A context hint is a natural-language sentence describing a conversation in which your ad would be useful. You attach it to an ad group in OpenAI’s Ads Manager, and it helps the system match a user’s conversation to your ads. Neither a keyword nor a targeting rule: one signal among several.
Updated 25 October 2026.
What exactly is a context hint?
A one-sentence description of the kind of conversation you are after, which OpenAI uses to decide whether an ad group is a candidate to be shown.
OpenAI’s documentation keeps returning to the same point: a hint describes a situation, not a query. Someone talking to ChatGPT does not type three words. They explain the problem, lay out their constraints and ask a whole question, so your hint should read like that exchange rather than like a line from a Google Ads account.
Nobody ever sees it, not the user and not the person scrolling past your ad, because it exists only for the selection system and has no place on the screen.
Ad group level: where hints live
Hints are entered as a JSON array, ad group by ad group, and every ad in the group inherits them.
The campaign sets the objective, budget and countries. The ad group carries the hints and the bid, while the ad carries the title, body, image and link. Because every ad in the group shares the same hints, an ad group should cover a single intent family, or the ad ends up talking to everyone and convincing no one.
OpenAI documents no maximum number of hints per ad group. In practice the count matters little: three hints describing the same situation from different angles beat a long list that points in every direction.
What a hint won’t guarantee
A hint is not an exact keyword, a targeting rule or a delivery instruction. It promises nothing. It forbids nothing, either.
| What people assume | What is documented |
|---|---|
| “The hint fires the ad as soon as the word appears” | The hint is one signal among others: title, body, landing page and the full context of the conversation also count |
| “I can target a city or a profile with a hint” | Country is set at campaign level. A hint such as “only show in London” is not an instruction the system executes |
| “A short, generic hint covers more ground” | A short, generic hint describes the situation poorly, so it matches poorly |
| “I need to list every synonym” | A natural sentence describing the situation replaces the synonym list |
The gap with search runs deeper than it looks. On Google Ads the keyword works as a contract (this query, this bid), whereas on ChatGPT Ads a hint stays a description the system may or may not match to a conversation. It steers, and that is all it does.
What format makes a good hint?
One sentence, one idea, three questions: who is talking, in what situation, with what need. Call it the who/when/what trio.
| Component | Question | Example fragment |
|---|---|---|
| Who | What profile is holding the conversation? | “fleet manager”, “freelancer in their first year” |
| When | What situation are they in? | “comparing tools before switching”, “planning a weekend away” |
| What | What need, product or constraint? | “cut the fuel budget”, “with no commission per booking” |
You do not need all three each time, and two often do the job. The floor does not move, though: a situation or a need, in everyday words. “Finance director” on its own is a label, and a label describes no conversation at all. Add the situation.
Write hints in the language of the user you are targeting, which means English for a UK campaign.
Before and after: five hints rewritten
A good hint reads like the summary of a real conversation, a bad one like a spreadsheet row.
| Bad hint | Good hint | What changes |
|---|---|---|
| “fuel, fleet, fuel card” | “fleet manager asking how to cut the fuel budget of a company vehicle fleet” | A list becomes a situation with a goal |
| “cheap invoicing software” | “freelancer comparing simple invoicing tools for their first year in business” | The profile and the moment replace the adjective “cheap” |
| “mattress, bedding, deal” | “someone asking how to choose a firm mattress for a guest room” | The actual need appears: a precise use, not a category |
| “lisbon weekend” | “traveller planning a long weekend in Lisbon on a fixed budget” | The budget constraint qualifies the conversation |
| “only show to restaurant owners in London” | “restaurant owner looking for booking software with no commission per cover” | The delivery instruction goes, the need stays |
These examples are generic and invented for illustration. Take yours from sales calls, support tickets and customer reviews, or from old Google Ads search terms that you rewrite as situations.
How do you group hints by intent family?
One ad group per intent family, each family gathering the conversations that call for the same promise and the same landing page.
The ones that come up most are the problem (“how do I fix…”), the comparison (“which tool should I pick for…”), the alternative (“replace… with…”), and then price, implementation and urgency. A single product can therefore yield several ad groups, each with its own hints, ad and landing page.
The method comes down to four moves. Collect your customers’ real phrasings first, sort them by family, write hints for each family that describe the situation from several angles, then pair every family with an ad and a page that answer exactly that situation.
One test settles most doubts. If two hints in the same ad group would send people to two different landing pages, they belong in two ad groups, however similar the wording looks.
The mistakes that keep coming back
Three dominate: pasting a keyword list, writing hints that run too long, and mixing two products in one ad group.
The first belongs to the advertiser arriving from Google Ads. They export their search terms and paste them as they are, so the system receives words with no situation and matches at random.
The opposite mistake is the bloated hint: three ideas in one sentence, two profiles, four constraints, and the system can no longer tell which situation it is being shown. One idea per hint, as many hints as ideas.
Mixing products breaks the hint, ad, page chain. An ad group that talks about invoicing and payroll together can write neither a precise ad nor a precise page, the relevance score drops, and the auction position follows. Costly, and avoidable.
A quieter one remains: writing hints in English for a French audience. The system matches against real conversations, and those happen in the user’s language.
Even well-written hints steer matching without guaranteeing it, and we do not claim otherwise. For a first set built from your product page, our context hints generator applies the format described here; read the output before it goes live, as you would with any draft.