How Do ChatGPT Ads Work? Matching, Context Hints, Auction, Placement & Measurement

Last verified: September 7, 2026

ChatGPT Ads work by evaluating whether a sponsored message is relevant to the user’s current conversation, determining which ads are eligible to compete, ranking eligible ads using factors that include relevance and advertiser bids, and then showing the selected sponsored placement below the ChatGPT response.¹²

That last part is important:

The ad system does not write the ChatGPT answer.

OpenAI says ads run separately from the model response, are clearly labeled, and do not influence what ChatGPT says.¹

Ads follow the conversation, not the keyword. That one difference changes how you structure everything.

For advertisers, the system is best understood as five connected layers:

1. Eligibility – can an ad appear in this user, conversation, market, campaign, and policy context?
2. Matching – how relevant is the advertiser’s offer to what the person is trying to accomplish?
3. Auction and ranking – among eligible ads, how do relevance, bids, and expected outcomes affect which ad is shown?
4. Placement and interaction – where does the sponsored unit appear and what can the user do with it?
5. Measurement – what happens after an impression or click, and how are outcomes attributed?

That is a more accurate mental model than “ChatGPT Ads are Google Ads with AI keywords.”

The Short Version: How ChatGPT Ads Work

A simplified flow looks like this:

User conversation
→ ChatGPT generates its answer
→ the ads system evaluates whether relevant advertising is eligible
→ eligible ads are matched using conversation context plus advertiser signals
→ eligible ads compete in a relevance-weighted auction
→ sponsored placement appears below the answer
→ user may click or interact
→ advertiser measures delivery and conversions

OpenAI documents several important inputs to that process.

On the user side, the system can consider the context and intent of the current conversation. Depending on the user’s ad-personalization settings, select signals from the broader ChatGPT experience may also contribute to ad relevance.¹

On the advertiser side, signals can include:

• targeting selections;
• the landing page;
• ad title;
• ad copy;
• advertiser-provided context hints;
• advertiser bids;
• conversion signals for supported conversion-optimized campaigns.¹²⁴

The system does not expose the user’s private conversation to the advertiser. OpenAI says advertisers do not receive chats, chat history, memories, names, email addresses, precise locations, IP addresses, or sensitive personal information. Advertisers receive performance reporting rather than the user’s conversation transcript.¹

Step 1: The User Has a Conversation With ChatGPT

The process begins with the user’s conversation, not with a traditional search-engine keyword.

Someone might ask:

I need comfortable running shoes for my first 5K. I overpronate a little, train mostly on pavement, and want to stay under $150.

That conversation contains substantially more context than a short search such as:

running shoes

OpenAI says its ads system considers the context and intent of the current conversation when selecting and delivering ads.¹²

That does not mean advertisers can read the conversation.

The context stays inside ChatGPT’s advertising system. OpenAI says conversations are not shared with advertisers.¹

This creates the first major distinction from traditional paid search:

The advertiser is not choosing the exact user query that must trigger the ad. The advertiser is giving the system information that helps it identify situations where the offer may be relevant.

Step 2: ChatGPT Generates Its Answer Independently

The organic answer and the advertising placement are separate.

OpenAI states that ads:

• do not influence ChatGPT’s answers;
• run on systems separate from the chat model;
• are clearly labeled;
• do not represent an OpenAI endorsement of the advertiser.¹

So if a user asks:

What is the best CRM for a small law firm?

an advertiser cannot pay OpenAI to make the answer itself say:

Product X is the best CRM.

The advertiser is competing for the sponsored placement, not control of the answer.

This also means marketers should separate two objectives:

Paid ChatGPT visibility

Buying eligible sponsored inventory below responses.

Organic AI visibility

Being mentioned or cited inside unpaid answers.

They can support the same business, but they are not the same ranking system.

Our ChatGPT Ads vs Organic AI Visibility guide covers that distinction in detail.

Step 3: The System Checks Whether an Ad Is Eligible

Before an ad can compete, it has to be eligible for the opportunity.

Not every conversation can show every ad.

Eligibility can be affected by factors such as:

• whether the user is in an ads-supported experience;
• age restrictions;
• sensitive or regulated conversation context;
• advertiser and campaign status;
• location and other campaign targeting;
• audience inclusion or exclusion settings;
• ad-policy eligibility;
• campaign dates and budget;
• whether the ad and landing page pass review.¹⁹

OpenAI currently says ads may appear for eligible users on Free and Go plans, while Plus, Pro, Business, Enterprise, and Edu experiences are ad-free. Ads are not shown to accounts identified as belonging to people under 18, and OpenAI excludes ads from sensitive or regulated conversations such as personal health, mental health, and politics.¹

Temporary Chats also do not show ads.¹

That means the theoretical total ChatGPT audience is not the same thing as addressable ChatGPT Ads inventory.

Step 4: Advertiser Targeting Narrows the Eligible Audience

At the campaign level, advertisers can define targeting and eligibility settings.

OpenAI’s current Ads Manager documentation includes options such as:

• geography;
• supported platform targeting;
• custom-audience inclusion;
• custom-audience exclusion.⁹³

Custom audiences can be built from advertiser-provided customer or prospect lists using supported identifiers. They can be used either to limit eligibility to selected audiences or to exclude audiences such as existing customers or recent purchasers.³

This is an important update because some earlier third-party articles described ChatGPT Ads as having “no audiences.”

That is no longer accurate.

The better statement is:

ChatGPT Ads combines campaign-level targeting controls with conversational relevance signals.

The system does not rely on only one or the other.

Step 5: Context Hints Tell OpenAI When the Offer May Be Useful

At the ad-group level, advertisers provide context hints.

OpenAI describes context hints as natural-language descriptions of:

• what the product or service offers;
• who it helps;
• when it may be useful;
• relevant customer needs, situations, or topics.⁷

For example, OpenAI contrasts a broad hint such as:

running shoes

with a more descriptive situation such as:

cushioned everyday running shoes for beginners training for their first 5K⁷

The purpose is to provide meaning that may not be obvious from the ad headline or landing page alone.

Context hints are not exact-match keywords

This is critical.

OpenAI explicitly says context hints are:

• not exact-match controls;
• not audience-targeting rules;
• not instructions to show an ad only for a specific conversation;
• not guarantees of delivery for particular words or topics.⁷

So an advertiser should not build a giant list of thousands of keyword variations and assume the platform will behave like exact-match Google Search.

A useful context hint describes a customer situation.

That can include language people actually use, but the goal is semantic relevance rather than matching one exact query.

Step 6: The Ad, Landing Page, and Context Hints All Contribute to Relevance

Context hints are only one signal.

OpenAI says its system can consider:

• the current conversation’s context and intent;
• the ad landing page;
• ad title;
• ad copy;
• advertiser-provided context hints;
• targeting selections;
• and, when ad personalization is enabled, select signals from the user’s broader ChatGPT experience.¹²

That means campaign relevance is not something the advertiser controls with one field.

The system is evaluating the relationship between:

what the person appears to need
↔ what the advertiser says it offers
↔ where the ad sends the person

That is why poor message alignment can be a structural problem.

If your context hint describes:

payroll software for restaurants managing tipped hourly employees

but the ad says:

Better Business Software

and the click lands on a generic homepage covering ten unrelated products, the three advertiser-controlled signals are not reinforcing the same use case.

The stronger architecture is:

specific customer situation
→ specific ad message
→ relevant landing page

Step 7: Personalized Ads Can Use Additional ChatGPT Signals

The current conversation is a core signal.

For users who have enabled ad personalization, OpenAI says the system may also use select signals from the broader ChatGPT experience to make ads more relevant. Depending on the user’s settings, that can include things such as past chats, memory, and interactions with ads.¹

Users can turn ad personalization off.

When personalization is off, OpenAI says ads can still be selected using the context of the current chat thread, but broader history is not used in the same way.¹

What the advertiser sees

The advertiser does not receive those private signals.

OpenAI says advertisers never receive:

• chats;
• chat history;
• memories;
• names;
• email addresses;
• precise location;
• IP address;
• sensitive information.¹

This distinction matters because “the system can use a signal for relevance” is not the same as “the advertiser can see that signal.”

Step 8: Eligible Ads Enter a Relevance-Weighted Auction

Once ads are eligible and relevant enough to compete, the platform has to decide which sponsored placement should win.

OpenAI currently documents that ChatGPT Ads uses a relevance-weighted, second-price auction among eligible ads.²

OpenAI also says that when multiple ads are eligible, it considers a combination of factors including:

• relevance;
• advertiser bids.¹

This means the highest bid does not operate in isolation.

A more relevant ad can be more competitive than an otherwise weaker ad with a large bid because the system is explicitly weighting relevance in the auction.

At the same time, relevance is not a free substitute for bidding.

An advertiser still competes economically.

What “second-price” means for advertisers

The important strategic point is not to reverse-engineer a proprietary formula that OpenAI has not published.

It is this:

Your maximum bid is part of the competition, but the auction also incorporates relevance, and the system is not simply ranking ads from highest max bid to lowest.

That gives advertisers two broad levers:

1. Economic competitiveness – bid and budget.
2. Relevance quality – campaign structure, context hints, creative, landing page, and offer fit.

You need both.

Step 9: Bidding Works Differently by Campaign Objective

OpenAI currently supports Reach, Clicks, and conversion-oriented campaign objectives.

Reach campaigns

Reach campaigns use CPM buying.

The advertiser sets a maximum CPM bid at the ad-group level and pays per 1,000 impressions.²

Clicks campaigns

Clicks campaigns use CPC buying.

The advertiser sets a maximum CPC bid at the ad-group level and pays for valid clicks.²

OpenAI currently recommends starting CPC campaigns around a $3–$5 USD maximum CPC bid, with Ads Manager potentially providing bid-strength guidance. That is platform guidance, not a universal performance benchmark.²

Conversion-optimized campaigns

OpenAI also supports conversion-optimized CPC, or oCPC, campaigns.

These campaigns are still billed on valid clicks, but delivery is optimized toward a selected supported conversion event.⁴

For oCPC campaigns, the advertiser sets a Bid Cap representing the maximum amount they are willing to bid for a conversion. OpenAI says the system uses predicted conversion likelihood to determine a per-click auction bid while optimizing toward that conversion event. The actual amount billed remains click-based and is determined through the auction.⁴

This distinction matters:

oCPC does not mean “pay per conversion.”

It means the click auction is informed by conversion optimization.

Step 10: The Winning Sponsored Unit Appears Below the Answer

When there is a relevant match, the sponsored unit appears below the ChatGPT response.¹²

OpenAI’s current documentation says a ChatGPT ad can include:

• advertiser name;
• favicon or advertiser logo;
• title;
• description;
• landing page;
• image creative.²

Ads are visually separated and labeled as sponsored.¹

How many ads appear?

Some early third-party guides described ChatGPT Ads as a single-ad-slot system.

OpenAI’s current user-facing documentation is broader: during the test, users may see one or more ad units below a response when there is a relevant match. A unit may contain one or more items from one advertiser or, in some cases, multiple advertisers.¹

Because the product is evolving, use current first-party documentation instead of assuming the earliest launch format is permanent.

Step 11: The User Can Click, Dismiss, Report, or Ask About an Ad

Advertising is not just a static impression.

OpenAI’s current user controls allow eligible users to:

• dismiss an ad;
• report an ad;
• see information about why it was shown;
• share a specific ad with ChatGPT using “Ask ChatGPT” and ask questions about it.¹

By default, ChatGPT cannot see the ad that was shown to the user unless the user explicitly shares that ad into the conversation through the supported interface.¹

Those interactions are part of the overall ad experience but should not be confused with advertiser access to the underlying conversation.

Step 12: The Click Goes to the Advertiser’s Landing Page

When a user clicks the ad, the destination is the advertiser’s selected landing page.

The landing page matters in two different ways:

Relevance

OpenAI includes the landing page among the inputs its system can consider when assessing ad relevance.²

Conversion

The landing page is where the advertiser typically begins measuring downstream actions.

That means landing-page quality affects more than traditional conversion-rate optimization.

It also helps communicate to the ads system what the advertised offer actually is.

A useful operating chain is:

conversation need
→ context hint
→ ad message
→ landing page
→ conversion event

If one step describes a different offer than the next, the campaign becomes harder for both the user and the platform to interpret.

Step 13: Conversion Measurement Connects the Click to Business Outcomes

OpenAI supports conversion measurement using:

• the OpenAI Pixel;
• the Conversions API;
• or both.⁵

OpenAI can report a conversion when:

1. an event is received from a connected data source;
2. that event matches the conversion event configured for the campaign;
3. it occurs within the applicable attribution window;
4. OpenAI can connect it to an eligible ad click using available measurement signals.⁵

One of those signals is OpenAI’s click reference, oppref, which can be appended to the landing-page URL and associated with later conversion events.⁵

The system can also use eligible advanced-matching and modeled-measurement signals where available.⁵

For stronger measurement resilience, OpenAI supports using both Pixel and Conversions API and deduplicating the same event when the same event ID is supplied.⁵

Step 14: Ads Manager Reports Delivery and Outcomes

OpenAI’s current Ads Manager reporting includes metrics such as:

• impressions;
• clicks;
• spend;
• CTR;
• average CPC;
• average CPM;
• conversions.⁶

For eligible accounts, Ads Manager may also report one-day view-through conversions separately from click-through conversions.⁶

OpenAI explicitly separates those view-through conversions from the primary Conversions metric and says they do not affect billing, bidding, CPA, or conversion optimization.⁶

Conversions may also take approximately 24–48 hours to appear in reporting.⁶

That means same-day reporting is not always a complete picture of downstream performance.

What Determines Which ChatGPT Ad Wins?

OpenAI does not publish every weighting or proprietary scoring detail.

But the current documentation supports four useful categories of inputs.

1. Eligibility

Can the campaign serve at all?

This includes user eligibility, policy context, targeting, campaign status, budget, and related delivery constraints.

2. Relevance

Does the offer fit what the user is trying to accomplish?

OpenAI can consider:

• current conversation context and intent;
• landing page;
• ad title;
• ad copy;
• context hints;
• targeting selections;
• select personalization signals when enabled.¹²

3. Bid

The advertiser’s bid affects auction competitiveness.¹²

The exact bid mechanic depends on the campaign objective.

4. Expected outcome

OpenAI describes selection and delivery as being based on expected relevance and outcomes.¹²

For oCPC campaigns, OpenAI goes further and explicitly says predicted conversion likelihood helps determine a per-click auction bid.⁴

Do not extrapolate beyond what is documented.

OpenAI has not published a public “ChatGPT Ad Rank” equation equivalent to a simple formula marketers can calculate independently.

Are ChatGPT Ads Keyword Targeted?

Not in the traditional exact-match paid-search sense.

OpenAI lets advertisers provide context hints that can include conversations, topics, or language associated with relevant use cases, but it explicitly says those hints are not exact-match controls.²⁷

So the useful comparison is:

Google Search mental model

“Which searches or keyword themes should this ad compete for?”

ChatGPT Ads mental model

“In what conversational situations is this offer genuinely useful, and what information should the system understand about that fit?”

That is a different planning exercise.

It places more pressure on:

• customer research;
• offer clarity;
• use-case definition;
• landing-page specificity;
• diverse creative coverage.

Do ChatGPT Ads Use Audiences?

Yes, in certain ways.

Current Ads Manager documentation includes campaign-level custom audiences that can be used for inclusion or exclusion.³

That means it is outdated to say ChatGPT Ads has “no audiences.”

However, custom audiences do not eliminate the importance of contextual matching.

A campaign might say:

Only these known prospects are eligible

while the ads system still evaluates whether a specific ad is useful in the current conversational context.

Audience eligibility and conversational relevance can work together.

Do ChatGPT Ads Use Personal Data?

OpenAI draws a line between data used internally for ad relevance and data exposed to advertisers.

Depending on the user’s settings, ChatGPT may use:

• the current chat;
• general context such as approximate location or language;
• select broader ChatGPT signals when ad personalization is enabled.¹

But OpenAI says advertisers do not receive the user’s private ChatGPT conversation, history, memories, identity, precise location, IP address, or sensitive personal information.¹

Advertisers receive campaign-performance reporting rather than a raw transcript of the user’s decision process.

What Happens in Sensitive Conversations?

OpenAI says ads are not eligible to appear near sensitive or regulated topics including:

• personal health;
• mental health;
• politics.¹

There are also advertiser-category policies and age restrictions.

This is part of the eligibility layer.

A campaign may be perfectly relevant to a broad topic and still be ineligible for a specific conversation because the context falls inside a restricted or sensitive area.

How Are ChatGPT Ads Different From Google Search Ads?

There are similarities.

Both can involve:

• bidding;
• auctions;
• landing-page relevance;
• paid clicks;
• campaign structure;
• conversion tracking.

But the matching environment differs substantially.

Google Search Ads	ChatGPT Ads
Search query is the primary expression of intent	Conversation can contain broader contextual intent
Keyword/search-theme controls are central	Context hints describe useful situations but are not exact-match controls
Search results page	Sponsored unit appears below a ChatGPT response
Audience tools can supplement search intent	Campaign targeting/custom audiences can supplement conversational relevance
Auction uses Google’s ranking system	OpenAI documents a relevance-weighted second-price auction
Advertiser sees search-term/reporting data according to platform features	Advertiser does not receive the user’s private ChatGPT transcript

The practical implication:

Do not simply duplicate a Google Ads keyword structure into ChatGPT Ads and call the migration complete.

Reuse the business intelligence.

Rebuild the campaign architecture for the new matching environment.

How Are ChatGPT Ads Different From Meta Ads?

Meta advertising often begins from the advertiser’s chosen audience, creative, and optimization event.

ChatGPT Ads begins with a live conversational environment in which the user is actively expressing a need, comparing options, researching, or trying to complete a task.

ChatGPT Ads still supports targeting and custom-audience controls, but conversational relevance is a central part of the system’s selection logic.¹³

So the useful question is not:

Is ChatGPT Ads search or social?

It has elements that resemble both, but the conversational context creates its own operating model.

What Does This Mean for Campaign Strategy?

Understanding the mechanics changes how you should build campaigns.

Structure around customer needs, not internal product taxonomy alone

If two products solve entirely different problems, they probably need separate ad groups.

Write context hints like use cases, not keyword dumps

Describe the actual situation in which the offer becomes helpful.

Build multiple genuinely distinct ads

OpenAI recommends creative coverage with multiple differentiated messages rather than repetitive variations.⁸

Send people to a page that matches the conversation

The landing page is part of the relevance environment and the conversion path.

Make measurement part of the launch, not an afterthought

Without conversion measurement, you can see delivery and click metrics but have a weaker view of downstream business outcomes.

Bid competitively without assuming money can compensate for irrelevance

OpenAI’s auction is explicitly relevance weighted.²

A Simple Example

Imagine two advertisers are eligible for a conversation about:

software for a 12-person plumbing company that needs scheduling, invoicing, and field technician dispatch.

Advertiser A

Context hint:

business software

Ad:

Grow Your Business

Landing page:

generic software-company homepage

Advertiser B

Context hint:

scheduling and invoicing software for small plumbing and HVAC field-service teams

Ad:

Run Your Plumbing Crew

Landing page:

field-service software for plumbing contractors

It would be wrong to claim that Advertiser B is guaranteed to win.

OpenAI does not publish a formula that lets us calculate that outcome.

But Advertiser B has created much clearer advertiser-side relevance signals:

customer situation
→ context hint
→ ad message
→ landing page

That is the type of alignment the documented system is designed to evaluate alongside bid and other eligible signals.²⁷

Can Advertisers See What Users Said Before the Ad Appeared?

No.

OpenAI says advertisers do not receive users’ chats, chat histories, memories, names, email addresses, precise locations, IP addresses, or sensitive information.¹

This is one of the most important misconceptions to correct.

The system may use conversational context internally to select a relevant ad.

The advertiser does not receive a transcript saying:

Here is everything this person told ChatGPT before clicking your ad.

Those are fundamentally different things.

Can Advertisers Pay More to Change ChatGPT’s Answer?

No.

OpenAI says advertisers cannot shape, rank, or alter ChatGPT’s responses through the advertising system.¹

The auction determines sponsored placement.

It does not auction control of the answer itself.

How Should Advertisers Measure Whether ChatGPT Ads Work?

Do not stop at CTR.

Ads Manager can measure:

• delivery;
• clicks;
• spend;
• CPC/CPM;
• attributed conversions.⁶

But the business should connect those metrics farther downstream.

For lead generation:

impression
→ click
→ lead
→ qualified lead
→ meeting
→ opportunity
→ revenue

For ecommerce:

impression
→ click
→ product engagement
→ checkout
→ purchase
→ revenue
→ margin

The platform tells you what happened inside the ad system.

Your analytics, CRM, and business systems should tell you whether that activity created economic value.

Should You Run ChatGPT Ads Yourself?

If you already operate paid media and have reliable conversion measurement, understanding the mechanics above gives you a reasonable foundation for building your own test.

Our How to Advertise on ChatGPT guide covers the launch sequence step by step.

If you would rather have WKND structure, launch, measure, and optimize the channel, see our ChatGPT Ads management service.

The key is to treat ChatGPT Ads as its own advertising system-not as a copied Google campaign with a new logo.

Frequently Asked Questions

How do ChatGPT Ads work?

ChatGPT Ads evaluate whether advertising is eligible and relevant to a user’s conversation, match advertiser signals such as context hints, creative, landing pages, targeting, and bids, rank eligible ads through a relevance-weighted auction, and show sponsored placements below the ChatGPT response.¹²

What are ChatGPT Ads?

ChatGPT Ads are paid sponsored placements delivered separately from ChatGPT’s organic answer. Current ad units can include the advertiser name, favicon, headline, description, image, and landing page.¹²

Do ChatGPT Ads influence ChatGPT’s answers?

No. OpenAI says ads run separately from the model’s answers and advertisers cannot pay to shape, rank, or alter those responses.¹

How are ChatGPT Ads ranked?

OpenAI says eligible ads are ranked using factors including relevance and advertiser bids. Its advertiser documentation describes the auction as a relevance-weighted, second-price auction.¹²

Does ChatGPT Ads use keywords?

Context hints can describe conversations, topics, and relevant language, but OpenAI explicitly says they are not exact-match controls and do not guarantee delivery for specific words or conversations.²⁷

Can ChatGPT Ads target audiences?

Yes. Current Ads Manager documentation supports custom-audience inclusion and exclusion at the campaign level, alongside geographic and other supported targeting controls.⁹³

Can advertisers see ChatGPT conversations?

No. OpenAI says advertisers do not receive users’ chats, chat history, memories, names, emails, precise locations, IP addresses, or sensitive information.¹

Where do ChatGPT Ads appear?

OpenAI says sponsored placements appear below ChatGPT responses and are clearly labeled and visually separated from the answer.¹²

Is there only one ChatGPT ad per response?

Do not assume that. OpenAI’s current documentation says eligible users may see one or more ad units below a response, and an ad unit can contain one or more items from an advertiser or, in some cases, multiple advertisers.¹

How are ChatGPT Ads conversions tracked?

Advertisers can use the OpenAI Pixel, Conversions API, or both. OpenAI connects eligible conversion events to qualifying ad interactions using supported measurement signals and the campaign’s attribution settings.⁵

---

Sources