ChatGPT Ads Context Hints and Targeting (No Exact-Match Keywords)

Table of contents

ChatGPT Ads do not give you exact-match keywords. If your media team is still building a 400-row keyword sheet for OpenAI, they are building the wrong object. Targeting is a stack: geography, platform, custom audiences, and context hints on the ad group. Hints describe the buying moment in ordinary language. They do not pin an auction to a string.

That is good news if you sell a Bangkok service people describe in ten different phrasings. It is bad news if you wanted the comfort of “match type = exact.” This article is the targeting brief we use before anyone hits publish. The productised version sits on ChatGPT Ads services.

The stack, in the order it actually constrains you

OpenAI’s targeting docs are consistent on one point: matching one setting does not override the others. A perfect hint will not save you if geography excludes the user. A Bangkok geo will not save you if the hint describes a product you do not sell.

  1. Geography — country and supported location IDs (Thailand, Bangkok, or tighter where the catalogue allows).
  2. Platform — ChatGPT iOS app, Android app, and web (desktop and mobile web). Web is not “desktop only.”
  3. Custom audiences — include, exclude, or bid-multiply people you already know, once the matched-user floor is met.
  4. Context hints — extra description of products, use cases, or needs the creative and landing page may not fully cover.

Exclusions win when someone is in both an include and an exclude list. Do not put the same audience in both.

Context hints are not keywords

OpenAI’s wording: hints are not exact-match keywords, and they do not replace geographic, platform, or audience targeting. An ad group can hold a large list (the API allows up to 2,000). You do not need 2,000. You need accurate sentences.

Their own contrast is the right teaching example: “Shoes” is a weak hint. “Lightweight trail running shoes for rocky terrain” is a useful one. “California customers” is the wrong tool — that belongs in geo targeting, not in a hint.

For a Bangkok company, weak vs useful looks like this:

  • Weak: “SEO”, “marketing”, “Bangkok”, “company”
  • Useful: “Owners of Thai factories comparing a Bangkok SEO retainer that includes AEO and GEO, not just blog posts”
  • Useful: “Clinic managers in Bangkok asking how to get more Google and ChatGPT booking enquiries this quarter”
  • Wrong place: “people in Wattana” — that is geography, if the location catalogue supports it

Write hints as 1–2 sentences of who is trying to do what, with what constraint. Then let the ad title, the landing page, and the bid do the rest. Hints steer relevance. They are not a promise that a particular chat will show the ad.

Updating the hint list replaces the list. If you add one hint in the API, send the full intended set. Empty array clears them. Omit the field to leave them alone.

Bangkok and Thailand geography

Set geography on the campaign. Country codes are ISO alpha-2 (TH for Thailand) but “valid ISO” is not the same as “available in your account.” Use OpenAI’s geographic lookup / geotarget catalogue for city-level IDs. Do not type “Bangkok” into a hint and call it geo.

A pattern that makes sense for most of our clients:

  • National Thai campaigns when the offer can be delivered anywhere in the country (ecommerce, remote retainers).
  • Bangkok (and surrounding provinces if the catalogue has them) when the service is on-the-ground: clinics, showrooms, installation, agencies that sell face-to-face.
  • Exclude regions you cannot serve, rather than hoping the hint will do it.

Product-feed campaigns, where relevant, are stricter: country-level geo only. If you are not running a catalogue, ignore that constraint.

Read the campaign back after every geo edit. Geographic and audience updates can replace what you already set. The number of times a team “adds Bangkok” and silently drops an exclusion is not zero.

Custom audiences: inclusion is not a bid bump

Custom audiences match emails, phones (E.164), hashed variants, and GAID to logged-in ChatGPT users. Files are UTF-8 CSV/TXT, large but not infinite. Processing is async. Uploaded rows ≠ matched users.

Two different tools, often confused:

  • Inclusion — only those people can see the campaign (plus your other targeting). Floor: typically 25,000 matched users for inclusion and for bid multipliers.
  • Bid multiplier — those people can be bid up or down (OpenAI documents a wide multiplier range) while non-members remain eligible.
  • Exclusion — suppress converters, staff, or junk. No matched-user minimum on a ready exclusion audience.

Most Thai SMEs do not have 25,000 matched ChatGPT users sitting in a CRM. That does not block you from running ads. It blocks you from restricting delivery to that list. Use exclusion for staff and recent converters if the list is small. Use geo + hints for prospecting. Do not upload a 800-row Excel and expect a remarketing wall.

EEA/Switzerland have extra audience limits in OpenAI’s docs; if you ever run those countries, read the current restriction before copying a Thai setup.

Platforms, and what “mobile” means here

You can include ios_app, android_app, and web. Omit them at creation and you get all three. A phone using ChatGPT in Chrome is web, not the iOS app. If your landing page is broken in in-app browsers, fix the page — or you will blame “targeting.”

How we brief a first Thai campaign

  1. Lock the legal entity and Ads Manager account — see can Thai businesses run ChatGPT Ads in 2026.
  2. Pick geo: Thailand vs Bangkok, with exclusions you mean.
  3. Write 5–15 hints that sound like real buyer chats, not a keyword dump.
  4. Ship separate EN and TH ads if you sell both languages.
  5. Put measurement on before scale. A hint that “works” without a conversion event is a story.

ChatGPT Ads are still a labelled card under an answer. They do not replace organic AEO/GEO, and they do not replace Google query coverage. They are the paid layer for conversational research. If you want that layer built under your company, not rented from an agency login, start at ChatGPT Ads for Thailand.

Frequently asked questions

Can I use exact-match keywords in ChatGPT Ads?

No. Context hints are not keywords. They describe the buying moment. Geography, platform, and audiences are set separately.

Enough accurate sentences, not a dump. Five to fifteen well-written hints beat hundreds of generic tokens. The API allows a large list; filling it is not a strategy.

Use campaign geography and OpenAI’s location catalogue, not a hint that says Bangkok. Confirm the location ID is available on your account.

Inclusion and bid multipliers need a much larger matched-user floor (typically 25,000). A small list is still useful for exclusions. Prospect with geo plus hints.

No. Hints help the auction understand the offer. The click still needs a page and measurement. That is why we sell setup, tracking, and pages together.