Conversational AI just crossed a line that separates a pilot from a real ad platform. OpenAI has quietly rebuilt ChatGPT Ads around conversion-based bidding, and Google is testing its own version of the same idea inside AI Mode. The story is not that a chatbot sells ads. It is that the chat window is becoming an auction, with the same optimization logic marketers already trust in Google and Meta.
The Bid Just Got Smarter
Until now, ChatGPT Ads ran mostly on cost-per-click delivery: pay for attention, hope it converts. OpenAI’s new conversion-optimized campaigns change the mechanics. Advertisers set a maximum cost-per-action bid on a single standard conversion event, such as a completed order, a generated lead, or a finished registration, and OpenAI’s auction optimizes delivery toward the clicks most likely to produce that outcome. Billing still happens per valid click, so the CPA bid functions as a targeting signal rather than a direct charge.
The catch is that only standard events qualify. Custom event definitions are not supported as optimization goals, and an account needs a working JavaScript Pixel or Conversions API connection, OpenAI notes the latter is the more reliable tracking source, before a campaign can even launch. Each campaign locks to one conversion event at creation and cannot switch later. A separate, beta-only version of conversion optimization applies to product-feed campaigns, gated to accounts with beta access.
Why This Had to Happen
The pressure behind this update is visible in the platform’s own pricing history. ChatGPT Ads opened to logged-in Free and Go-tier users in the United States on February 9, 2026, and by late March had reached a $100 million annualized revenue run rate as OpenAI expanded into Canada, Australia, and New Zealand. But CPMs fell from around $60 at launch to as low as $25 by April, largely because advertisers had no conversion proof to justify premium pricing. A chat surface with no way to prove downstream value is a novelty buy, not a budget line. Conversion-optimized bidding is the fix: it gives ChatGPT Ads the same “pay for outcomes” credibility that already anchors budgets in Google and Meta.
The platform now covers seven markets, including the U.S., Canada, Australia, New Zealand, the U.K., Japan, and South Korea, with Mexico and Brazil described as coming soon. AppsFlyer and Adjust integrations, which shipped alongside the bidding changes, let app advertisers attribute installs and in-app events back to ChatGPT placements using the same measurement partners and dashboards they already use for Meta, Google, and TikTok. That is the real unlock: advertisers can finally compare ChatGPT against established channels using shared attribution rules instead of taking performance on faith.
Not Just OpenAI
Google is building toward the same destination from the search side. Its Gemini-powered ad formats for AI Mode and Search, unveiled at Google Marketing Live, include Business Agent for Leads, which embeds a conversational AI agent directly in an ad so a prospect can ask questions and get answers pulled from the advertiser’s own site instead of filling out a static form. Conversational Discovery ads use Gemini to generate creative tailored to the exact query, and AI-powered Shopping ads pair product listings with an AI-written explanation of why a given item fits the search. None of these formats are click-and-hope; all are built to be measured and optimized like the auction products that fund the rest of Google’s ad business.
Put OpenAI’s and Google’s moves side by side and a pattern emerges: the interface layer of advertising, not just the inventory, is being rebuilt around agents and conversation, and every major platform is racing to make that layer auction-native rather than experimental. Marketing teams that have spent the past year treating AI agents inside marketing platforms as a governance problem now need to treat them as a media-buying one too.
What This Means for the Marketing Leader
Conversion-optimized bidding only pays off if the tracking underneath it is already solid. Teams that have not wired a Conversions API integration, not just a pixel, will not be eligible for the more reliable signal path, and a campaign’s single locked conversion event means the first test needs to target the highest-confidence outcome in the funnel, not the broadest one. Budget owners should treat an initial ChatGPT Ads conversion test as they would a new channel test on any auction platform: a fixed, small tranche, one clear event, and a direct CPA comparison against the same event on Google and Meta rather than against ChatGPT’s own history. If CPMs stabilize or climb as conversion data accumulates, that is the signal the fix is working. If they do not, the trust problem sits somewhere other than measurement.
The Trust Problem Still Sits Underneath
Better bidding mechanics do not resolve the skepticism that met ChatGPT Ads at launch. Reaction to OpenAI’s original ads announcement ran heavily negative on social platforms, driven by users who saw the product as a paid intrusion into a space they treated as a clean, ad-free assistant. Brand safety carries a different shape here too: an ad now sits next to a generative response that can still be wrong, which is a harder problem than an ad sitting next to a static page. Regulators have already flagged the category. Senator Ed Markey sent letters on January 22, 2026, to OpenAI and six other AI and platform companies, warning that ads could be woven into conversational responses closely enough to be mistaken for the chatbot’s own answer, and asking specifically how each company would keep advertising away from sensitive conversations about health, mental health, or politics. OpenAI faced a February 12 deadline to respond. Conversion-optimized bidding makes the ad mechanics look more like Google and Meta. It does not make the trust questions go away, and measurement across AI answer surfaces remains unsettled well beyond just the ad layer.
What to Do Now
Marketing leaders piloting ChatGPT Ads should move the Conversions API to the top of the setup checklist, ahead of creative or targeting work, since it gates access to the better bidding path entirely. Pick one conversion event, the one closest to revenue, for the first campaign rather than trying to optimize the whole funnel at once, since that choice cannot be changed mid-flight. Run the test against a fixed budget and compare CPA directly to the same event on an established platform, not to ChatGPT’s own early numbers. And watch the regulatory track alongside the performance numbers: a platform that solves its measurement problem before it solves its trust problem is still carrying risk that will not show up in a CPA report.
Source: OpenAI
