How ChatGPT Ads Work: First Insights and Takeaways on Advertising in ChatGPT 

Ads in ChatGPT are just beginning to emerge and may change the way we think about digital marketing. User intent and real-time conversational context are becoming more important than keywords alone. This is a new model for reaching audiences that could completely redefine performance marketing.

ChatGPT Ads are no longer just a preview of another performance marketing channel. OpenAI is developing its own Ads Manager, enabling the creation of standard and product campaigns, conversion tracking, and ad targeting based on the context of the user’s conversation.

And this last element is the most interesting here.

ChatGPT Ads don’t work like another version of Google Search Ads. You don’t simply choose a set of keywords you want to show up for. The system takes into account the intent and context of the current conversation, the content of the landing page, the ad copy, and additional signals provided by the advertiser.

This represents a shift in how we think about campaigns. Instead of focusing solely on what the user will type, the more important question becomes:

at what moment in the conversation can my offer genuinely be useful to them?

In this article, I show what ChatGPT Ads currently look like from an advertiser’s perspective, what campaign types are available, and what insights can already be drawn from the first configurations in Ads Manager.

• ChatGPT Ads use conversational context. When selecting ads, the system may take into account user intent, the current conversation content, the landing page, the ad itself, and Context Hints.
• Both standard campaigns and product-feed-based campaigns are available. In retail use cases, the system can select specific products from an uploaded catalog.
• Context Hints are not traditional keywords. They help the system understand in which types of conversations an ad may be relevant, but they do not function like exact match targeting.
• Conversions can be measured via the OpenAI Pixel and the Conversions API. OpenAI recommends using CAPI whenever possible.
• In product campaigns, feed quality is critical. Ad titles and descriptions are pulled directly from product data.

ChatGPT Ads is an advertising system developed by OpenAI that enables the display of paid ads within ChatGPT.

Ads are separated from model responses and clearly marked as sponsored. The presence of an ad does not influence the content generated by ChatGPT itself – OpenAI states that the advertising system operates independently from the model that produces responses.

A standard ad may include:

• advertiser name,
• favicon,
• headline,
• description,
• destination URL,
• image.

At first glance, this feels familiar. Title, text, image, URL – nothing particularly groundbreaking.

The real difference appears in how users are selected and when the ad is served.

This is where things get most interesting.

In traditional Search Ads, the primary signal is the query entered into the search engine. We can build structure around keywords, match types, user queries, or audience lists.

In ChatGPT, the user often does not enter a single phrase.

They may have a long conversation, for example:

I’m looking for a solution for a small apartment

then:

I also need the option to host guests

and later:

show me something high quality under £2,000

For the ad system, the entire conversation can provide signals about user intent.

OpenAI indicates that when selecting ads, it may analyze context and intent of the current conversation, the landing page, the ad title, ad copy, and contextual hints provided by the advertiser.

Because of this, ChatGPT Ads is much closer to intent-based targeting than traditional keyword bidding.

In standard campaigns, one of the most important configuration elements is Context Hints.

They can be used to describe:

• conversations where the offer may be relevant,
• specific user needs,
• relevant topics,
• products,
• purchase situations,
• phrases and expressions related to the offer.

However, OpenAI emphasizes that Context Hints are guidance for the system, not strict matching rules.

This is a very important distinction.

Entering:

electric sofa

does not mean:

show my ad to everyone who types exactly “electric sofa”

Similarly, entering:

people comparing electric sofas before buying

does not mean the system will look for conversations containing that exact string of words.

Instead, the hint helps the model understand the broader intent space in which the ad may be considered relevant.

At this stage, it makes sense to combine two approaches to describing intent.

The first involves short expressions close to traditional commercial queries:

• product + category,
• product + location,
• “best…”,
• “buy…”,
• specific product attributes.

The second is more natural descriptions of needs and scenarios:

• a user comparing multiple solutions before purchase,
• a user looking for a product with a specific feature,
• a user working within a defined budget,
• someone trying to solve a specific problem.

However, the most important factor is consistency within a single ad group.

If an ad group is focused on one product or category, it is not advisable to add loosely related topics just to increase potential reach.

In ChatGPT Ads, broader does not necessarily mean better.

This is another key difference compared to the simple “choose an audience and set CPC” approach to advertising.

OpenAI uses a relevance-weighted auction model. When selecting an ad, the system considers not only the advertiser’s bid, but also predicted relevance and expected outcome.

In practice, it is important to ensure that all campaign elements communicate the same message:

Context Hints → ad → landing page

If Context Hints describe a very specific product category, but the ad is generic and the user lands on a homepage, the system receives mixed signals.

A much more coherent structure would be:

specific need → specific ad → specific category or offer

This is likely to be one of the key factors differentiating well-structured ChatGPT Ads campaigns from setups built on a “put everything into one ad group and see what happens” approach.

From the advertiser’s perspective, two approaches are particularly interesting:

Standard campaigns

Here, the advertiser prepares the creative themselves.

The core elements include:

• headline,
• description,
• image,
• landing page.

This approach provides greater control over the message, making it well-suited for promoting specific services, product categories, USPs, or more lifestyle-oriented product benefits.

Targeting can additionally be supported by the previously mentioned Context Hints.

Product Feed campaigns

The second option is product campaigns.

The advertiser uploads a product catalog to the system, and OpenAI can generate ads based on the data in the feed and match relevant products to the user. 

This leads to a very important takeaway for e-commerce.

People working with Google Shopping are very familiar with the importance of the feed.

In ChatGPT Ads, this can be even more interesting.

OpenAI states that in feed-based campaigns, the ad title and description are taken directly from the product feed.

This means it is worth moving away from treating the feed as purely a technical file containing:

brand + name + variant + size

A well-described product should also communicate information that helps understand:

• what the product is,
• its most important features,
• who it may be suitable for,
• what problems it solves,
• how its variants differ,
• in what situations it can be useful.

A product feed can include not only basic data such as price or availability, but also additional information and metadata that support working with the catalog.

Existing data can be used as a starting point.

However, this does not mean that a feed prepared years ago for Google Shopping is automatically the best feed for ChatGPT Ads.

In a classic feed, we often see titles created primarily around keywords searched on Google.

In the case of an ad appearing within a conversational context, it may be important to provide the system with clearer information about the product itself and its use case.

Therefore, before launching a product campaign, it is worth primarily reviewing:

title, description, categories, variants, and additional product attributes

Targeting is only half of the equation.

The other half answers a much simpler question:

did the ad actually deliver a business result?

OpenAI offers two main ways of sending events.

OpenAI Pixel

The OpenAI Ads Measurement Pixel is a browser-based script that allows tracking actions taken by users after clicking an ad.

It can be used to track purchases, form submissions, or registrations, among others.

The pixel can be implemented via the website or a tag management system – in practice, Google Tag Manager will be the most convenient option for many companies.

Conversions API

The second option is the Conversions API.

Here, event data is sent server-to-server.

OpenAI indicates that CAPI is a more reliable data source than relying solely on the pixel and recommends using it whenever possible.

Both solutions can also be used simultaneously.

Tracking first, ads second

This is a basic principle that is still easy to overlook.

If the business goal is a purchase, lead, or registration, it is worth configuring the appropriate event before the campaign starts spending budget.

Otherwise, after the first week, you will only know how much traffic cost.

And that is not the point of performance marketing.

ChatGPT Ads currently supports both CPM and CPC models.

For campaigns focused on reach, you can use the Reach objective with cost per thousand impressions billing.

For traffic-focused campaigns, the Clicks objective is available, with a maximum CPC bid set at the ad group level. OpenAI currently indicates $3–5 as a general starting point for CPC, although Ads Manager may also provide its own assessment of how competitive a given bid is.

The platform also supports optimization using measured conversions, so properly configured tracking can be important not only for reporting but also for later campaign optimization.

Ads Manager Beta currently allows analysis of, among others: 

Data can be analyzed at campaign, ad group, and ad level, and exported to CSV.

At this stage, however, I would approach any kind of benchmarking with caution.

ChatGPT Ads is still a beta product, the list of available markets is changing, and OpenAI regularly updates the Ads Manager and its documentation.

Far more valuable will be collecting your own data for a specific business.

After working with the current system, a few things stand out as particularly important.

1. Don’t copy Google Ads structure 1:1

ChatGPT Ads operates in a different environment.

The user does not necessarily enter a single query – they often have a conversation, gradually refining their need.

Because of this, you need to think not only in keywords, but primarily in intent and conversation scenarios.

2. With a small budget, it’s better to start narrow

One well-defined category or intent provides more control than a very broad set of topics.

If delivery is too limited, the scope can always be expanded later.

It is much harder to evaluate early results from a campaign that, from the start, tries to reach users interested in everything.

3. Context Hints should support the ad, not replace it

A great set of hints won’t help if the headline says something different and the landing page leads to a homepage.

It is best to think of the entire setup as one coherent message being delivered to the system.

4. In e-commerce, a lot of work starts before Ads Manager

The product needs to be well described.

In feed-based campaigns, data quality is not an aesthetic issue – it directly influences the material the system uses to build ads. OpenAI indicates that feed-based ads have historically been among the best-performing formats in the program.

5. Tracking should be treated as seriously as in Google or Meta

Pixel, Conversions API, proper events, and correct attribution should all be part of the pre-launch checklist.

Clicks alone do not yet tell you anything meaningful about channel value.

For Polish advertisers, the answer is: yes, but without artificial FOMO.

The Ads Manager is not currently available in Poland. OpenAI is rolling out the system in stages and expanding availability to additional countries.

However, this is still a good moment to understand how the platform works.

Especially since preparing for ChatGPT Ads does not have to mean launching campaigns tomorrow.

For e-commerce, it may involve:

• cleaning up the product feed,
• improving product descriptions,
• preparing analytics,
• reconsidering product categories and customer intents.

For service-based businesses:

• mapping key customer problems,
• defining stages of the decision-making process,
• preparing dedicated landing pages,
• thinking about in what kind of conversation the company’s ad would actually be helpful.

And this is probably the most important change ChatGPT Ads introduces today.

For years, marketers learned to answer the question:

“What does the user type into the search engine?”

Now it is worth starting to ask a second one:

“What is the user talking about before making a decision?”