What do ChatGPT ads look like? When 90% of users are unwilling to pay, AI is heading towards "fewer paying, more watching ads"

ChatGPT adding advertisements is no longer a question of “if” but “when.” Bryan Kim, a partner at the well-known Silicon Valley venture capital firm a16z, recently pointed out through data on LLM usage demand and monetization capabilities that relying on subscriptions for profit is unrealistic. Almost inevitably, high-tier users will pay while free users see ads. Data shows that the ARPU of the top 10% of AI products is only $30, making it difficult to support expensive development costs. Most users utilize LLMs for information retrieval, document editing, and seeking advice, rarely accessing advanced features.

AI products’ ARPU of the top 10% is only $30, making monetization difficult

For a long time, Marc Andreessen, founder of venture capital firm Andreessen Horowitz (a16z), has stated: “If you fundamentally oppose advertising, you are also opposing the large-scale adoption of technology.” From Google and Facebook to TikTok, free products plus advertising have always been the core model for internet services expanding to billions of users. Now, this logic is being replayed in the AI industry.

AI product average revenue per user (ARPU), source: a16z

According to a16z’s analysis of over 70 generative AI subscription products, converting free users into paying subscribers is challenging. The data shows ARPU:

Bottom 25% products: $11.05

Median: $16

Top 25%: approximately $23.98

Top 10% leading products: $30.40

This indicates that most AI products have limited monetization potential, while model inference costs remain high. For low-frequency or light users, subscription models are difficult to sustain in a commercial cycle, making advertising a necessary supplement to long-tail revenue.

Most users cannot access advanced LLMs; high-value scenarios account for only 4%

OpenAI’s published usage data shows that user needs mainly focus on daily productivity:

Writing and editing: 28.1%

Practical advice (life, health, learning, etc.): 28.3%

Information retrieval: 21.3%

In contrast, computer programming accounts for only 4.2%, with data analysis, mathematics, and other high-value scenarios even lower.

User purposes for ChatGPT, source: a16z

This also explains the difference in willingness to pay: engineers or professional users can gain several times the productivity from AI and are willing to pay $20–$200 per month; but for general users who only search for information, write emails, or ask questions, free services are sufficient.

Subscription-only monetization is unrealistic; advertising is inevitable for LLMs

According to Sensor Tower data, as of January 2026, ChatGPT mobile active users number about 800 million, Gemini around 200 million, with others like Perplexity, Character.ai, and Claude far behind. Even with a 5–10% paying rate, this still represents 40–80 million subscriptions. But to reach the remaining 90% of users and even scale to 1 billion, advertising is almost the only feasible free model.

LLM weekly active users, source: Sensor Tower

What might AI advertising look like? Three possible forms:

1. Intent-based ads (Search-like): For example, when asking about travel or recipes, recommending hotels or ingredient brands, marked as sponsored content.

2. Context-based recommendations: Similar to Instagram, predicting needs based on context to provide personalized products and services.

3. Agent commercialization (Affiliate Commerce): AI directly completes price comparisons, orders, and purchases, with platforms taking commissions.

Additionally, industry discussions include:

• Targeted bidding (investing more computing power for specific questions)

• Token-based pricing

• Diverse models such as AI companions and entertainment subscriptions

Currently, the core issue with AI is the immediate and enormous cost, while most users’ willingness to pay is very low. However, internet history repeatedly proves that when services aim to reach billions, advertising will inevitably appear. For OpenAI and the entire AI industry, subscriptions will not disappear, but the future structure is likely to be a small number of high-value paying users supported by subscriptions, with the majority of free users supported by advertising.

This article, “What does ChatGPT advertising look like? When 90% of users are unwilling to pay, AI is heading toward ‘few pay, many watch ads’,” first appeared in ABMedia.