ChatGPT Ads moved from an (essentially) invite-only experiment to an open marketplace in a matter of weeks. Here's what changed, what's hype, and where we're telling clients to actually test.
A few months ago, running an ad inside ChatGPT required a $200K commitment, a managed-service onboarding, and a polite conversation with OpenAI’s sales team. But now, as of this month, any U.S. business with a credit card can log into ChatGPT’s self-serve Ads Manager and launch a campaign with whatever they want to spend.
That meaningful change opens up ads within ChatGPT’s Free and Go Tiers, which are the bulk of its weekly active base. OpenAI has stated publicly that it aims for $2.5B in ad revenue this year, on its way to a projected $100B by 2030. Whether or not those numbers land, the channel is no longer hypothetical. ChatGPT Ads are buyable, loosely measurable, and show early performance signals worth paying attention to.
There are two distinct ways in, built for different brands solving different problems. Brands shouldn’t question whether to test but instead ask: which door should we walk through, with what budget, and with what aim?
DOOR 1: AD TECH PARTNERS—THE COMMERCE-LED PATH
The partner path is built for the part of ChatGPT where shopping intent shows up most clearly. When a user asks ChatGPT for product recommendations, gift ideas, or “what should I buy for X,” partner APIs surface sponsored products pulled from retailer catalogs.
- Buying model: Performance commerce: the home turf of partners like Criteo and Pacvue. Campaigns are built around product feeds and optimized toward clicks and conversions, managed through your brand’s chosen partner platform alongside existing retail media spend. Partners approach this differently by channel strength: Pacvue is built for retail media and cross-channel performance (Amazon, Walmart, etc.); StackAdapt extends programmatic buys across CTV, native, and display into ChatGPT; Adobe plugs into Experience Cloud workflows; Kargo adds a premium/creative-programmatic lens. The right partner depends on where ChatGPT fits within your existing channel mix.
- Ad format: Sponsored products. Ads are auto-generated from the same product catalog file that retailers already send to Google Shopping (title, image, attributes), meaning there is no separate creative build. Up to 1M SKUs per advertiser; onboarding typically starts with a sample of around 100.
- Targeting: Contextual, powered by conversation intent; for Criteo specifically, layered with its commerce graph—roughly $1T in annual ecommerce transactions and around 5B SKUs of behavioral signal. Matching is based on conversation intent, not keywords.
- Minimums: Pilot commitments were initially pitched at $50K–$100K, but minimums have since decreased to $25K, meaningfully widening the pool of brands that can run a credible test without overcommitting incremental budget.
- Geographic availability: U.S., plus Australia, Canada, and New Zealand currently available via Criteo; other partners vary. OpenAI also confirmed broader geographic expansion of the ads pilot to the U.K., Mexico, Japan, Brazil, and South Korea; availability through individual partners will vary.
- Best for: Retailers and DTC brands with a working product feed, especially in the categories that early partner data has flagged as outperformers: consumer electronics, lifestyle and wellbeing, and home and garden. In these categories, AI-referred conversion rates are running close to 2x traditional search and CTRs roughly 3x comparable formats in other environments.
DOOR 2: OPENAI ADS MANAGER: THE SELF-SERVE PATH
OpenAI’s own Ads Manager became broadly available on May 5th. The interface will feel familiar to anyone who has run paid search (campaign, ad group, ad setup) and it is open to any U.S. business with no minimum spend.
- Buying model: Two campaign objectives. Reach (CPM) and Clicks (CPC). OpenAI recommends a starting max bid of $3–$5 per click and defaults CPM max bids to $60. The move from CPM-only at launch to a CPC option is the bigger story than the dollar threshold—it pulls ChatGPT ads closer to how paid search teams actually plan demand.
- Ad format: A single unit called a “chat_card”—a 3–50 character title, up to 100 characters of body copy, an image, a favicon, and a destination URL. No video, carousel, or rich interactive unit in self-serve yet, though OpenAI has signaled more formats are coming. OpenAI has also added a “product feed” campaign type that lets merchants auto-generate ads directly from their existing product catalog—using the same structured file they already send to Google Shopping. Retailers set filters for eligible SKUs and the platform handles the rest, generating ads from product names, images, attributes, and pricing automatically. The system supports up to 1M SKUs per advertiser; new partners start with a 100-product sample feed. Ads land below the AI response, clearly labeled as sponsored. By default, clicks send shoppers to the merchant’s own site or app—supporting both DTC and drive-to-retailer destination URLs.
- Targeting: Contextual, not keyword-based. Advertisers provide “context hints”— broad descriptions of the questions or situations a relevant user might bring to ChatGPT—and the system matches against conversation topics, prior chat history, and previous ad interactions. Budgets, dates, and country live at the campaign level; bids and context hints at the ad-group level.
- Measurement: Reporting covers impressions, clicks, spend, CTR, average CPC, average CPM, and conversions. Functional, but thin compared to what most teams are used to in Google or Meta. What’s more, the attribution story is still being written.
- Best for: Brand and consideration-stage testing, lead-gen advertisers without a product feed, and any team that wants to understand how its messaging behaves inside an AI conversation before the channel matures and competition compresses returns. This path is also increasingly relevant for ecommerce brands that want self-serve, no-minimum access to feed-based shopping ads—particularly those already using StackAdapt, where catalogs will automatically flow to OpenAI once supported.
OUR POV: TEST, BUT TEST ON PURPOSE
The first-mover narrative around ChatGPT Ads is real, but it is also oversold. The early performance numbers that Criteo has shared—2x conversion rates and 3x CTRs—come from a tightly curated pilot in specific verticals, with a small set of well-funded brands. Treat them as a directional signal, not a benchmark.
Here is what we are telling Stella Rising clients:
- If you are a retailer or ecommerce brand with a clean product feed, feed-based shopping ads are now available through both doors. The question is, which path fits your setup?
- The ad tech partner path (Criteo, Pacvue, and others) is the stronger choice if your brand already runs retail media through these platforms, wants the benefit of Criteo’s commerce graph optimization, or is in the early-outperformer categories (consumer electronics, lifestyle, home). Minimums start at $25K and brands can fund the test out of incremental retail media budget.
- The OpenAI self-serve path offers no minimum, is fully self-managed, and has a product feed campaign type that now lets brands auto-generate ads from the same catalog they already send to Google Shopping. Self-serve is the right entry point if you want to test the format without a partner commitment or existing platform relationship.
- If you are a brand or lead-gen advertiser, the OpenAI self-serve door is the lower-stakes way in. With no minimum and CPC bidding live, you can run a controlled test, with clear context hints, a tight geo, a single landing page, and learn how ChatGPT users respond to your brand message before the auction gets crowded. We treat this as discovery spend, not a performance line item, until measurement matures.
- We are not recommending that clients shift meaningful budget out of Google or Meta. The data is too new, the formats too thin, and the measurement story still under construction. The right frame is incremental discovery: small, instrumented, and learning-led.
WHAT WE’RE WATCHING
A few things will tell us how seriously to take this channel over the next two quarters:
- Whether OpenAI ships richer self-serve formats (video, multi-card, interactive). Product feed ads have already shipped; the remaining format gaps are video and interactive units
- If third-party measurement and attribution catch up to the inventory. Note that cost-per-action (CPA) bidding went live inside ChatGPT this week, in the first real sign the platform is moving into performance advertising; early access will be offered to accounts with conversion tracking set up by June 1
- If CPCs hold near the $3–$5 recommended range or compress as auction density grows
- Whether ChatGPT’s Plus and Pro tiers ever open to advertising at all
ChatGPT ads are no longer a thought experiment, but buyable through two doors, with two different buying logics, built for two different kinds of advertisers. The right move is to walk through the one that fits your business—with clarity around what you will and will not learn.
If you are weighting a ChatGPT Ads test for your brand, our paid search team can talk you through a structured pilot—including budget, hypotheses, and a measurement plan—addressing how the test will fit alongside the rest of your demand mix.
Note: This post, originally published on May 27th, was updated on May 29th to reflect changes to the partner path.
COMMENTS