Context hints vs keywords: what changes in ChatGPT Ads
ChatGPT Ads don't use keywords. Context hints describe the situations your ads are for. Here's how the two differ, and how to turn a keyword list into hints.
If you've run search ads, the first thing you'll look for in ChatGPT's Ads Manager is a keyword list. There isn't one. Targeting in ChatGPT Ads is built around context hints: short descriptions, set on each ad group, of the needs and situations its ads are for.
That sounds like a small change of format. In practice it's a different way of thinking about who you're reaching, and why.
Keywords match words. Context hints describe situations.
A keyword says “show my ad when someone types this”. A context hint says “my ad is useful to someone in this situation”. OpenAI describes hints as a guide for matching rather than exact-match controls: they don't guarantee that an ad appears for particular words, and people don't need to use a hint's exact words for it to match.
| Keywords in search ads | Context hints in ChatGPT Ads | |
|---|---|---|
| What you write | Words and phrases people type | Short descriptions of a need, situation or topic |
| How they match | Match types decide how close a search must be | A guide for matching, with no exact match and no guarantee |
| What people are doing | Searching, often close to a decision | Talking a problem through, often still comparing |
| What you learn | A search terms report shows what people typed | Results by ad group and country; the conversations stay private |
| How you organise | Ad groups of related keywords | One ad group per conversation theme |
The last two rows matter most. Because you can't see the conversations behind your results, the ad group is the only lens you have on which conversations work. Put three themes in one ad group and you'll never know which one drove the results.
Why a pasted keyword list underperforms
Keywords are compressed: “sofa bed”, “sofa bed small space”, “best sofa bed”. People in ChatGPT don't compress. They write something like “I live in a studio flat and my parents visit twice a year, what can I do for a guest bed?” A hint that describes that situation, “renter in a studio flat who needs a bed for occasional guests”, says far more about whether your ad will help than the keyword does.
OpenAI's own guidance uses the same idea, turning a broad hint like “running shoes” into “cushioned everyday running shoes for beginners training for their first 5K”.
Short, keyword-like hints aren't useless, though. One early test reported by MarTech found a keyword-style ad group reached more conversations than one written as questions. Treat the style of your hints as something to test, in separate ad groups, rather than a fixed rule.
How to turn a keyword list into context hints
- Group your keywords by intent: what the person is trying to get done, not which words they share.
- For each group, write down the situation behind it: who the person is, what they're trying to do, and the constraint that makes your product a good answer, such as space, budget, timing or skill.
- Turn each situation into a short hint, written the way the person would describe it, in the market's own language.
- Give each theme its own ad group, with ads that answer that situation directly.
- Leave out keywords that only make sense as a quick search, such as a competitor's login page or your own brand name on its own.
What stays the same
The fundamentals carry over: ads that match what the person needs, a landing page that continues the same message, several ads per ad group to test, and budget that follows results. What changes is the unit you plan around. In search it's the query. In ChatGPT it's the conversation.
For the rules we use when writing hints, read Writing context hints that match real conversations. For ready-made patterns, see context hint examples for eight industries.