How ChatGPT Memory Shapes Which Ads People See
ChatGPT may use what it remembers across chats to pick relevant ads. Here is what that means for targeting, creative and privacy-safe copy.
On most ad platforms, targeting is something you configure: ages, interests, lookalikes, keyword lists. ChatGPT ads work differently. You describe the situations your ad belongs in with context hints, and the system decides which conversations match. One of the inputs it can use is memory: what ChatGPT remembers about a person across their chats.
That has real consequences for how you write context hints and ads, and for what you must never write. This guide explains how memory fits into ad selection based on what OpenAI has described publicly, what it changes for advertisers, and the privacy lines to respect.
The short version
- ChatGPT ads are selected using context hints, plain-language descriptions of conversations and situations an ad belongs in.
- ChatGPT may also use what it remembers across a person's chats (memory) to pick relevant ads, not only the current conversation.
- Ads appear below ChatGPT's answer, clearly labeled as sponsored, and do not change the answer.
- Advertisers do not see individual conversations or memories, and ads must never imply they know what someone said.
What ChatGPT memory is
Memory is a ChatGPT feature that lets it carry useful details from one conversation into later ones. If someone mentioned they have a toddler, are training for a half marathon, or rent a small apartment, ChatGPT may remember that and use it to make later answers more relevant. Users can view and manage what is remembered, and can turn memory off, through ChatGPT's settings.
For ads, the relevant point is simple: the system choosing which ad to show may consider not only the question in front of it but also what it knows about the person's broader situation, where memory is in use.
How memory changes ad matching
Take a hypothetical example. Someone asks ChatGPT, "What should I make for dinner tonight? Something quick." On its own, that question could match ads for meal kits, recipe apps, cookware, or grocery delivery.
Now suppose ChatGPT remembers from earlier chats that this person has two young kids and avoids dairy. An ad whose context hints describe "parent looking for quick weeknight dinners their kids will eat" or "household cooking dairy-free meals" is a closer fit than a generic cookware ad.
That is the practical effect of memory: it gives the system more of the situation, so ads written for specific situations have more chances to match the right person, even when the current question is short or vague.
What this means for your context hints
Memory rewards the same kind of context hints that work well anyway: descriptions of real situations, needs and intent, not keyword lists.
| Weak hint | Strong hint | Why the strong one fits memory better |
|---|---|---|
| dinner recipes | parent looking for quick weeknight dinners their kids will eat | Describes a life situation memory might hold |
| running shoes | beginner runner training for a first half marathon who wants cushioned road shoes | Captures a goal that persists across chats |
| apartment furniture | renter furnishing a small studio apartment on a budget | Situation, constraint and intent |
| skincare | adult with sensitive, dry skin looking for a fragrance-free moisturizer | Specific need and preference |
| dog food | owner of a senior dog with a sensitive stomach | Ongoing circumstance |
Notice that the strong hints describe circumstances that last beyond one chat: being a parent, training for an event, living in a small space, having a pet with a health need. Those are exactly the kind of details memory tends to carry. Write hints around them.
The context hints generator turns a product description into situation-based hints you can edit, and our guide to context hints covers the fundamentals.
Write hints for durable situations and in-the-moment needs
A balanced set of context hints for one ad usually includes both:
- Durable situations that memory may capture: "parent of a newborn," "home cook who follows a gluten-free diet," "person who works from home full time."
- In-the-moment needs that the current conversation reveals: "looking for a last-minute birthday gift," "comparing options before buying this week."
An ad for a $45 insulated water bottle might use:
- Hiker planning day trips who wants water to stay cold for hours
- Parent packing drinks for kids' sports practice
- Person trying to drink more water during the workday
- Shopper comparing reusable bottles that fit a car cup holder
Each one describes a real reason to buy, some long-lived and some immediate.
The line you must not cross
Memory makes ads more relevant, but OpenAI's policies forbid ads that exploit it in the wrong way. Your ad must never imply it knows what someone said in their conversation or what ChatGPT remembers about them.
Not acceptable:
- "Still looking for that dairy-free dinner?"
- "Since you're training for your half marathon..."
- "We noticed you've been stressed about money."
Acceptable:
- "Dairy-free dinners ready in 20 minutes"
- "Cushioned shoes for half marathon training"
- "A simple budget app for everyday spending"
The acceptable versions are relevant to the same situations, but they speak to anyone in that situation, not to one specific person. That is both a policy requirement and good practice: ads that feel like they are watching you erode trust in the advertiser.
Also avoid exploiting hardship or insecurity. Memory may help an ad reach someone in a difficult moment, which is precisely why copy that preys on fear, shame or financial distress is prohibited. Our ChatGPT ad policies guide lists the full rules, and the ChatGPT ad policy checker flags risky phrasing.
Example ads written for memory-aware matching
Here are three hypothetical ads that work well when matching considers a person's broader situation, while never referring to it directly.
Meal kit
- Headline: Weeknight dinners kids will eat
- Body: Family-size meal kits with 20-minute recipes. Pick your meals each week and skip anytime.
Senior dog food
- Headline: Gentle food for older dogs
- Body: Easy-to-digest recipe made for senior dogs. Vet-formulated. Try a small bag first.
Home office cushion
- Headline: Comfort for long desk days
- Body: Memory foam seat cushion that fits most office chairs. 30-day returns on every order.
Each headline is under 50 characters and each body under 100. Each speaks to a situation, not to an individual, and every claim (like "vet-formulated" or "30-day returns") must be true and shown on the landing page.
What memory does not do
It is worth being clear about the limits, because there is a lot of speculation:
- Advertisers do not get access to memories or chats. You write hints; the system matches. You see aggregate performance, not individual conversations.
- You cannot target "people whose memory says X." There is no audience builder for memory contents. Memory is one input the system may use, not a targeting option you control.
- Ads do not change ChatGPT's answers. The answer is the same whether or not an ad appears.
- Paid-tier users are not shown ads. Memory on a Plus or Pro account does not lead to ads, because those tiers are ad-free.
Measuring the effect
You cannot isolate memory's contribution in reporting, but you can see its effects indirectly. If ads with situation-based hints consistently earn higher click-through rates than ads with generic hints, that is consistent with better matching. Similarweb reported an average ChatGPT ads CTR around 0.68%, with top brands near 1.57%. Compare your ads against that range and against each other, and judge final success on cost per conversion rather than CTR alone. Track it with the CTR calculator.
How SecondWin uses this
SecondWin builds every ad around a specific buyer question and writes context hints that describe the situation behind it, both the durable circumstances memory may capture and the in-the-moment need. Every ad is checked so it never implies knowledge of anyone's conversation, against OpenAI's policies and our own rules. Campaigns run in your own OpenAI ad account, and you can see every hint we write.
To see the buyer questions and context hints we would use for your product, get a free ChatGPT ads analysis of your site.
FAQ
Does ChatGPT use memory to show ads?
ChatGPT may use what it remembers across a person's chats to pick relevant ads, in addition to the current conversation and your context hints. This is one input to matching, not a targeting setting you control. Users can view, manage and turn off memory in ChatGPT's settings, and paid-tier users do not see ads at all.
Can advertisers see what users told ChatGPT?
No. Advertisers write context hints describing the situations their ads belong in, and the system does the matching. Advertisers see aggregate results like impressions, clicks and conversions, not individual conversations or memories. Ads also must not imply they know what someone said, which OpenAI's policies prohibit.
How should I write context hints with memory in mind?
Describe real situations rather than keywords, and include both durable circumstances and immediate needs. Durable situations like being a new parent, training for a race, or owning a senior dog are the kind of details memory may carry. Immediate needs like a last-minute gift capture what the current conversation reveals.
Can my ad mention something the user said?
No. Copy like "Still looking for..." or "Since you mentioned..." implies the ad knows the person's conversation, which breaks OpenAI's ad policies and feels intrusive. Write ads that speak to anyone in a given situation, such as "Gentle food for older dogs," so they are relevant without referring to an individual.