses. The working mental model is a relevance auction over conversation context: you sharpen the descriptions, widen or narrow the campaign constraints, and let ranking combine relevance with your bid.
Point Targeting at Real Measurement
The ChatGPT Ads Manager setup covers where each of these controls lives in the campaign flow, and the conversion tracking guide covers the measurement layer that turns delivery into learnings - the two-source pixel and conversions API model with event_id deduplication. Targeting quality is only observable through that measurement layer: without reliable conversion tracking - server-side above all - the business impact of a context-hint hypothesis cannot be measured accurately. MOST delivers tracked conversions server-side to ChatGPT Ads alongside Meta, TikTok, Reddit, Snapchat, and Pinterest, and the free Pixel Activator validates a single landing at pixel.way2.us.