How are OpenAI's advertising sales actually doing?

OpenAI started announcing the rollout of ads on February 13th. Two months have passed—how well has it actually been selling?

In this article, we’ll dig into it.

I. Looking at OpenAI ad test progress from the data

According to a report from The Information on March 26, OpenAI’s ChatGPT advertising business ARR (annual recurring revenue) has reached $100 million.

This definition is pretty odd, because the ad industry almost never uses the ARR metric.

It’s also a very inaccurate indicator, since advertising isn’t a subscription revenue model like SaaS.

So, we should look at more granular data—

The report says that currently less than 20% of ChatGPT users can see ads. OpenAI has expanded to 600+ advertisers and plans to launch a self-serve ad dashboard in April.

From the advertiser side, ChatGPT ads are being rolled out with both brand ads and performance ads.

On the brand ads side, the three major ad groups—Dentsu, Omnicom Group, and WPP—were invited to participate in the test. The minimum prepaid ad budget for the participating brands is $200,000.

On the performance ads side, the first programmatic partner selected is Criteo.

Criteo is an ad company listed on the Nasdaq but not especially large. It brought in 17,000 advertisers and has also started collaborating with Shopify’s Shop Campaigns.

Media reports say OpenAI and The Trade Desk (the largest independent programmatic advertising platform globally) have also already begun preliminary talks. Because of this news, The Trade Desk’s stock price jumped quite a bit.

According to Sensor Tower’s monitoring data—within two weeks, 100+ brands placed ads in ChatGPT.

I looked—it’s all big brands.

The reason for choosing big brands is probably also to ensure a good user experience during the test period, leaving a strong first impression for the industry and users.

Same from Sensor Tower’s monitoring data: retailers like Walmart and Target are the main industries for brand advertising.

So how good are ChatGPT’s ad results, really?

NP Digital collected data from five advertisers, and the results look like this—

In lead quality, GPT is 256% higher than Meta, but 49% lower than Google;

On cost, CPA is 46% lower than Meta and 38% lower than Google.

What clients ultimately care about is the combined effect of these two metrics: you not only have lower cost per individual lead, but also a higher rate of effective leads.

Viewed this way, GPT’s ads are about comparable to Google’s search ads, and clearly better than Meta.

Of course, this is just small-sample data. Once it goes live for real, the bidding environment will be more competitive, and whether the results can be maintained is still a question.

So, it seems like the ChatGPT ad testing is progressing pretty well, doesn’t it?

But the reality isn’t that simple—some parts are very much “makeshift.”

Two pieces of evidence can prove how makeshift it is—

First, it actually sends performance data to advertisers using CSV files.

According to AdWeek, it sends impression and click data to advertisers every week using a single CSV file.

That’s hard to imagine. As an AI company, even if it just did “vibe coding” a data dashboard, it could—yet it hasn’t.

Second, as of now, there are no screenshots of its ad dashboard anywhere on the entire internet, so we can infer that the infrastructure behind the dashboard is likely very rudimentary.

Of course, there’s also the hiring process: according to Digiday, roughly seven ad-related engineering roles are currently being recruited.

The next question is—

Why does the rollout of GPT ads look so unprofessional?

In my view, this is the result of OpenAI’s strategic wobble. There are frictions between its monetization function and its research and growth functions.

Let’s look at the timeline:

On December 2, 2025, Sam Altman sent a red alert (Code Red) internally, saying the entire company would focus on improving GPT’s core product, including the business delays affecting ads.

Only 45 days later, on January 16, the ads were announced as live.

If you’re going to do ads, then do them seriously—so what happened?

Two months later, on March 16, The Wall Street Journal reported that OpenAI planned to pause side projects and focus on coding and enterprise services.

Judging from the makeshift actions in the ad business above, ads are likely being counted as part of the side business.

Let’s connect the dots on why OpenAI keeps swinging strategically. First, why did it rush to roll out ads?

Some people say it’s due to financial pressure, but I think that’s just the surface reason.

The deeper reason is: free users, to a certain extent, are liabilities.

In the mobile internet era, free users were purely assets—you could use them to optimize recommendation algorithms.

But in the AI era, the logic has changed. To a certain extent, free users have become liabilities, because every question they ask consumes real computing power.

Minimax’s Yan Junjie said in a previous interview with LatePost that he had the following two quotes—

“Better models can lead to better applications, but better applications and more users won’t necessarily lead to better models.”

“In everyday use, the model is smarter than most users. Most users’ queries are basically not simulated as well as the model itself.”

That is to say, in Yan Junjie’s view, those “garbage questions” that free users ask still aren’t useful enough to generate good data.

GPT has 800 million weekly active users, and 95% of them are free.

And ads are the only way to convert those free users from a cost center into a profit center.

Now, why has it started wavering again recently?

Let’s look at two charts. The first—

From Ramp data: although OpenAI is still the number one in market share among enterprise users, it has started to decline, while Anthropic is rising rapidly.

If this trend continues, it’ll be caught up soon.

The second chart is clearer—

Also from Ramp: among new enterprise customers buying AI, 70% have chosen Anthropic—more than double OpenAI—and this overtake process has happened within the past month or so.

In other words, Anthropic has launched a surprise attack on OpenAI’s mainline at the intelligent layer, forcing it to put ads aside for now.

Another reason is that ads erode the research culture—

On the day ads went live on GPT, a researcher resigned.

Not only did he resign, he also wrote a dedicated article in The New York Times saying OpenAI is repeating Facebook’s past mistakes, and that’s why he chose to quit.

That article received 8,800 likes.

For OpenAI’s technical brand, of course, that’s a hit—it would reduce its ability to attract top AI talent.

So OpenAI’s core reasons for being conflicted about ads are basically two points—

First, the cost pressure from free users forces it to consider ads;

Second, friction between ads and research, growth, subscription revenue, and the technical brand makes it impossible to raise ads to a higher priority.

II. Google and Anthropic’s attitude toward ads is different from OpenAI’s—but also quite nuanced.

At this point, we can compare Google and Anthropic’s attitudes toward ads—it’s actually pretty interesting.

Google—its also been grappling—

Right now, Google is adding ads into AI Overviews and AI Mode in Search, while there are no ads yet in the Gemini main app.

However, Google executives’ statements about ads have been quite subtle—

In January this year, Demis Hassabis (CEO of DeepMind) said at the Davos forum—

“Gemini currently has no plan to add ads. It’s interesting that OpenAI adds ads so early—maybe they need more revenue.”

In December 2025, when global ad VP Dan Taylor was interviewed—

“There are no ads in the Gemini app, and there are no plans to change that.”

In early 2026, Google SVP Nick Fox also said—

“We can’t rule out ads being shown in Gemini. The ad experience in AI Mode could extend to Gemini.”

These executives’ messages don’t match each other, which shows that Google internally is still fighting over how to approach ads.

Now look at Anthropic: even though it doesn’t add ads, it still spent $8 million on Super Bowl ads to mock its competitors for adding ads. And even though its B2B revenue is strong, Anthropic still hasn’t said anything in stone—

On the very day its Super Bowl ads were running, it updated an official blog post explaining why it isn’t adding ads.

In that post, there’s a line that says:

“If we need to reassess our strategy (change our mind and add ads), we will explain the reasons publicly and transparently.”

In other words, it also leaves a “trapdoor” in case it gets challenged—it hasn’t put a final stamp on the decision.

III. Ads may no longer be a “sexy” industry

With the attitudes of these three companies toward ads being so big, let’s talk a bit further out: re-examine the business model of advertising itself.

First, a few takeaways—

First, advertising’s share of GDP has basically stayed unchanged for years.

US data: from 1991 to 2017, the share of advertising in US GDP stayed largely stable, hovering long-term between 2% and 2.5%.

China is the same. The advertising revenue curve and the nominal GDP curve basically overlap after adjustments—

The reason we see internet advertising growth outpacing GDP growth is that internet ads are replacing traditional ads—

This chart makes it even more obvious: traditional advertising is declining, while the pink segment—internet advertising—is rising rapidly—

These are the latest data I saw just a couple of days ago—

By 2028, the total revenue of traditional media advertising combined still won’t match the advertising revenue of Amazon alone.

US analyst Mary* Meekle previously clarified one key point: total advertising revenue and total media time ultimately need to match.

Let’s directly compare data from 2009 to 2018: print media dropped from 26% in 2009 to 7% in 2018.

Internet rose from 13% in 2009 to 51% in 2018, and the internet also happens to account for 51% of people’s time. (Although these aren’t the newest figures, the big logic that ad revenue matches time is clear.)

With that, we can easily draw a conclusion—

When the substitution effect on traditional ads starts to fade, the growth rate of the overall internet ad pie also declines.

You can see that reflected in the revenue growth rates of companies whose core business is ads.

At the core, it comes down to the fact that it’s hard for online attention to take a significantly higher share of total human attention, and ads ultimately need attention.

Similarly, we can easily infer—

Ads have increasingly become a less “sexy” business model than before. In a sense, it becomes a “quasi non-depleting” market.

If an AI company uses ads as its business model, it’s hard to support a wild market-implied valuation multiple.

Ads are ultimately a subset of productivity. If AI can directly become productivity—as Musk said, 10x in 10 years—that would obviously have far more upside than the imagination around ads.

Anthropic is exactly this kind of model, and it has shown strong momentum.

The above is just my personal hot take.

Back to reality: even though the logic above is correct at a macro scale, at this stage, advertising is still a business with extremely high certainty—and for C-side user-scale-leading companies like OpenAI, it’s a pragmatic commercialization choice.

So next, we’ll do the math together: how much can OpenAI earn in its first year?

In my previous article, “Five Analysis Points on OpenAI Adding Ads,” I calculated that—

The two methods’ estimates don’t differ by orders of magnitude. The conclusion is that OpenAI’s ad revenue in its first year is likely in the range of $2.0–$8.0 billion per year.

Based on the somewhat makeshift start so far, I’m inclined to endorse the lower bound of this forecast: $200 million.

That was for first-year short-term revenue. What about long-term revenue?

According to a forecast by a bank analyst Ken Gawrelski at Wells Fargo: by 2030, ChatGPT could capture 30% of global search ad spend.

That would correspond to $100 billion in GPT ad revenue by 2030—

I think it’s possible for this “1000” to happen, for three reasons—

First, OpenAI has something it can copy.

The ad system is a mature framework. Google and Meta spent 20 years building an advertising money-printing machine. The core architecture and methodology are not too hard to transfer.

There are no non-compete agreements in Silicon Valley. It can hire people from mature ad teams and quickly build a complete ad tech stack. Also recently, OpenAI recruited Dave Dugan, a former Meta ad executive, to lead ad sales.

Second, ad revenue is highly scalable.

GPT has 800 million weekly active users and 2.57 billion conversations per day. As long as you maintain it, ad inventory is huge. Once ad load increases from an extremely low level, revenue will grow exponentially.

It’s hard to ramp up early, but once the system is polished and mature, operations during the scaling phase will be relatively easier.

Finally, to wrap up the core conclusions of this article—

1. Free users are, to a certain extent, a liability for OpenAI, and the cost pressure from free users forces it to consider ads.

2. The exponential demand for Agents and Coding represented by the “lobster” approach—drives a more intense need for direct intelligence selling, which is more “sexy” than selling ads.

3. Friction between ads and research, growth, subscription revenue, and the technical brand prevents OpenAI from raising the priority of ads.

4. As a mature business model, ads have substantial monetization potential and could become an important part of OpenAI’s revenue in the long run.

Conclusion

Finally, a note on what could happen with Chinese large-model companies in the advertising space—

Currently in China, apart from Baidu, everyone else hasn’t added ads into their model products. The core reason is that China’s AI C-side products are still in the battle stage, and no one is short on money.

But as user scale grows and inference costs rise, the top two C-side AI product user bases in China will most likely add ads to balance cost pressure.

As for timing, it’s expected that within 1–2 years, 200 million DAU is likely a key milestone.

And according to the past rule that only the “first-place” C-side player is qualified to be the first to add ads, the most likely first advertiser among Chinese players is ByteDance’s Doubao.