ChatGPT Ads, Explained: The 2026 Reference

ChatGPT ads are sponsored placements that appear inside ChatGPT's answer flow, shown to users as part of OpenAI's response UI. OpenAI began rolling the product out in February 2026 after years of public hedging on the question. This is the reference document for how ChatGPT ads work, what they cost, who is buying them, and what brands should plan for if they want to reach ChatGPT's roughly 700 million weekly users.

The shape of the product matters because ChatGPT is now the single largest AI interface by user count. Even a modest change to how ads appear inside that surface moves budgets at scale. What follows is what we know, what we can reasonably infer, and what is still unknown as of April 2026.

What ChatGPT Ads Shipped in February 2026

OpenAI announced the ad product on February 6, 2026, in a blog post signed by Sam Altman and Fidji Simo. The launch was narrow. Ads went live first in the United States, on the free tier of ChatGPT only, for a limited set of query categories: shopping, travel, software, and financial services. The announcement confirmed several things the industry had been speculating about since 2024.

First, the placement is inside the answer, not alongside it. ChatGPT does not show a sidebar, a top banner, or a sponsored link carousel the way Google Search does. Instead, sponsored recommendations are woven into the model's response when the query has commercial intent. A user asking "what's the best running shoe for plantar fasciitis" may now see a branded product named inside the answer, labeled as sponsored.

Second, ads are opt-out for Plus, Pro, and Enterprise subscribers. Paying users do not see them by default, though OpenAI has said this may change. Free users see ads on an estimated subset of queries — OpenAI has not published the exact share, but early third-party testing suggests ads appear on somewhere between 8 and 15 percent of commercial-intent queries in the covered categories, per public reporting from The Information and Platformer in March 2026.

Third, OpenAI is running the auction directly. There is no third-party ad server, no Google Ad Manager integration, no IAB-compliant bidstream. Advertisers buy through a new interface at ads.openai.com (access is gated and invite-only as of April 2026), and OpenAI handles targeting, auction, and delivery in-house. This matters because it means no third-party measurement, no industry-standard verification, and limited interoperability with existing ad stacks.

The product shipped with four format types, per OpenAI's launch materials: named product mentions inside answers, linked citations with a sponsored label, comparison-table sponsored rows, and follow-up suggested prompts that lead to a branded destination. All four are text-based. There are no image ads, no video ads, and no interactive units in the initial release.

If you want a primer on how ChatGPT advertising sits inside the broader landscape, the guide at /blog/how-to-advertise-on-chatgpt covers the options available before and after OpenAI shipped its own product.

The Format: Sponsored Results Inside Answers

The defining characteristic of ChatGPT ads is that they appear inside the response, not next to it. This is a break from every dominant ad format of the last thirty years. Search ads sit above organic results. Display ads sit next to content. Social feed ads sit between posts. ChatGPT ads sit inside the text of an answer the user asked for.

The UI looks like this. A user sends a prompt. ChatGPT generates an answer. Inside that answer, a product name may appear with a small sponsored label (the exact glyph has varied during rollout, but as of April 2026 it is a subtle "Sponsored" tag on hover plus an italic S mark next to the product name). Tapping the product name opens a card with a brief description, a link to the advertiser's site, and a disclosure line.

For comparison-style queries, ChatGPT may show a table with sponsored rows highlighted. A query like "compare Notion, Obsidian, and Roam" could return a three-row table with a fourth sponsored row added — a tool that paid for inclusion in that category comparison. The sponsored row is labeled, but it sits in the same table as the organic results, which is a deliberate choice and a contested one.

OpenAI has said publicly that it refuses to let sponsorship influence the ranking of organic results. A sponsored product cannot buy its way into being recommended as the top choice. The sponsored placements are additive, not substitutive. Whether this holds up under commercial pressure over the next two years is one of the open questions advertisers should watch.

The format most advertisers are buying first is the follow-up suggested prompt. After a ChatGPT answer, the UI shows two or three suggested follow-up questions the user might want to ask next. One of those can now be sponsored — a prompt like "Compare this to [Brand X]" that the advertiser has paid to place. When tapped, the follow-up prompt runs as a real query with the brand name seeded, which tends to surface the brand prominently in the next answer.

Pricing and Auction Mechanics

Pricing details are not fully public. OpenAI has not released a rate card. What follows is based on reporting from Ad Age, The Information, and Digiday between February and April 2026, plus conversations with advertisers who have run campaigns. All figures should be read as reportedly, not confirmed.

The auction is reportedly a second-price auction, similar in structure to Google's. Bids are set per impression, not per click, though OpenAI also reports click data. Minimum bids reportedly start around $2 CPM in low-competition categories and climb to $40+ CPM in financial services and software, per Digiday's March 2026 reporting. Effective CPCs land in the $8 to $35 range depending on category, which is higher than Google Search in most verticals.

A few structural notes from public reporting. Budgets are set daily, like most programmatic channels. Targeting is reportedly limited in the initial release — advertisers can select category, region (US-only at launch), and a rough intent tier, but there is no demographic targeting, no retargeting, and no custom audience upload. This is a deliberate privacy stance by OpenAI, though it also reflects the fact that OpenAI has less deterministic user data than Meta or Google.

Measurement is the weakest part of the product. OpenAI reports impressions, clicks, and landing-page arrivals. It does not yet support third-party pixel measurement (Innovid, DoubleVerify, IAS are not integrated as of April 2026) and conversion tracking is limited to OpenAI's own pixel, which advertisers install on their destination pages. Attribution windows are short — seven days click, one day view — and cross-device tracking is limited to signed-in ChatGPT users.

For brands accustomed to a Google Ads or Meta level of measurement fidelity, this is a step down. For brands used to TV or podcast measurement, it is a step up. Where exactly ChatGPT ads sit on the measurement sophistication curve is still being worked out.

Who's Buying, Who's Not

The initial cohort of advertisers on ChatGPT ads skews heavily toward direct-to-consumer brands, software companies, and travel. Per reporting from Ad Age in March 2026, the launch partners included names across those verticals plus a handful of financial services firms (lenders, brokerages, insurance).

What is notably absent is the largest category of Google Search advertisers: local services. Plumbers, dentists, attorneys, contractors. ChatGPT's targeting granularity does not yet support local-intent advertising at the ZIP-code level, which rules out most local services buyers. OpenAI has said local targeting is on the roadmap but has not committed to a date.

Also absent, at least publicly, is the pharma category. ChatGPT ads reportedly cannot be used for prescription drug promotion at launch, which is a category that spends north of $8 billion annually on search and display. Pharma's regulatory requirements around fair balance and disclosure are hard to satisfy in a text-response format where answers vary from session to session.

The brands that are buying share a few characteristics. They have strong commercial intent queries (people searching "best [X]" type phrases), they have short decision cycles, and they can afford to test a channel with weak measurement. This is the profile of an early adopter — brands that lean experimental and have budget for channels that haven't proven themselves on ROI yet.

The brands that are not buying, as far as we can tell, fall into two groups. Large CPG advertisers (Procter & Gamble, Unilever) who are accustomed to reach-based TV and programmatic display, and for whom the ChatGPT format doesn't yet support the volume or measurement they need. And performance brands with tight CAC targets who are holding off until conversion tracking improves.

How It Compares to Third-Party Chatbot Ad Networks

Third-party ad networks for AI chatbots have been running for over a year — the space Surfacedd operates in. The distinction between OpenAI's first-party ads and third-party networks matters because they serve different use cases.

OpenAI's ads are available only inside ChatGPT. If you are a brand and your goal is to reach ChatGPT users specifically, you buy through ads.openai.com. If you are a developer building a chatbot, agent, or assistant on top of OpenAI's API (or on top of Anthropic, Google, or an open model), you cannot use OpenAI's ad product — it does not serve ads into third-party apps.

Third-party networks serve the other side of the market. They place ads into ChatGPT-wrapper apps, Claude-based agents, Perplexity-alternatives, and the long tail of AI products built by developers. A brand that wants to reach AI-native users broadly — not just inside ChatGPT — uses a third-party network to access the long tail of surfaces. See /ai-ad-network for how this works on the buy side.

The pricing is different. Third-party networks typically run on a CPC or CPA model with broader category targeting and higher fill rates. OpenAI's first-party ads run on CPM with tighter targeting and stricter brand-safety review. The two are complementary, not substitutes. Advertisers running integrated AI strategies in 2026 buy both.

Measurement is also different. Third-party networks support IAB VAST-equivalent conformance for conversation ads and integrate with standard attribution tools. OpenAI does not, at least not yet. For a brand whose entire marketing stack runs through Google Analytics 4 or Adobe Analytics with third-party pixels, the third-party network experience is closer to what they already use.

For a brand evaluating AI advertising more broadly, both are worth testing. The question is not which one to pick — it is how to allocate between them based on the audience shape you want and the measurement fidelity you need.

What to Plan for If You're a Brand

Five practical points for brands planning ChatGPT ad spend in 2026.

First, apply for access now even if you don't plan to spend for another quarter. Access to ads.openai.com is invite-only and the waitlist is reportedly long. Getting an account doesn't commit you to spending.

Second, start with the follow-up prompt format. It is the easiest to reason about, the easiest to measure (because it creates a clear user action), and the easiest to write creative for. A well-designed follow-up prompt acts like a soft call-to-action that surfaces your brand in a second answer.

Third, accept the measurement limitations for now. ChatGPT ads are a top-of-funnel brand-exposure channel today, not a last-click performance channel. If you treat it like Google Ads and demand a seven-day click-through ROAS, you will be disappointed. If you treat it like a podcast ad or a connected TV placement and measure incremental lift, you will have a more honest read.

Fourth, build your creative for a text-first surface. ChatGPT ads don't support images or video. Your brand has to land in a single product name, a short description, and maybe a comparison row. This requires creative thinking about how your brand reads in text alone — a skill most performance creative teams don't have yet.

Fifth, set up direct measurement fallbacks. Install the OpenAI pixel, tag destination URLs with clear UTM parameters, and run regular geo-holdout or Ghost Ads-style lift tests if budget allows. The measurement deficit in the channel means you need to build your own truth source for what's working.

Brands that want to reach ChatGPT users in a way that layers on top of OpenAI's first-party product can also work with third-party networks that serve ads into the broader AI surface. See /advertisers/reach-chatgpt-users for details on that approach.

ChatGPT ads shipped smaller and more carefully than many observers expected. The product is narrow, the measurement is thin, and the targeting is rough. But OpenAI has shipped it, which means the next six to twelve months will produce the first real data on how users respond to in-answer commerce at the scale of hundreds of millions of queries per day. For brands, the smart move is to test now, keep spend proportional to the measurement quality, and pay attention to what OpenAI ships next.