ChatGPT remembers more than you think

Most business owners picture ChatGPT ads matching ONLY the live conversation.

Someone asks about dog food, a dog food ad shows up. That's true. It's also barely half the story.

Unfortunately, stopping there means you're only writing for one of the three layers OpenAI uses to pick your ad.

The juicy one is "past chats," and almost nobody is writing context hints for it yet.

I may be wrong, but I’m pretty sure that OpenAI had previously said they wouldn’t incorporate chat history into ad decision trees.

…if that was the case before, clearly we’re in a brave new world.

Why almost nobody targets this layer

• They assume the targeting system only reads the current conversation. It also pulls from past chats and memory, weeks of context that most advertisers never write hints for.
• They write hints like search keywords: "when users ask about X." That catches one moment. The memory layer can match someone who's been researching your category for three weeks straight.
• They run pushy, billboard-style creative without knowing that every time someone hides an ad, that signal goes back to OpenAI's system and reduces future delivery. Bad creative compounds against you here in a way it doesn't on Google.
• They write one hint per product. Your customer probably has 3-5 different conversations with ChatGPT in the weeks before they buy. Each one is an unmatched placement.

Once you can see all 3 layers, writing for them is just different sentences.

Step 1: Learn what's actually in the targeting stack

3 layers shape which ads a user sees:

1. The live conversation: the topic being discussed right now.
2. Past chats and Memory: saved preferences, ongoing projects, interests ChatGPT has noted over time. Users can toggle this off in their ad controls.
3. Ad interactions: ads they've clicked, asked about, or dismissed.

Knowing the stack exists is half the work.

But the other half is writing hints that speak to all 3 layers, which is what Steps 2 and 3 are about.

Step 2: Write context hints for the slow burn (CPM), not just for the moment (CPC)

Most advertisers write hints for single questions. "When users ask about X." One moment, one match, done.

The memory layer makes that approach leave money on the table.

Big purchases, software, courses, equipment, B2B services, anything with a research phase, don't happen in one conversation. They happen across weeks of questions, comparisons, and revisits.

The memory layer sees all of that. A hint written for the journey catches a buyer who's been building toward a decision. A hint written for a moment catches them once and loses them.

Here's the difference in practice:

• "when users ask about cabinet prices" catches a moment.
• "when users have been planning a kitchen renovation and are comparing options or costs over time" catches a project in progress.

The second hint matches a person who's already 3 weeks into the decision. That's the person closest to buying.

What this looks like for different businesses:

• Coffee gear brand: "when users have been exploring home espresso setups and ask about grinders, beans, or technique across conversations"
• B2B SaaS: "when users are evaluating tools for a process change, have mentioned team size or budget, and return to compare specific options"
• Course creator: "when users keep asking how to learn [your skill] and have discussed timelines or career goals"

The exercise: think about the 3 conversations your customer has with ChatGPT in the month before they buy. Write one hint per stage. That's a funnel built out of language, and it costs nothing to set up.

Step 3: Respect the dismiss button

OpenAI lists "ads you hide or engage with" among the signals that shape which ads get shown in the future.

So when someone dismisses your ad, the platform logs it and quietly starts showing your ads less to similar users going forward.

As it stands, ChatGPT will not tell you if people are clicking “Block” on your ad.

Those signals stack up.

• On Google, a bad ad wastes one impression and nothing else.
• Here, every dismissal trains the platform away from you. Run enough bad creative and you're effectively shut out of the conversations you most want to be in.
• The flip side works the same way: helpful, low-key, conversation-matched ads earn positive signals that compound in your favor over time.
• The platform grades every impression. The grade sticks.

This is why ad quality matters more on ChatGPT than on any platform you've run before.

The math behind future delivery punishes bad creative in ways Google never did.

Hints for projects beat hints for moments every time

The targeting stack rewards advertisers who think in terms of customer journeys.

That's a soft, holistic skill shift - and the business owners who make it early get the cheap end of the auction.

Everything above is sourced, so read it yourself: openai.com/index/testing-ads-in-chatgpt‍