ChatGPT Ads context hints: how conversational targeting works

If you have run Google Ads, your instinct will be to treat context hints as keywords. That instinct is wrong in a way that will cost you money, and OpenAI's own documentation is unusually direct about why.

This is the mechanic that makes conversational advertising different from search advertising. Getting it right is most of the skill in the channel right now.

Where they live

Context hints belong to the ad group, not the campaign and not the individual ad.

Each ad group contains a title and a set of context hints that describe, in OpenAI's words, "what your product offers, who it helps, or when it may be useful." Their job is to help the system understand when your ads may be relevant to conversations or user journeys in ChatGPT.

So the structure is: campaign holds the objective, budget and geography. Ad group holds a theme and its context hints. Ads hold the message. If you have built search campaigns, the shape is familiar even though the contents are not.

What they are not

OpenAI states this plainly, and it is worth quoting in full because almost every misconception about this channel dies here:

"Context hints are not exact-match controls, audience targeting rules, or instructions to show your ad only for specific conversations. They do not guarantee that your ad will appear for particular words, topics, audiences, or situations."

Three things follow.

You are not buying a keyword. There is no match type, no bid per term, no negative list acting as a hard block. You are describing a situation and the system decides relevance.

You are not defining an audience. A context hint is not a demographic or an interest segment. Describing "homeowners aged 35 to 55" is using the wrong tool.

You are not guaranteeing placement. Writing a hint does not reserve the conversations that match it. Your ad competes on relevance and bid like everything else.

The corollary is liberating once it lands: people do not need to use your exact words. OpenAI notes that being specific does not require the user to phrase things the way you did. The system is built to understand broader needs, situations and natural variations in how people describe a problem. You are writing for meaning, not for string matching.

OpenAI's own example, and why it works

The documentation gives a worked comparison. Instead of a broad hint like:

running shoes

write something like:

cushioned everyday running shoes for beginners training for their first 5K

Look at what the second one adds. It carries a product characteristic (cushioned, everyday), an experience level (beginners), and a situation with a deadline attached (training for a first 5K). None of that is in "running shoes," and all of it helps the system distinguish a conversation worth showing your ad in from one that merely mentions running.

The second hint is also more useful commercially. Someone training for a first 5K is a different buyer from a competitive marathoner, and if you sell to one and not the other, the broad hint invites the wrong conversations and you pay for them.

What a good hint contains

Drawing directly from OpenAI's guidance, effective context hints:

• Add information beyond your ad title, copy and landing page. If the hint just restates your headline, it has contributed nothing. The system already reads all three.
• Connect product features to real customer needs or use cases. Not "24/7 dispatch" but what having it solves at eleven at night in February.
• Describe needs or situations rather than broad audience labels. Situations are what conversations contain.
• Use clear, natural phrases rather than disconnected terms. This is prose, not a keyword list. Comma-separated fragments waste the mechanism.
• Include only genuine, relevant use cases. Padding the list with adjacent-but-untrue situations invites irrelevant traffic you pay for and do not convert.
• Stay focused on the theme of the ad group.

That last point has a structural consequence. If your hints start describing meaningfully different products or use cases, ones that would need different ad messaging or a different landing page, that is the signal to split them into separate ad groups. OpenAI says so explicitly, and it is the same discipline that keeps search accounts readable.

How to actually write them

The mental shift is from "what would someone type" to "what situation is this person in."

A useful exercise: think of the last five customers you genuinely helped, and write the sentence each of them would have said to a friend before they found you. Not the search they ran. The sentence.

For a heating company that might be:

our furnace stopped overnight and we have young children in the house

For an accountant:

we have grown past the point where a bookkeeper is enough and I am not sure what a controller actually does

For a renovation contractor:

we want to open up the main floor but we are not sure which wall is structural

Those are context hints. They describe circumstances, constraints and the shape of the problem. They read like sentences because conversations are made of sentences.

Then check each one against the test above: does it add anything my ad copy and landing page do not already say? If your landing page already screams emergency furnace repair, the hint that earns its place is the one about young children in the house, because that is the constraint that makes the job urgent and the customer decisive.

The organizing principle

One theme per ad group. Enough hints inside it to cover the genuine variations of how people describe that theme. A different theme gets a different ad group, its own messaging and its own landing page.

That structure is not bureaucracy. It is what makes results readable later. If one ad group holds four unrelated situations, a mediocre overall result tells you nothing about which situation worked, and you will end up guessing at the optimization.

What this means for the work

Context hints reward businesses that genuinely understand their customers' circumstances, and punish generic messaging in a way keyword bidding never quite did. You cannot brute-force your way in with a large term list and a big budget, because the mechanism is not built to be brute-forced.

That is good news for a well-run small business competing against a larger, blunter advertiser. The advantage goes to whoever can describe the customer's situation most precisely and truthfully, which is a writing and customer-knowledge problem more than a media-buying one.

It is also why this work does not get delegated to a keyword tool. The best context hints in a business usually already exist, in the words the owner uses when describing the job they do best.

Last reviewed against OpenAI documentation: August 2026. Ad group mechanics on this platform are still changing, so verify before rebuilding an account around them.