95% of ChatGPT users haven't paid a dime, but they might be the most valuable group.

Author: Apoorv Agrawal

Edited by: Deep Tide TechFlow

Deep Tide Overview: This is the third installment in the author’s series on ChatGPT’s business model. The first two installments respectively argued for usage (900M weekly active users, 70% market share) and stickiness (retention smile-curve, usage depth comparable to Slack). This one tackles the most crucial question: How much are these actual attention values worth? The core conclusion is counterintuitive— for top AI applications, the ceiling on advertising revenue could be higher than subscription revenue. And right now, ChatGPT’s 95% free users contribute nearly zero revenue, representing an untapped monetization opportunity that the entire industry hasn’t fully opened up yet.

The full text is as follows:

The first two parts of this series respectively showed ChatGPT’s user scale and real engagement. Those two parts discussed the “quantity” in the equation revenue = price × quantity—how many users there are, how many times they come back, and whether their habits are real. This part discusses the “price.” How much can you actually charge?

Time spent is the bridge connecting the two. In consumer tech, time is the raw material for monetization. Subscriptions convert time into perceived value and willingness to pay, while advertising converts time into ad inventory. The starting point is the same: how much time does your product take up from users?

The conclusion first: I believe the advertising revenue opportunity for leading consumer AI applications may be larger than the subscription revenue opportunity. The reason is simple: consumer AI is accumulating the same raw materials as the biggest internet companies—time and attention. The advertising revenue formula is straightforward: advertising revenue = total time spent × ad density × ad price. From these three variables, the data shows:

The total time spent on AI apps is exploding in growth. The distribution of AI attention shares follows the same power-law distribution as user counts—even after adjusting for time spent per user.

Time spent per user is increasing, meaning more and more sustained growth in ad inventory. AI apps currently lag behind consumer benchmarks, but they’re starting to approach enterprise applications. ChatGPT’s behavioral patterns look more like a work and productivity tool than a social information feed. This is a strong signal that ad density could improve in the future.

ChatGPT’s query intent signals are stronger than search, implying a higher ad price. See Section 3 below.

1. Total time spent: ChatGPT captures 68% of consumer AI attention

Total time spent on generative AI applications has grown about 10x over the past two years, and it grew 3.6x just in 2025. No app category expansion is that fast.

There are a few points worth noting. First, the inflection points around early 2025 are very clear. Driven by the expansion of ChatGPT’s voice, image generation, and search features, total time spent in the first half of 2025 roughly doubled. Second, Gemini emerged in mid-2024 and achieved meaningful growth, but it still lags far behind the leader.

ChatGPT accounts for 68% of total AI time spent, Gemini for 16%, and all other apps combined for about 16%. This level of concentration makes ChatGPT the most likely place for the first scalable AI-native advertising business. This also helps explain why OpenAI tried to monetize earlier and more aggressively than peers with smaller attention shares. This matters because you can’t do advertising on a platform that doesn’t have enough scale.

Where users spend their time is where ad buyers can place inventory. And this 68% of inventory is concentrated in a single product: ChatGPT. For ad buyers evaluating AI-native ad placement, the fact that attention is highly concentrated in one product is hard to ignore.

2. Time spent per user is rising, meaning more ad space

Every AI app in this chart is trending upward. Since early 2023, ChatGPT’s average time spent per user has grown by roughly three times. Claude, Gemini, and Grok have all surged sharply over the past year. The trend is clear: people are spending more and more time on AI apps—not just downloading and then giving up.

But compared with the consumer and enterprise application benchmarks we already know, how much time is that?

Compared with consumer apps: still much lower. ChatGPT spends 16 minutes per day, far below TikTok, YouTube, Instagram, and others. But this gap isn’t a fair comparison, because ChatGPT is missing two key elements that drive consumer apps to generate massive time spent.

First, it doesn’t have social network effects. TikTok and Instagram are sticky in part because your friends, creators, and communities are all there. The content is personalized for you and generated by the people you follow. This creates a pull that keeps you coming back and keeps refreshing your feed. ChatGPT doesn’t have these—no feed, no followers, no social graph.

Second, it doesn’t have a dopamine loop. You don’t open ChatGPT to scroll cat videos or check your ex’s updates. Consumer social apps are designed for variable-reward interactions: you never know whether the next swipe will be boring or interesting, and that unpredictability keeps you staring at the screen. AI assistants are the opposite. You come with a specific task, get answers, and then leave.

Compared with enterprise apps: higher than most. The enterprise comparison is more informative, but there’s an important caveat: these are pure mobile data from SensorTower, so desktop-dominant products like Slack, Gmail, and Google Docs are underestimated. Even so, the signal is still worth watching. Even on mobile alone, ChatGPT already looks like a high-frequency productivity tool. This matters because productivity products can monetize well even when time spent is far below consumer entertainment apps.

Slack charges $7–$12 per user per month. If AI assistants can operate at the consumer scale and already capture daily time spent at that level, the monetization opportunity is substantial.

Higher time spent per user means two things in the revenue equation: more perceived value to support willingness to pay for subscriptions, and more space to place ads. Both point in the right direction.

3. Monetization opportunities

3a. Why ads can beat subscriptions

Now let’s look at the pricing side: what are these attention assets actually worth? At consumer scale, the two most important monetization models are subscriptions and advertising. The key isn’t that the subscription model is weak, but that—historically—the biggest consumer internet companies have generated far more revenue through advertising than through subscriptions.

The biggest consumer subscription businesses:

The biggest consumer advertising businesses:

The magnitude gap is the core. Google’s advertising business revenue alone is about 5x Netflix, Meta’s about 4x Netflix. Even Amazon’s advertising business—almost nonexistent a decade ago—now exceeds Netflix. A natural question is whether higher subscription ARPU can make up for a smaller base of paying users. For some businesses, that’s indeed true. But at the scale level, the larger monetization base of advertising often wins.

Alphabet and Amazon are especially interesting because they have both models. In both cases, the advertising business is larger than the subscription business and grows faster. Netflix, Disney, and Spotify rely almost entirely on subscriptions, while Meta and TikTok rely almost entirely on ads. OpenAI currently relies almost entirely on subscription revenue, plus some recent ad revenue from a small-scale rollout among roughly 5% of free users in the U.S. This is an enormous attention pool that currently contributes almost zero revenue. OpenAI is the early mover here, but the same free-user monetization problem applies to every AI app with a large unpaid user base.

3b. How AI attention is priced

Ad revenue = total time spent × ad density × ad price

Total time spent: We already covered total time spent in Sections 1 and 2. ChatGPT has about 900M weekly active users. The DAU:MAU ratio is 45%, and on mobile it’s about 16 minutes per day.

Ad density (i.e., ad load) is a product decision. How many ads are shown per session? For example, Google shows 3–4 ads on each search results page, and Meta inserts one ad for every 3–5 posts in the feed. For ChatGPT, at most one ad is shown per conversation so far, and only for about 5% of mobile users. This restraint is wise for maintaining trust, but it means the ad density variable is currently very low.

Ad price (CPM) is the price an advertiser is willing to pay per one thousand impressions. This is where it gets interesting, because not all attention is priced the same. CPM ultimately comes down to whether the question has purchase intent: will this user buy something? This breaks down into three elements: intent (is the user actively making decisions?), attribution (can the advertiser trace ads back to purchase behavior?), and audience quality (does this user have purchasing power?).

Different major ad businesses rely on different advantages. Google Search has strong intent signals, because when someone types “2026 best mortgage rates,” they are expressing commercial intent in real time. CPM ranges between $15 and $200+, varying by category, and Google’s annual revenue per user is about $84 worldwide. Meta has weaker intent signals but massive time spent. Users browse 30–90 minutes per day, and Meta makes up for it with extraordinarily precise targeting, inferring intent from behavior and the social graph; annual revenue per user is about $57. YouTube sits in between: mid-level CPM, long sessions, and video creative.

Summary: Google sells intent, Meta sells attention, and YouTube sells watch time.

3c. Positioning of AI assistants—taking ChatGPT as an example

Use ChatGPT as the test case, because it has the largest free-user base and the most ad data. ChatGPT’s ad pricing is likely closer to Google than to Meta, and it may have an edge in categories where conversational context can strengthen commercial intent.

When someone opens ChatGPT to ask for laptop recommendations, compare insurance plans, or plan a family vacation, the interaction is similar to search, but with richer context. Users often include budget, preferences, constraints, and intent within a single prompt. This can make the commercial signal easier for ad buyers to interpret, even if it doesn’t automatically make every AI query more valuable than a search query.

I expect ChatGPT’s actual CPM to be at least comparable to Google Search, and possibly higher in some categories. Early data supports this. OpenAI’s pricing for premium ad placements is about $60 CPM, far above display ads, and sits in the price range typical of high-intent search ads.

Currently, ChatGPT has about 800–900M free users (95% of weekly active users). If ChatGPT can generate $30 in annual ad revenue per free user, then at the current scale that implies $25B in annual ad revenue. For reference: Meta generates $57 per user and Google generates $84 per user. So for a high-intent, login-required product, $30 isn’t aggressive.

Early data shows no impact on trust metrics, but the testing is still early-stage. Scaling ad volume by 20x without harming the experience of building user habits—that’s the real execution challenge. The main reason this opportunity hasn’t been validated yet is that not all AI time spent has commercial value. A substantial portion of ChatGPT usage is information-seeking, creativity generation, or productivity-oriented—not transaction-oriented. Moreover, unlike an information feed or a search results page, the conversational interface has fewer obvious positions to insert ads without harming trust. So the upside space is real, but the execution constraints are real too: OpenAI has to monetize without damaging the product experience that builds the habit of using it.

There’s also a more optimistic possibility. AI isn’t just creating ad inventory—it can also create entirely new ad formats. Conversational ads—where product recommendations are woven into the dialogue rather than attached to a sidebar—could actually improve user experience instead of reducing it. Imagine you ask ChatGPT to plan a weekend trip, and it recommends a relevant hotel deal within the conversation based on your unique preferences and memories. That isn’t a disruption—it’s a feature. If AI can enable hyper-personalization, agentic behavior, and truly conversational brand interaction moments, the ad opportunity might not only be enormous in scale, but also radically different from any ad experience that exists today.

3d. Why Google can wait

Google’s strategy is clearly different from OpenAI’s. Google has repeatedly said it has no plan to run ads in Gemini. In January at Davos, DeepMind CEO Demis Hassabis said he was “surprised” by OpenAI pushing ads in ChatGPT so quickly. Google Ads VP Dan Taylor wrote in December 2025: “There are no ads in the Gemini app, and there are no plans to change this situation.”

This is a strategic luxury unique to Google. Google already has an ad money-printing machine in search with $295B in annual revenue. It can use Gemini as a loss-leader to subsidize growth, expanding users and deepening engagement with an ad-free experience, while monetizing AI through existing search infrastructure (AI Overviews and AI modes already run ads). OpenAI doesn’t have this luxury. Without an independent cash cow to rely on, it must monetize directly from the chat interface.

For now, Gemini monetizes only through subscriptions. Compared with the advertising + subscription model OpenAI is building, this is a much smaller revenue opportunity per user. But Google is playing a different game: protecting the assistant to retain users, while aggressively monetizing on the search results page. Whether this strategy still holds when free-user inference costs rise and Gemini’s user base surpasses 750M monthly active users is an open question. At some point, the economics of operating a large-scale free AI assistant may force even Google to make a move.

Putting it all together

This series asked a simple question: Is consumer AI just having a lot of usage, or is it becoming a real business? The first part showed reach, the second part showed user habits, and the third part suggests that the monetization opportunity may be larger than many people think.

One thing that’s underestimated is that top AI assistants—especially ChatGPT—already have the best features for monetizing like the biggest consumer internet businesses: scalable, repeatable attention. About 95% of its weekly active users are still free users, meaning most attention today still has almost no monetization.

This doesn’t guarantee that the ad business will reach the scale of Google or Meta. Monetizing from a conversation interface is harder, and trust is the most valuable asset for a product. But if OpenAI can prove that ads can exist in high-intent assistants without breaking user experience, then the long-term ad opportunity may ultimately surpass the subscription business. And if that happens, Google’s real strategic problem won’t be whether Gemini should stay ad-free—but how long it can afford to.

Thanks to Sarah Friar and Fidji Simo for reviewing this article and the draft of this series.