TL;DR

OpenAI flipped ChatGPT ads from CPM-only to cost-per-click today, with bid ranges of $3 to $5 per click surfaced in the ads manager. That moves ChatGPT out of the “reach buy” bucket and into direct competition with Google Search and Meta for performance dollars. The catch: early pilot data puts ChatGPT’s ad CTR at 0.91%, about seven times below Google Search. At that click rate, a $5 CPC has to clear the effective CPM math before anyone calls it a Google killer.

What OpenAI turned on today

According to Digiday’s reporting on April 21, cost-per-click pricing is now selectable inside ChatGPT’s ads manager, with bid amounts between $3 and $5 per click visible in advertiser-facing screenshots. Before today, the pilot ran purely on CPM. Advertisers paid for a thousand impressions whether anyone clicked or not.

The shift is a big deal for one reason. Performance advertisers (the Targets, the Fords, the Expedias running “drive me a purchase” campaigns) largely refuse to pay on an impression basis. They want the platform to eat the risk of non-clicking users. CPM gets you brand-awareness budgets; CPC gets you a slice of the much, much larger performance pie.

This is the first time OpenAI has offered pricing that a performance advertiser would recognize from a Google Ads or Meta Ads dashboard.

The 10-week price collapse that got us here

The CPC launch is easier to read once you see what happened to CPM. ChatGPT ads opened on February 9, 2026 at a $60 CPM with a minimum commitment of $200,000–$250,000. Aggressive pricing, aimed at premium brand buyers. Ten weeks later, that pricing is gone.

DateCPMMin spendNotes
Feb 9, 2026$60$200K–$250KLaunch, US only, Free + Go tiers
Mar 2026~$40$200KExpands to CA, AU, NZ; David Dugan (ex-Meta) hired to head global ad solutions
Apr 13, 2026$25–$35$50KMinimum slashed 4×
Apr 17, 2026As low as $15$50KSome buyers report floor pricing
Apr 21, 2026CPC live$50K$3–$5 per-click bids in ads manager

Two months. The CPM got cut in half (at least), the minimum got cut by 75%, and now pricing has a second axis entirely. That’s a platform rushing to fit whatever shape the buyer wants.

Ashley Fletcher, Adthena’s CMO, called it out in trade coverage: the decline “feels like everything is coming down, preparing for a wider auction accessibility to fit [a] global rollout.” Read: OpenAI is trying to get to a working global auction, and it’s willing to rewrite the pricing page every couple of weeks to get there.

The 0.91% CTR problem

Here’s the line every advertiser is staring at this morning. Adthena ran an analysis of early ChatGPT ad impressions in April and landed on a 0.91% click-through rate, against Google Search’s 6.4% benchmark. That’s roughly 7× worse, and it’s not a rounding-error gap. It’s structural.

ChatGPT users are simply not in a Google mindset. Someone searching “best running shoes for flat feet” on Google is shopping, and three or four of the top ten results will be ads pointing at exactly what they want to buy. Someone typing the same query into ChatGPT is usually asking for a synthesis. They want the chatbot to tell them the answer, not hand them a carousel of retailers. Clicking out to a store breaks the workflow they came for.

That behavioral difference drops to a simple arithmetic problem for a media buyer:

0.91%
ChatGPT ad CTR (pilot)
6.4%
Google Search CTR
$36
Implied effective CPM @ $4 CPC × 0.91%

At a $4 CPC and a 0.91% CTR, the effective CPM works out to about $36. At the top of the range ($5 CPC), it’s roughly $45. Either number is above the $25 CPM OpenAI is currently charging brand buyers on the same inventory. The CPC move is only a win for advertisers if their click-to-conversion economics justify the math, and for most retail, the answer depends entirely on what lies on the landing page after the click.

Who’s actually buying

Trade coverage from earlier this month names the pilot advertiser list: Target, Ford, Mrs. Meyer’s, Adobe, Expedia, Qualcomm, Best Buy, Enterprise Mobility, and The Knot Worldwide. Big, brand-forward names. Not a dev-tools startup in sight, which tracks, because the minimum spend was $250K until a week ago.

The early behavior is telling. According to reporting collected by Adweek and MediaPost, one enterprise buyer spent only 3% of a $250,000 pilot budget after several weeks. Agencies tell a mixed story. Omnicom secured placements for 30+ clients; Dentsu says volume is “building week over week.” The clients showing up are running small tests, not scaling campaigns.

And there’s a structural problem inside the ads manager itself: a glitch has prevented advertisers from viewing their own campaign performance data during parts of the pilot. That kind of outage burns trust before CPC even gets a chance.

The paid-user exemption

Every current ChatGPT ad impression goes to a Free tier or Go tier ($8/month) user. Plus, Pro, Business, and Enterprise subscribers are ad-free. OpenAI’s own wording is that paid plans “will not include ads for the time being” and that users “will always have a way to not see ads.” The framing is soft, but the message is clear: if you pay, you don’t watch ads.

It’s a consumer-friendly trade-off. It also means the audience an advertiser reaches through ChatGPT is, by design, the lowest-willingness-to-pay cohort of the platform. The people with corporate cards and Pro subscriptions, the ones Target’s brand team would most love to hit, sit on the wrong side of the paywall. That caps the audience you can monetize, and the cap gets tighter as OpenAI grows its paid base.

David Dugan, OpenAI’s new head of global ad solutions (hired from Meta in March), has the job of making the non-paying half of ChatGPT look like a valuable enough audience to move performance budgets off Google. Today’s CPC flip is his first public lever.

What Google and Meta have to worry about

Not much, yet. Here’s the realistic read for each incumbent.

Google: ChatGPT sends roughly 190× less traffic than Google Search despite taking an estimated 12% of total search volume. Clicks are the atomic unit of Google’s business, and ChatGPT isn’t producing them at scale. What Google should worry about is attribution: if the “best running shoes” research happens on ChatGPT and the purchase click happens on Google, Google still gets paid, but the discovery credit moves. Google’s last-click attribution model masks the erosion for now.

Meta: Less exposed. Meta’s ad system sells attention during passive scroll, which isn’t what ChatGPT does. The overlap only matters in direct-response shopping, where Meta and Google are already competing.

Performance ad networks (Amazon, The Trade Desk, programmatic open web): This is where OpenAI’s $2.5B 2026 revenue target has to come from. If ChatGPT becomes a viable CPC platform, it’s pulling from the same performance budgets these players already compete for.

OpenAI is reportedly telling investors to expect roughly $11 billion in ad revenue by 2027 and $100 billion by 2030. To hit those numbers, the 0.91% CTR has to move a lot, or the per-click price has to move a lot, or the audience has to include paid users. All three, probably.

My take

The CPC launch is an honest move, more than a bold one. Two months of the “$60 CPM is worth it” story didn’t hold up (advertisers couldn’t prove the ads worked, and the minimum had to come down four-fold to keep the pilot breathing). Going per-click is how you keep performance advertisers at the table when the reach numbers aren’t there yet. It’s the right call. It’s also a tacit acknowledgment that the brand-advertising product wasn’t landing on its own.

The interesting question isn’t whether $3–$5 CPC bids are competitive with Google (at today’s CTR, they basically aren’t). The real question is whether ChatGPT can redesign the ad unit itself: sponsored recommendations, sponsored follow-up questions, contextual carousels inside the reply, so clicks happen in places where clicks already feel natural. That would actually move the CTR. Bidding $5 on the current sponsored-result slot will not.

Frequently Asked Questions

How much do ChatGPT CPC ads cost?

Current bids in the ads manager are between $3 and $5 per click, per screenshots surfaced in reporting on April 21, 2026. Minimum campaign spend is $50,000, reduced from $200,000–$250,000 at launch in February 2026.

Are there ads in ChatGPT Plus, Pro, or Business?

No. ChatGPT ads only appear for users on the Free tier and the Go tier ($8/month). All paid subscription plans (Plus, Pro, Business, and Enterprise) are ad-free. OpenAI’s wording is that paid plans “will not include ads for the time being” and that there will “always be a way to not see ads.”

Where can I buy ChatGPT ads?

ChatGPT ads are currently available in four markets: the United States, Canada, Australia, and New Zealand. The international expansion landed on March 26, 2026. A wider global auction is reportedly in development but has no public rollout date.

Can I see how my ChatGPT ad campaigns are performing?

Partially. Reporting exists in the ads manager, but multiple agencies have flagged that the dashboard has had bugs preventing advertisers from viewing their own performance data during parts of the pilot. Click-through rates for the overall pilot have been measured externally at around 0.91%, compared to Google Search’s 6.4% benchmark.

How is ChatGPT CPC different from Google Ads CPC?

Google Ads runs on an auction where your bid competes for slots on a search results page densely optimized for commercial intent. ChatGPT’s CPC slots sit inside conversational responses where users are reading and asking follow-up questions, rather than scanning for links. The mechanics are similar (you bid per click), but the environment and user behavior are different, which is why the early CTR data is 7× lower.

Sources

Bottom line

Today’s CPC flip is the sign of a platform that knows its CPM experiment didn’t clear the bar and is willing to re-tool in public. The $3–$5 bids are competitive on paper, but with a 0.91% CTR they price out to a $45 effective CPM, worse than what brand buyers are paying now. The real test is whether OpenAI can invent an ad format that belongs inside a chat, rather than porting a search-result slot into one.