Context Hints, Not Keywords: How ChatGPT Ads Targeting Actually Works (With Examples)

There is no keyword bidding in ChatGPT ads. There are no match types, no audience segments to stack, no lookalikes. There's a free-text box called "context hints" - and how you fill it decides which conversations your ad enters. Here's how to write hints that work, with the framework we use for our own campaigns.

What are context hints?

Context hints are plain-language descriptions - OpenAI recommends 5-15 per ad group - of the conversations where your ad belongs. Each hint is one or two sentences describing a buyer moment. Hints guide ChatGPT's conversational matching; they are not exact-match targeting rules (OpenAI Ads Manager).

That last clause is the part most advertisers miss. In the Ads Manager, the hints field sits where the keyword builder would sit in Google Ads, so search-trained marketers treat it like one - and get burned. A hint does not say "serve my ad when these words appear." It says "here is the kind of conversation I belong in," and the matching model decides how far to generalize. Your hint about "founders researching AI search visibility" can match a conversation that never uses the word "visibility" - while a bare keyword can match almost nothing, because a bare keyword describes no conversation.

The practical consequence: the writing itself is the targeting. There is no bid modifier, no match-type toggle, no negative list to rescue a badly written hint. The rest of this post is about how to write them well.

Why keyword thinking fails here

Search ads intercept a query. ChatGPT ads join a conversation - after the model has already answered. The user asked, got a real answer (OpenAI is explicit that ads never change the answer), and then sees one sponsored unit matched to the discussion.

So the unit of targeting isn't a keyword, it's a conversation. "ai visibility tool" is a keyword. "A DTC founder asking why their brand never comes up when shoppers ask ChatGPT for recommendations" is a conversation. The hints field holds the second kind.

Context hints vs keywords

The two systems look adjacent and behave nothing alike. Here is the full delta:

The ad-tech industry has reached the same conclusion

This is not just our reading of the field. Criteo - the first ad-tech partner in OpenAI's ChatGPT ads pilot (Criteo press release, March 2026) - draws the identical line in its Prompt Smart Ads beta materials: a keyword is a target word you match against a search, working through keyword ranking; a context hint is a signal that helps the model understand what topics, products, or audiences are relevant to a chat, working through OpenAI's relevance and intent matching. Their example pair: the keyword is "running shoe"; the context hints are "marathon training" and "fitness journey."

Criteo's whole product bet confirms where the leverage sits. Prompt Smart Ads, launched June 2026, auto-generates creatives and context hints from catalog data and prompt-level insights - and Criteo reports roughly 4x higher spend after activation, meaning campaigns match into far more conversations (Criteo, June 2026). When the largest ad-tech player on the surface builds its differentiator around hint-writing, hint-writing is the game. We break down the product, its numbers, and its creative-report lessons in our Criteo Prompt Smart Ads analysis.

Two rows deserve emphasis. Negative control: search marketers lean on negatives as a safety net, and that net does not exist here - precision has to be built into the hint itself. Measurement: because you never see the matched conversations, the discipline shifts from post-hoc query mining to upfront hint quality. You cannot fix in reporting what you failed to specify in writing.

A format that works: audience + situation + questions

We write every hint block with three ingredients:

1. Who's talking - "ecommerce founders, DTC brand marketers, heads of growth"
2. What's happening - "researching why their brand doesn't appear in AI shopping answers"
3. The literal questions - the long-tail phrasing buyers actually type

Example - one of our real ad groups:

Conversations where ecommerce founders or DTC marketers are asking about AI search visibility: why their brand doesn't appear in ChatGPT or Perplexity shopping answers, how to see AI referral traffic in Google Analytics, why GA4 shows unexplained direct traffic, and how answer engine optimization works for online stores.

Every clause is a conversation we've seen buyers have. None of it is a keyword list.

Three worked examples

The format holds across categories. Here is the same audience + situation + questions structure applied to three different businesses.

DTC ecommerce - a skincare brand.

Conversations where someone with sensitive or acne-prone skin is asking for a simple skincare routine: which ingredients to avoid, whether they need both a moisturizer and an SPF, what actually helps with hormonal breakouts, and which affordable brands dermatologists recommend for reactive skin.

The audience is implied by the situation ("someone with sensitive or acne-prone skin") rather than stated as a demographic - there is no demographic targeting to hand you, so the conversation itself carries the qualification. Every clause is a question shoppers ask an assistant before ever visiting a product page - exactly when the single sponsored slot appears. Note what's absent: no "buy skincare online," no brand-name stuffing - those are queries, not conversations.

B2B SaaS - an expense management platform.

Conversations where a startup finance lead, controller, or founder is trying to fix expense chaos: how to stop chasing receipts at month-end close, when to move off shared corporate cards, how expense tools integrate with QuickBooks or NetSuite, and what expense software costs for a 50-person company.

B2B buying conversations are longer and more comparative than DTC ones, so the hint leans into process pain ("chasing receipts at month-end close") and evaluation questions ("what it costs for a 50-person company") - the phrasing of someone mid-research, not someone ready to type a product name. The integration clause matters: "does it work with my stack" is a constant B2B qualifying question, and it lets the hint reach conversations where your category is implied but never named.

Services - an immigration law practice.

Conversations where a skilled worker or their employer is navigating a US work visa: whether they qualify for an H-1B or O-1, what happens if a transfer petition is denied, how long green card sponsorship takes from an H-1B, and when a situation is complicated enough to need an immigration lawyer instead of a filing service.

Services buyers ask assistants "do I need a professional at all?" before they ever search for one - so the last clause targets that exact fork in the conversation, which is where a services ad earns its click. The specificity of the scenarios (denied transfer petition, H-1B to green card) keeps the hint out of general immigration-news conversations that would waste the slot. This is the negative-control problem from the table above, solved the only way it can be: with precision in the writing.

Across all three, the test is the same: read the hint aloud. If it sounds like a paragraph describing a person mid-conversation, it will match. If it sounds like a keyword report, rewrite it.

Where the questions come from

The research is the same work as answer-engine SEO - which is the point. Sources we mine, in order of yield:

• Sales calls. The questions prospects ask in the first ten minutes are the questions they asked an AI assistant the week before.
• Your organic AI citations. Where assistants already cite you reveals the conversations you're matched to organically. A Citation Rank scan shows this per surface.
• Support and community threads. Reddit, Discord, and forum phrasing is conversational by nature - closer to ChatGPT prompts than any keyword tool's output.
• Your own blog's question posts. If a how-to post pulls organic AI citations, its title belongs in a hint block.

Then split by intent, not topic: problem-aware ("why can't I see AI traffic?"), solution-aware ("how do I connect my catalog to ChatGPT?"), and comparison ("X vs Y, what does it cost?"). Three ad groups, three hint blocks, three landing pages. Comparison conversations are the smallest and most valuable - bid accordingly.

Structure ad groups by intent - and match the landing page

The destination should answer the conversation's question, not restate your homepage. Our problem-aware group lands on the free scan; comparison conversations land on pricing - because we publish ours and most competitors don't. Sending a comparison-stage buyer to a homepage hero wastes the most expensive click in the account.

The global clock is ticking

ChatGPT ads currently serve in the United States, United Kingdom, Canada, Australia, New Zealand, Japan, Korea, and - since August 27, 2026 - India, with more markets announced through 2026 (OpenAI expansion posts; The Hindu BusinessLine, Aug 2026). Every new market opens with thin advertiser competition for that single ad slot - the cheapest conversational inventory that market will ever have. Brands that have already done the question research can clone a working campaign into a new market in an afternoon; brands starting from scratch will pay the learning tax at higher CPCs.

The same research compounds on the organic side: the questions in your hint blocks are the questions your content and catalog should be winning without paying per click - the play behind the results in our case studies. Paid rents the conversation. Organic owns the answer. The brands that do both are the ones AI assistants can't route around.

Where Tru Commerce fits

Context hints are one half of a two-layer game. The sponsored slot is rented: it appears only for Free and Go users, it is labelled, and it disappears the day you stop paying. The organic recommendation above it - the answer itself - is earned, and it is the only layer Plus and Pro users ever see. The question research that makes your hints sharp is the same research that should be making your brand the answer.

Start with the earned layer: run a free Citation Rank scan to see where AI assistants already cite you - and where competitors hold the conversations your hints are chasing. Then, when you're ready to add the paid layer, our ChatGPT ads setup guide walks through the full Ads Manager flow, hints field included.

The same measurement-first loop drove our Amazon India engagement, where Top-5 presence rose +2.4% in 6 weeks - see the Amazon case study. If you want to see where you stand before committing, start with Visibility Score.