Everything You Need to Know About ChatGPT Ads

A person researching a purchase can reveal a lot through a conversation: what they need, what they can afford, what they have already considered and what is holding them back. For advertisers, that context is what makes ChatGPT Ads worth paying attention to.

I’ve been looking more closely at ChatGPT Ads, particularly at what it would take for a smaller business to run a campaign. OpenAI began testing ads in February 2026 and rolled out its self-serve Ads Manager in May. The platform is still in beta, but advertisers now have more options for buying, targeting and measuring campaigns than they did at launch.

Here’s what that looks like in practice.

What are ChatGPT ads?

ChatGPT ads can appear below a response, labelled as sponsored and kept separate from the answer. A standard ad includes the advertiser’s name and logo, headline, copy, an image and a landing-page link. One thing worth making clear is that paying for an ad does not get your brand written into ChatGPT’s answer. OpenAI states that advertisers cannot influence, rank or change those responses. The ad is a paid placement, not an endorsement.

Retailers also have product-feed campaigns. These use catalogue information to create ads featuring product images, titles, ratings and prices. That feed makes products available for advertising; it does not automatically get them included in organic ChatGPT responses. I’d keep paid advertising and organic AI visibility as separate conversations when planning and reporting.

Who can see the ads?

Ads may appear for people using ChatGPT’s Free and Go plans. Plus, Pro, Business, Enterprise and Edu remain ad-free. Ads are not shown to accounts identified as belonging to people under 18, or in Temporary Chats. So, from a planning perspective, we should not use ChatGPT’s total user base as the audience we can reach through advertising. The addressable audience depends on the plan people use, their location and eligibility.

What about advertiser eligibility?

I would check category eligibility before briefing Creative or committing to a launch. OpenAI restricts certain advertising categories and sensitive placements. Its current policy allows specified financial advertisers in the U.S. on a case-by-case basis. Being approved as an advertiser also does not mean ads can appear alongside sensitive conversations. For regulated businesses, I would get that confirmation upfront. General access to Ads Manager is not enough to establish that a particular product or service can be advertised. For Canadian businesses, Canada is listed as available for self-service Ads Manager access, alongside the U.S., U.K. and other markets. Eligibility is based on where the Advertiser’s legal entity is located.

How can businesses buy ChatGPT ads?

Directly through Ads Manager. Eligible businesses can create an account, complete verification, add billing details and build campaigns. They can then give their agency access to help manage the activity.

Through advertising partners. OpenAI has named partners including Adobe, Criteo, Kargo, Pacvue and StackAdapt. These partners can support campaign management, but OpenAI’s advertising system still controls delivery.

Through tools they already use. HubSpot’s integration supports campaign creation, reporting and lead follow-up within the CRM. The Shopify app connects product catalogue data and measurement with ChatGPT Ads. For smaller teams already using these tools, these could be useful ways to get started, subject to availability and account eligibility.

Amazon Ads has also announced an integration. However, its September 10 announcement describes a pilot with selected U.S. advertisers, so I wouldn’t assume it is available to every Amazon advertiser or in every market.

When speaking with a partner, I’d ask them to separate the media requirement from their own fees and service minimums. The amount quoted for a managed campaign may be different from what it takes to start directly.

Is it accessible to small businesses?

Yes, eligible smaller businesses can participate. The minimum daily campaign budget for a CAD-billed account is currently $25. That is the minimum the platform allows you to set, not a recommendation for how much a useful test should cost. That distinction matters. Being able to launch at $25 a day does not tell us how many qualified enquiries or sales that budget will generate. There is also a pacing detail to be aware of, the daily budget is an average, not a hard daily cap. Spend can reach twice that amount on an individual day, within the applicable weekly limit. A campaign’s total budget gives you a cap on total campaign spending.

For a smaller business, I’d work backwards from what a customer is worth and what the business can afford to spend acquiring one. Then I’d decide whether the available budget gives the campaign a reasonable chance to produce enough results to assess.

What are you charged for?

For CPC campaigns, OpenAI suggests starting with a maximum bid of USD $3–$5 per click. I would treat that as bid guidance, not an expected CPC or industry benchmark.

How does targeting work?

The system considers the context and intent of the conversation, along with the ad, landing page and information supplied by the advertiser. One feature I find interesting is context hints. These let advertisers explain more about their product, who it helps and when it might be useful. Take a furniture business selling fold-away desks. It could describe the product as useful for someone setting up a workspace in a small apartment, where the room also needs to function as living space. That gives more useful context than a general list of keywords like “desk,” “office” and “furniture.” But I wouldn’t approach context hints like exact-match keywords. They guide relevance, but don’t guarantee that an ad appears in a particular conversation. There are two targeting details I’d check carefully:

Geography: New product-feed campaigns currently support geographic targeting and exclusions only at country level. Other campaigns can offer more granular locations, depending on the market.

Custom audiences: Inclusion targeting and audience-based bid adjustments require at least 25,000 matched users. Smaller audiences can be used for exclusions.

What creative do you need?

OpenAI’s guidance favours specific, benefit-led, direct messaging, with different creative variations leading to relevant landing pages. It also requires those pages to be accessible to its relevant crawlers. OpenAI has also introduced AI assistance for copy and imagery, with advertiser review before suggestions are added to campaigns. That could save production time, but I would still keep the usual creative QA in place. Claims, prices, images and the connection to the landing page all need checking.

Can you measure more than impressions and clicks?

Yes. Ads Manager reports impressions, clicks, spend, CTR, average CPC, average CPM and attributed conversions when measurement is set up. Conversion tracking can use the OpenAI Pixel, Conversions API, or both. Reporting supports 7, 14 or 30-day click-through windows. Reporting also needs time to catch up. OpenAI says spend can take approximately 7-8 hours to update, while attributed conversions can take 24-48 hours to appear. I wouldn’t make an optimization decision from an incomplete same day snapshot.

Where I land

What interests me is the possibility of reaching someone while they are actively researching and working through their decision making process. The broader access and lower entry budget make that easier for smaller businesses to explore. But I’d still approach it as a test. I’d want to understand the quality of the traffic, what people do after landing on the website and whether the resulting business justifies the spend.

With consumers increasingly using the platform to research products and services, the opportunity is difficult to ignore, but with limited benchmarks and industry learnings, the case for significant investment is still being built. For now, test, learn and let the results determine what role it deserves in the media mix.

Source: https://openai.com/