Context hints: how to target advertising in ChatGPT without keywords
Context hints are short descriptions of the conversations, topics or needs in which your product or service is useful. They are set at ad group level in the OpenAI Ads Manager and replace keywords as the main targeting mechanism for advertising in ChatGPT. According to OpenAI’s documentation, hints are not exact-match keywords, are not audience rules and do not guarantee display in a specific conversation — they help the system understand broader needs and situations. A good hint describes what the product offers, whom it helps and when it is useful: not “running shoes”, but “soft everyday running shoes for beginners training for their first 5 km”. Each ad group should cover one topic or intent, with several ads inside, and different products or use cases go into separate groups. Below is the method I use to write hints for Bulgarian businesses.
I am Hristiyan Panov — a Performance Marketing consultant with over 5 years of experience and over €2,000,000 in advertising budget managed for over 90 companies in Google and Meta. Below is a breakdown of context hints based on OpenAI’s official documentation and the practice of the first advertisers.
What are context hints and how do they differ from keywords?
In Google Ads you target by the words a person types into the search engine: short, transactional, with match types. In ChatGPT a person describes their situation in sentences: what they want to achieve, what their constraints are, what they have already tried. OpenAI’s ad system understands the intent of the conversation semantically and compares it with the descriptions the advertiser has submitted. That is why a hint is a description of a scenario, not a list of words.
OpenAI states explicitly: hints are not an exact-match control, are not audience targeting rules and are not an instruction for the ad to appear only in certain conversations. People do not have to use the same words — the system recognises natural variations of the same need. This is also the reason a good hint should be specific, but not narrow.
How does OpenAI recommend writing hints?
The documentation gives three questions every hint must answer: what else the system needs to know about the product, whom it helps and when it is useful. Good hints add information beyond the headline, the text and the landing page; connect product features to real needs; describe situations, not just audience labels; use clear, natural phrases instead of fragmented terms; and include only genuine, relevant use cases.
The practical guidance of the first agencies is consistent: hints are written as a buyer’s real questions, not as a list of keywords; each ad group covers one product or one topic; hints are specific enough to reach conversations with purchase intent. According to published breakdowns of the Ads Manager structure, the hints field of an ad group has a limit of 280 characters, and the recommended volume is between five and fifteen hints per group — these two figures are not in the official documentation, so treat them as a guide, not a rule.
What does a good hint look like for a Bulgarian business?
Take an online tool store. A weak hint: “socket sets, ratchets, tools”. Strong hints, split by intent: “someone who repairs their own car and needs a socket set with a ratchet and extensions” · “someone who is equipping a home garage and comparing complete tool sets” · “someone who lacks the exact socket size or extension in the middle of a repair” · “someone looking for a useful gift for a relative who repairs cars”. Each of these four is a separate ad group with its own ads and its own category page.
Take a travel agency. A weak hint: “holiday, excursion, offers”. Strong ones: “a family from Bulgaria planning 4–5 days in October with a direct flight and a hotel on the beach” · “a couple comparing a city weekend in Europe with a budget under a certain sum per person” · “someone looking for an organised guided tour for a first visit to a distant destination”. Note the structure: who, in what situation, with what constraint.
How are ad groups structured around the hints?
OpenAI recommends that each ad group be focused on one product category, topic or area of intent; separate groups for meaningfully different audiences or use cases; no mixing of unrelated products in one group; and several ads in each group, to test different messages. If the hints in one group describe products that require different ads or different landing pages, that is a signal to split.
The working structure for a first test is one campaign per offer, three to six ad groups by intent, three to five different ads in each group and one suitable page for each group. The headlines and descriptions must match the hint: the “gift for a craftsman” group gets an ad for a gift, not for a discount on the set.
What are the most common mistakes with hints?
The first is carrying keywords over from Google Ads one to one — the first advertisers note that the two are completely different. The second is a hint that is too broad (“tools”), which reaches conversations without purchase intent and raises the cost per click. The third is mixing different intents in one group, which makes the ad imprecise for all of them. The fourth is describing an audience instead of a situation (“men 30–50”) — the system works with intent, not demographics. The fifth is a hint that does not match the landing page: relevance is also assessed by the page, so the mismatch is paid for in the auction.
How do you check whether the hints are working?
Ads Manager does not show which specific conversations triggered the ad. This is one of the limitations of the beta phase that the first advertisers point out. So results are read by ad group: impressions, CTR, cost per click and, where tracking is working, conversions and cost per result for each group. Groups with impressions but no engagement usually have a hint that is too broad or too vague. Groups with no impressions have a hint that is too narrow or a bid that is too low. With a bulk upload, hints are submitted as a complete list: the upload replaces the old hints rather than adding to them.
FAQ — Frequently asked questions about context hints
Can hints guarantee that an ad is shown for a particular question?
No. OpenAI states explicitly that hints do not guarantee delivery for specific words, topics, audiences or situations. They steer the matching, and the final choice also depends on the auction.
How many hints should I write for one ad group?
The official documentation does not set a number. The practical guideline from the first advertisers is between five and fifteen specific hints per group, each describing a different situation within one topic.
Can I use my keywords from Google Ads?
As source material for understanding intent, yes. As hints, no. A hint is a sentence describing a situation, not a word.
Should hints be written in Bulgarian or in English?
OpenAI does not set a language. It makes sense to describe the situation in the language your customers use in their conversations: Bulgarian for a Bulgarian audience, and a separate group for each language in multi-country campaigns.
Can I see which conversations triggered the ad?
No. The report shows impressions, clicks, spend and conversions by campaign, group and ad, but not the conversations themselves. OpenAI does not share conversations with advertisers.
Conclusion
Context hints are at the heart of targeting in ChatGPT Ads. They are not keywords but descriptions of situations: who needs your product, when, and with what problem. A good structure is one topic per ad group, specific hints for real needs, several different ads and a page that matches the same situation. The result is read by group, not by individual conversation, and it improves by refining the hints, not by raising the bid.
Request a quote for ChatGPT ads →
Related articles: Advertising in ChatGPT: how it works · ChatGPT Ads vs Google Ads · Negative keywords in Google Ads
Sources: OpenAI Help Center — Create Ad Groups for ChatGPT Ads · OpenAI Help Center — Bulk Upload Checklist · HawkSEM — ChatGPT Ads Context Hints
