Context hints, explained
ChatGPT ads have no keywords and no audiences. You describe the conversation instead. Writing that description well is the entire craft.
A context hint is a one or two sentence description, in plain English, of the conversation where your product genuinely belongs. ChatGPT ads have no keywords, demographics or audiences — OpenAI reads your hint alongside the live conversation, your ad copy and your landing page to decide whether to show you, so the hint is the whole targeting layer.
What is a context hint?
It is the sentence you write to tell the auction which moments you want to buy.
OpenAI describes hints as the conversations, topics or situations where your product may be relevant. They are not exact-match rules. They are thematic signals, weighted against everything else the system knows about the moment.
| Google Ads | ChatGPT ads |
|---|---|
| Keywords and match types | Plain-language context hints |
| Audiences, lookalikes, remarketing | None |
| Demographic and device targeting | Country, plus platform (iOS / Android / web) |
| Negative keywords | Separate ad groups, tighter hints |
| Quality Score | Relevance weighting in a second-price auction |
How do you write a good context hint?
Three parts, in this order. Who is talking. What they are trying to do. The detail that narrows it to your buyer.
That last part does most of the work. Without it you are buying a category; with it you are buying a customer.
| Hint | What happens | |
|---|---|---|
| Too broad | "People looking for CRM software." | Matches students, researchers and enterprise buyers you cannot serve. Spend without shape. |
| Too narrow | "Heads of sales at 40-person Series B fintechs in Berlin switching off Salesforce this quarter." | Almost no delivery. You have described a lead, not a conversation. |
| Right | "When a small sales team of five to twenty people is comparing CRMs because their current one is too complex and too expensive." | A real, recurring conversation with a buying trigger in it. |
What makes a hint fail?
- Writing a keyword. "espresso machine" is not a conversation. It carries no situation, no constraint, no trigger.
- Describing yourself. Hints are about the buyer’s moment, not your feature list.
- Stacking unrelated hints in one ad group. If two hints need different landing pages, they need different ad groups. OpenAI says this outright.
- Ignoring the landing page. The system reads it. A hint that promises something your page does not deliver gets weighted down.
- Never retiring anything. Hints have performance. Rank them on cost per customer and cut the bottom.
Paste what you sell. Our generator returns five context hints written in the shape that delivers. No email wall, no signup.
Try the free generatorHow many hints should an ad group have?
Enough to cover the genuinely different ways people arrive at the same need, and no more. In our own account that has usually meant five to nine per ad group.
The test is simple: if two hints would be served by the same ad and the same page, keep them together. If either changes, split them. Hints are cheap to write and expensive to muddle.
How do you know which hints are working?
Reporting is thinner than on Google or Meta, so structure is your instrument. Keep ad groups tight enough that the ad group tells you which hint family paid off.
Then judge on cost per customer, measured server-side. Click-through rate flatters hints that catch attention in conversations that were never going to buy. We have retired high-CTR hints that never produced a sale, and kept quiet ones that did.
Last verified 23 Aug 2026 against OpenAI’s published documentation and our own live account. This channel changes monthly, so we re-check every figure on this page and restamp the date.
Sources: OpenAI help centre, Create Ad Groups for ChatGPT Ads; OpenAI, Ads in ChatGPT: The Basics; and our own live account.