In short: Context hints are free-text descriptions of which conversations your ad belongs in, and they replace the keywords you know from Google Ads. They are written at the ad group level, matched semantically against the whole conversation, and they guide the system rather than control it. This guide gives you the template we use ourselves, five patterns that work, the seven most common mistakes and a simple testing methodology. The main rule first: Short, concrete hints of one to two sentences almost always beat long essays.
What context hints are, and why they are not keywords
In ChatGPT Ads there is no keyword list. Instead you write context hints in the ad group: Short descriptions of which conversations, needs and situations the ad is relevant for. The system reads the whole conversation the user is having with ChatGPT, understands topic and intent, and judges whether your ad fits. Your hints are the argument in that judgement, not a switch you control it with.
That sounds like a detail, but it turns the whole craft upside down. A keyword in Google Ads is an exact contract: If the search matches, the ad shows. A context hint is a semantic description that covers every phrasing of the same need. Where Google required fifty keyword variants, one good hint covers the whole cluster, because the model understands that "tool for measuring AI visibility", "how to check whether ChatGPT recommends us" and "GEO tools for businesses" are the same conversation. It is exactly the same semantic matching that decides which sources an AI answer is built on, just applied to ads. If you know GEO, you know more about context hints than you think.
Short or long hints?
This is the most common question, and the answer from early advertisers is fairly clear: Short, concrete hints work best as the default. A hint should be one, at most two sentences. Short hints keep the focus on the core message, give the system room to match broadly within the need, and are quick to write, debug and adjust. Long hints lock the matching to rigid phrasings and increase the risk that the system overlooks what matters most.
The nuance is that full questions, hints phrased the way the user actually asks, seem to reach slightly different conversations than the short variants. So our recommendation is not "never long", but: Build the foundation with short hints, and test the question format in a separate ad group if you want to explore. A good rule of thumb is 5 to 15 hints per ad group, all about the same need.
The template: Who, stage, need and signals
A good context hint answers four things in one sentence:
- Who is the user? Role or situation, not demographics.
- Which stage are they in? Exploring, comparing or ready to choose.
- Which need drives the conversation? The problem, not your product category.
- Which signals do they typically mention? Platform, budget, industry or requirements.
Put together: "Marketer in a mid-sized company comparing AI visibility tools and wondering how ChatGPT talks about their brand." Notice what is not there: No product names, no keyword list, no superlatives. Just a precise description of a conversation you want to be present in.
The most important advice in this whole guide sits in point three: Describe the user's situation, not your category. "Team struggling to keep track of deliveries" performs better than "project management tool", because people describe problems in conversations, not categories.
Five patterns that work
- Persona plus intent. "Owner of a trade business who wants more inquiries without increasing the ad budget." Connects who with what they are trying to achieve.
- Question format. "How does my business become visible in ChatGPT and AI search?" Written the way the user actually asks, and worth testing in its own group.
- Topic plus disqualifier. "Business evaluating an agency for AI visibility, not students or private individuals writing a paper about AI." The disqualifier filters out conversations that look similar but never convert.
- Outcome focus. "Wants to be recommended when customers ask ChatGPT to suggest suppliers in their industry." Describes the outcome the user is after.
- Comparison. "Comparing tools and agencies for generative engine optimization and AI visibility." Captures the conversations where the choice is actually made.
For our own campaign we use hints like "how to become visible in ChatGPT and AI search", "business evaluating AI visibility tools" and "marketer wondering how AI talks about their brand". Short, concrete, one need: That is the whole recipe.
The seven most common mistakes
- Pasting in the keyword list from Google Ads. Term lists without context give the system little to work with, and this is by far the most common mistake.
- Repeating the ad copy. The hint should describe the conversation, not sell the product. That is the ad's job.
- Mixing several audiences in one group. When the hints point in three directions, you cannot read what is working. One need per ad group.
- Role without problem. "Marketing managers in Norway" is not a conversation. "Marketing manager watching Google traffic decline" is.
- Dropping the disqualifier. Without negative signals you pay for conversations that resemble your audience but are not it.
- Ignoring the buying stage. A hint about "what is AI advertising" and one about "ready to choose a supplier" belong in different groups, with different ads and landing pages.
- Never revising. The hints are a hypothesis, not a one-time setup. Those who adjust weekly in the first months build the head start.
Consistency: Hint, ad and landing page must tell the same story
The system does not judge the hints in isolation. The conversation's context, the ad's title and copy and the landing page are weighed together, and gaps between them weaken delivery. If the hint promises a conversation about AI visibility for businesses, the ad must speak to businesses and the landing page must be about exactly that. This is the same consistency principle that governs organic AI visibility, and if you have already cleaned up your content and your technical setup, you harvest the gains on the paid side too.
How to test your way forward
You get no search term report in ChatGPT Ads, so the effect of the hints has to be read indirectly:
- Low delivery suggests hints that are too narrow or too specific. Broaden the descriptions or add more variants of the same need.
- High delivery with low CTR suggests the hints are reaching the wrong conversations. Sharpen the descriptions and add disqualifiers.
- Many clicks with poor engagement points to the landing page or to a gap between hint and ad, not necessarily to the hints themselves.
Run changes in separate ad groups instead of overwriting what works, and give each test a few weeks of data before you conclude. The tracking that makes this reading possible is set up with the OpenAI pixel and your own UTM parameters.
Where to go next
Context hints are the single most important lever in ChatGPT Ads, but they work best when the rest of the setup is in place: The campaign structure and ad format from the step by step guide, the buying model and budget from the guide on CPM, CPC and oCPC, and the measurement from the pixel guide. The whole paid AI landscape, with the platforms, the prices and the interplay with organic visibility, lives in the pillar guide on AI advertising. And remember the organic backdrop: The ad buys you a place next to the answer, but what ChatGPT actually recommends is decided by the visibility you have built. Both are the same semantic craft.