Context hints are short phrases that describe the situation your buyer is in, and ChatGPT uses them, alongside your ad copy and your landing page, to decide whether your ad belongs in a live conversation. They are the single biggest lever on what a campaign costs and whether it converts, and they are the thing almost every advertiser gets wrong on the first try, because the platform asks for something it calls keywords and instinct fills the box with product terms.
How ChatGPT decides which ad to show
Three inputs feed a relevancy score: your landing page, your ad copy, and your context hints. The system reads the conversation a user is having, weighs it against those three signals, and shows at most one ad, below the answer, when the match is strong enough. This is a conversational relevance judgment, not a lookup against a list of exact terms.
That single difference changes everything downstream. On a keyword engine, you win the auction by bidding on the right words. Here, you win by describing the right moment, because the model is matching meaning, not spelling. Relevance is also priced in: a closer match costs less per placement, so getting hints right is how you pay less for the same slot.
The keyword habit that wastes budget
Twenty years of Google trained everyone to think in product terms. The field says keywords, so people type what they sell: project management software, running shoes, tax advice. Those describe your catalogue. They do not describe the situation a person is in the moment before they need you, which is what the system is actually trying to match.
The result is a campaign that either barely serves, because your product terms rarely match a natural conversation, or serves against the wrong moments and burns budget on people who were never close to buying. Both failures trace back to the same root: describing the product instead of the person.
Describe the situation, not the product
The method is to write the moment your best customer is stuck in, in plain language, in about four or five words. Not what you sell, and not the words you would use internally, but the words that describe their problem as they would feel it. Think about the sentence they would say to a friend, not the category on your pricing page.
The test for a good hint is simple: could you picture the exact conversation where it belongs? “Running shoes” could sit under a thousand unrelated chats. “Knee pain on long runs” belongs to one specific moment, and that specificity is what earns a cheap, relevant placement.
Weak keyword vs strong hint
The pattern is the same across every industry: move from the noun you sell to the situation your buyer is living. A few worked examples:
| Weak keyword | Strong context hint |
|---|---|
| Project management software | Team losing track of deliverables |
| Running shoes | Knee pain on long runs |
| Accounting services | Behind on VAT before the deadline |
| CRM platform | Leads slipping through the cracks |
| Meal delivery | No time to cook after work |
| Language app | Moving abroad in three months |
Read the right-hand column and you can hear the conversation each one belongs to. That is the bar. If a hint could sit under dozens of unrelated chats, it is a keyword wearing a disguise, and it will cost you.
How many hints, and how to test them
Hints live at the ad-group level, so every ad in a group shares the same situation targeting. Group ads by intent, not by product, and give each group a tight set of hints that all point at the same moment. A handful of sharp, related phrases beats a long list of loose ones, because breadth dilutes relevance and relevance is what the auction rewards.
Treat hints as testable. Launch with distinct situations across your groups, watch which ones actually draw relevant placements, and reallocate toward the moments that convert. This is exactly why a real launch spreads across many ads and intent clusters: you are not guessing the winning situation up front, you are giving the system enough distinct moments to find it. When the budget is there, we do this research and build for you, because it is the same intent work our organic AI-visibility clients already pay for.
