The door to ChatGPT ads opened by degrees, not all at once
a handful of brands
pilot + agencies
self-serve beta
ChatGPT ads are now something a normal marketing team can actually buy. On May 5, 2026, OpenAI began rolling out a beta self-serve Ads Manager, added cost-per-click bidding alongside its original CPM buying, and shipped a Conversions API and pixel so advertisers can finally see what happens after the click. Three months earlier this was a closed pilot with a six-figure minimum. That is a fast door to open, and the temptation to fund it by raiding an existing budget line is going to be strong.
Here is what actually changed, what a click inside ChatGPT is currently worth on the evidence available, what to test in your first cycle, and, the part most coverage skips, which budget you should not touch to pay for it. Every figure below carries a named, dated source. Nothing is modelled or estimated, and where two credible outlets disagree, we say so rather than picking the tidier number.
What changed with ChatGPT ads in May 2026?
Three things changed at once: buying went self-serve, bidding went click-based, and measurement got a pixel. OpenAI announced on May 5, 2026 that advertisers could begin signing up directly through a beta Ads Manager rather than going through a managed relationship. In the same announcement the company added cost-per-click bidding to the CPM buying that had defined the pilot, and confirmed it had recently launched a Conversions API and pixel-based measurement so advertisers can attribute purchases, leads and sign-ups.
The self-serve piece is the one that matters for most businesses reading this. OpenAI says companies can now register as advertisers, add payment information, set budgets, bids and pacing, upload creative, launch campaigns and view performance in a portal: the ordinary furniture of an ad platform. The company was explicit that the rollout is gradual and still in beta, so access is not guaranteed on day one.
Alongside that, OpenAI named its partner bench: agency groups Dentsu, Omnicom, Publicis and WPP, and technology partners including Adobe, Criteo, Kargo, Pacvue and StackAdapt. If you already work with one of those, you may have a route in that doesn’t depend on the self-serve queue.
How do ChatGPT ads work, and who actually sees them?
Ads sit below the answer, are labelled sponsored, and are shown only to logged-in adults in the United States on the free and Go tiers. OpenAI began that test on February 9, 2026. Plus, Pro, Business, Enterprise and Education subscribers do not see ads at all. Free users who would rather not see them can opt out in exchange for fewer daily messages; Go users can upgrade to remove them.
Targeting works off conversation context rather than a keyword you bid on. OpenAI says it matches advertiser submissions against the topic of the current conversation, past chats, and previous interactions with ads. Ask about recipes and you may see a meal kit. Advertisers never see the conversations or personal details.
They get aggregate performance data such as views and clicks.
The exclusions are a real constraint on inventory and deserve more attention than they get. OpenAI confirmed it will not show ads in conversations about health, mental health or politics, and will not serve them to accounts identified as under 18. If your category lives near any of those topics, the addressable surface is narrower than the headline user numbers suggest. And those user numbers are genuinely large: Sam Altman told employees in February 2026 that ChatGPT had roughly 800 million weekly users, as reported by CNBC.
The measurement function is younger than the ad product. Digiday reported in April 2026 that OpenAI was hiring its first advertising marketing science leader. A pixel and a Conversions API are the plumbing; the benchmarks, incrementality standards and normal-versus-abnormal ranges that let you argue with a CFO are the part that takes a year of data to build. Plan your first quarter as if you will be reading your own numbers without an industry baseline to check them against, because you will be.
What do ChatGPT ads cost right now?
Less than they did at launch, on every reported measure, but the launch figures themselves are disputed. Media buyers were quoted roughly $60 per thousand impressions during the initial February 2026 test, per Digiday. By April 2026, Digiday reported CPMs had fallen to as low as $25 in some cases, and that a subset of pilot advertisers were seeing cost-per-click bids in the $3 to $5 range in an early ads manager build.
On the entry ticket, two credible outlets published different numbers. Adweek reported in February 2026 that OpenAI had confirmed a $200,000 minimum commitment for early ChatGPT ads. Digiday reported the minimum spend commitment as $250,000 at launch, falling to $50,000 by April 2026. Both agree the floor dropped sharply and quickly; they disagree on where it started. We have not found a public reconciliation, so treat the launch minimum as reported-and-contested and the April figure of $50,000 as the number to plan against.
What a $3 to $5 click is worth depends entirely on what it sits next to in your account. Adthena data cited by Digiday puts Meta CPCs at three to five times cheaper than Google Search, not because Meta’s inventory is worse, but because the intent behind the click is different. Nicole Greene, VP analyst at Gartner, told Digiday that the move to CPC is precisely what lets advertisers compare ChatGPT results against other platforms on the same terms. That comparison is the whole exercise.
Run it before you commit, not after.
Should you move budget out of Google or Meta to fund this?
Not yet, and the strongest evidence against it is the conversion data on ChatGPT’s organic referrals. A working paper analysing twelve months of first-party Google Analytics data from 973 ecommerce sites (August 2024 to July 2025, $20 billion in combined revenue) compared more than 50,000 ChatGPT-driven transactions against 164 million from traditional channels. ChatGPT referral traffic converted below organic search and well below affiliate, and trailed both paid and organic search on revenue per session.
Two caveats matter and cut in different directions. The study measures last-click attribution on organic ChatGPT referrals, not paid placements, so it is a proxy for the environment rather than a read on the ad unit. And the trend was moving in ChatGPT’s favour: conversion rate and revenue per session both improved over the period, though average order value declined. The authors’ own projections suggested continued gains but no parity with organic search within the following year.
Two further studies point the same way. Amsive found no significant conversion difference between LLM and organic traffic, with LLM referrals making up under 1% of total sessions. A SALT.agency analysis found organic search visitors more likely to engage or complete key actions across most sectors, with LLMs ahead only in a few niches such as health and careers. The consistent read across all three: AI referrals are growing, and they are still small and still behind search on purchase intent.
None of that says don’t test. It says fund the test from a test budget. We made the same argument about where agentic commerce should sit in your paid search budget: new surfaces earn share by proving it, not by being handed it on narrative.
What should you test first on ChatGPT ads?
Test the thing the platform is structurally best at: reaching someone mid-decision who has not yet named a brand. Because targeting runs off conversation context rather than a keyword you choose, this is closer to a category-consideration buy than to bottom-funnel search. Four practical priorities for a first cycle:
Buy on CPC, not CPM. OpenAI says both remain available. CPC gives you a per-click cost you can put side by side with your Google and Meta numbers, which is the only comparison that will settle the internal argument. On CPM you are buying an impression in an environment with no published viewability or engagement benchmarks.
Install the pixel and the Conversions API before the first dollar spends. The measurement tools shipped in May 2026. Aggregate view and click counts will not survive contact with a finance review; you need the post-click event.
Send the traffic to a page built for a reader who arrived without a query. Someone clicking from a ChatGPT answer has been reading prose, not scanning a SERP. They have context and no keyword. A landing page that opens by restating the problem in plain language will do better than one that assumes intent you cannot see.
Set a fixed learning budget and a decision date. With a $50,000 reported minimum on the managed side and no published performance benchmarks, this is a channel where “we’ll see how it goes” quietly becomes a standing line item. Decide in advance what CPA would make you continue, and what would make you stop.
What should you not defund to pay for it?
Do not fund ChatGPT ads out of the work that gets you mentioned inside the answer itself. This is the trap, and it is a specific one. Ads in ChatGPT sit below the response and are separated from it. OpenAI has been consistent on this point since January 2026, when it published its advertising principles: answers stay independent of advertising. Fidji Simo, OpenAI’s CEO of Applications, put it plainly: ads will not influence the answers ChatGPT gives you.
Read that as an advertiser and the implication is uncomfortable but clear: buying ads does not buy you a place in the recommendation. If ChatGPT names three vendors in the body of an answer and yours isn’t one of them, no amount of sponsored placement underneath changes that. The two are separate scoreboards, and the one above the ad is earned through the content, entity and citation work that has nothing to do with your ad account.
Which means the budget most at risk is exactly the one that should be protected. Organic search still converts above ChatGPT referrals on the best data available, and it feeds the corpus that AI answers are assembled from. Cutting it to fund a beta ad channel trades a proven line for an unproven one and weakens the unpaid position at the same time. The same logic applies to the Google Ads automation shift toward AI Max: the channels are converging on AI-driven matching, and starving the signal you feed them is the one move that makes every channel worse at once.
Fund it from experimental budget, from a reallocation inside paid media rather than out of it, or from genuinely underperforming spend you were going to cut anyway. Our full view on the discipline sits in our paid search practice.
ChatGPT ads went from a closed pilot with a six-figure minimum in February 2026 to a beta self-serve Ads Manager with CPC bidding and pixel measurement on May 5, 2026: a genuinely fast opening, and worth testing. But the economics are still being discovered in public: CPMs reportedly fell from about $60 to as low as $25, CPCs surfaced at $3 to $5, the minimum spend dropped to $50,000, and OpenAI was still hiring its first advertising marketing science leader in April. Meanwhile the best available conversion evidence (973 ecommerce sites, $20 billion in revenue, 12 months to July 2025) puts ChatGPT referral traffic below organic search and well below affiliate, ahead of only paid social. Test it on CPC with a fixed learning budget and a decision date. Do not fund it by cutting the organic and content work that determines whether ChatGPT names you inside the answer, because the ad below the answer cannot buy you a place in it.
Before you buy the ad, find out whether ChatGPT already names you
We’ll check how your brand shows up inside AI answers for the questions your buyers actually ask, and where competitors are being named instead. Free, no obligation.
Sources
- OpenAI, “New ways to buy ChatGPT ads,” May 5, 2026: announcement of the beta self-serve Ads Manager; addition of cost-per-click bidding alongside CPM; recently launched Conversions API and pixel-based measurement; advertisers able to register, add payment details, set budgets, bids and pacing, upload ads and view performance in the portal; gradual rollout still in beta; agency partners Dentsu, Omnicom, Publicis and WPP; technology partners including Adobe, Criteo, Kargo, Pacvue and StackAdapt; OpenAI’s ads system controls all delivery decisions; advertisers charged on a click outcome under CPC
- Search Engine Journal, “OpenAI Launches Self-Serve Ads Manager for ChatGPT,” Brooke Osmundson, May 5, 2026: independent reporting on the same launch, including the limited beta status and the implications of CPC buying for performance marketers
- Search Engine Journal, “OpenAI Begins Testing Ads In ChatGPT For Free And Go Users,” Matt G. Southern, February 9, 2026: ads live for logged-in adult users in the US on the free and Go tiers; Plus, Pro, Business, Enterprise and Education tiers ad-free; ads placed below responses and labelled sponsored; targeting based on conversation topic, past chats and prior ad interactions; advertisers receive only aggregate performance data; no ads in conversations about health, mental health or politics and none served to accounts identified as under 18; free-tier ad opt-out in exchange for fewer daily messages; $200,000 minimum commitment per Adweek; approximately $60 CPM quoted to media buyers per Digiday; OpenAI’s advertising principles first published January 16, 2026 alongside the US launch of ChatGPT Go
- Search Engine Journal, “ChatGPT Ads Now Offer CPC Bidding Between $3 And $5: Report,” Matt G. Southern, April 21, 2026: reporting Digiday’s findings: CPC bids of $3 to $5 visible in an early ads manager build shown to a subset of pilot advertisers; CPMs down from $60 at launch to as low as $25; minimum spend commitment down from $250,000 at launch to $50,000; OpenAI hiring its first advertising marketing science leader; Nicole Greene, VP analyst at Gartner, on CPC enabling direct cross-platform comparison; Adthena data on Meta CPCs running three to five times cheaper than Google Search because of the difference in intent
- Search Engine Land, “ChatGPT, LLM referrals convert worse than Google Search: Study,” Danny Goodwin, October 23, 2025: reporting a working paper analysing 12 months of first-party Google Analytics data (August 2024 to July 2025) from 973 ecommerce sites with $20 billion in combined revenue, comparing more than 50,000 ChatGPT-driven transactions with 164 million from traditional channels: ChatGPT referral traffic approximately 0.2% of total sessions, roughly 200× smaller than Google organic; more than 90% of LLM-originating ecommerce traffic from ChatGPT; affiliate conversion rates 86% higher and organic search 13% higher than ChatGPT, with only paid social converting worse; ChatGPT trailing paid and organic search on revenue per session; lower bounce rate than most channels but generally lower session depth; conversion rate and revenue per session improving while average order value declined; model projections suggesting no parity with organic search within the following year
- Amsive study, reported by Search Engine Land: no significant conversion difference between LLM and organic traffic, with LLM referrals accounting for under 1% of total sessions
- SALT.agency analysis, reported by Search Engine Land: organic search visitors more likely to engage or complete key actions across most sectors, with LLM traffic performing better only in a few niches such as health and careers
- CNBC, February 9, 2026: Sam Altman told employees that ChatGPT had approximately 800 million weekly users
- OpenAI, “Our approach to advertising and expanding access to ChatGPT,” January 16, 2026: the five advertising principles, including answer independence and conversation privacy; Fidji Simo, CEO of Applications, on ads not influencing the answers ChatGPT gives
Keyword and People Also Ask demand for this topic could not be verified against a live keyword tool for this article — the section headings are grounded in the questions raised directly by the sourced reporting rather than in confirmed search volume, and are not presented as measured demand.