Who Pays for Consumer AI, and What Could Bring It to Everyone Else?

a16z editorial partner Elena Burger talks with investors Olivia Moore and Josh Elman about the seventh edition of the firm’s consumer AI rankings. The conversation combines panel data, product observations and the investors’ own predictions.

The report ranks web and mobile products by usage and adds observed US consumer-card spending. Moore says there were fewer first-time entrants than in earlier editions, even as new kinds of agents began to attract attention.

The panel cautions that rankings capture a particular moment. Products discussed heavily by technologists may take time to reach a mainstream audience, and traffic does not by itself describe revenue or depth of use.

They discuss early personal-assistant products and company-reported launch numbers. One founder’s reported credit-card connection and transaction figures are presented as claims about a specific product, not a general measure of consumer behavior.

A personal agent can become more useful as it connects to services and learns preferences, but the same connections create privacy and trust questions. The panel asks how users will decide what information and permissions to share.

Moore cites a community of early agent users where coding and technical automation were prominent tasks. She says mainstream-oriented assistants may cost less to serve than the intensive workloads of early power users.

Elman says a blank prompt box can be hard for a newcomer: people may not know what to ask or how to make an assistant useful. Examples and ways to learn from others could help bridge that gap.

Moore says usage is broad while paid subscriptions remain concentrated. In the report’s US e-receipt panel, 4.5% of eligible consumers had an active personal subscription to at least one of ChatGPT, Gemini or Claude in August 2026; this is a panel estimate, not a census.

Among observed AI payers, the report says the top 10% generated roughly half of consumer AI spending and the top 1% generated about one fifth. The concentration does not describe all users, most of whom were not in the paying group.

The report puts the top 1% of payers’ average observed consumer-card spending at $903 per month, compared with $25 for the median payer. These figures cover the panel and period measured, not every subscriber or all business spending.

Heavy spenders disproportionately buy developer, productivity and creative tools. Elman sees a market of people paying to build and create, although the group is narrower than the audience for mainstream entertainment.

The panel notes that not everyone wants a product mainly to save time. Entertainment, social connection and enjoyable ways to spend time are also large consumer needs, leaving opportunities beyond productivity.

Most consumer AI revenue discussed comes from subscriptions or usage charges. Moore argues that ads or transaction fees might eventually reach people unwilling to subscribe, although which models will work remains uncertain.

The group also discusses the cost side: serving heavy users can be expensive, and a flat subscription may be difficult to sustain without limits. Pricing must balance accessibility, usage and operating costs.

They discuss reports about ads in AI assistants and how advertising might fit into a conversational product. Their revenue and rollout figures are contemporary reports, not independently audited results in the interview.

An ad placed in a personal conversation can feel intrusive or influence a recommendation. The panel argues that disclosure, relevance and user trust will matter if assistants use advertising.

Elman gives a shopping example in which an assistant surfaced a supplier he had not known about. The panel uses it to illustrate potential discovery value while discussing how commercial incentives could affect such recommendations.

The report compares ChatGPT, Claude and Gemini through traffic, subscriptions and user mix. The speakers describe different product and distribution strategies; their rankings are specific to the report’s sources and observation window.

Specialized creative products can still attract users when they offer a distinctive model, workflow or audience-specific experience. The panel argues that a general model does not automatically replace a tool built for a particular craft.

Some products spread from individual use into teams and enterprises. The discussion describes how easy onboarding and a strong experience can help a small product gain adoption inside larger organizations.

The investors ask how startups can improve as underlying models improve without being absorbed by a model provider’s own interface. They point to audience knowledge, useful workflows, data and distribution as possible differentiators.

A personal assistant’s accumulated preferences and playbooks may make it more useful over time, but portability and privacy matter too. A rich product involves more than a model call, the panel argues.

The conversation ends with open areas such as shopping, dating, entertainment and social experiences. The speakers expect products to go beyond a text box when people need to browse visually, connect with others or spend leisure time.

The investors stress that successful consumer products need an experience people return to, not just impressive underlying technology. Their discussion is a snapshot of October 2026, not a prediction that any particular startup will succeed.