Four Key Findings
- AI adopters are pruning from the bottom. Power users fell to 8.5% of consumers, down 24% from September 2025, while their most-used activities held firm or grew.
- Product discovery is a core task. Twenty-three percent of U.S. consumers and 93% of AI adopters who research purchases used AI for discovery. Research and comparison was the only activity to post meaningful growth.
- Gemini took the product-discovery lead, reaching 63.2% of AI discovery users versus ChatGPT’s 56.5% after a surge among Android owners.
- Search wins the broad market, but power users go direct. Overall, 42% chose AI inside search while 57.1% of power users chose a standalone platform.
AI power users prune the tail and protect the core
The persona mix is moving toward the middle. From September 2025 to July 2026, light users rose 19% to a series high of 13.7% and mainstream users rose 7% to 33.1%. Power users fell from 11.2% to 8.5%. Three-month averages tell the same story: mainstream and light users gained 6.3 people per 100 consumers as non-users and power users declined by a combined 6.2.
The independent monthly samples cannot track people moving between tiers. Still, the pattern suggests new adopters enter through the light tier while experienced users narrow their routines after testing more uses.
Among power users, 37 of 52 measured activities declined between the September-to-November and May-to-July averages. The least-used activities fell 12%, while the most-used rose 2%. Power users did about 1.6 fewer activities on average, yet they were no less likely to use AI for high-complexity work. They cut from the fringe rather than abandoning the core.
Travel use collectively declined 10% across seven activities, including the four show in the chart below. Practicing a language fell 29.5% and currency research for travel fell 21.3%. At the other end, researching and comparing products rose 12.5%, editing personal writing rose 10.6% and saving, debt and credit tips rose 10%.
Activity count alone can misread maturity as retreat. Power users appear to be building a shorter, more dependable roster. Retaining them will depend on becoming indispensable in the uses that survive repeated testing.
In July, 22.9% of consumers had used generative AI to discover what they wanted to buy in the prior 12 months. Among AI adopters, the reach was 92.6%. It was nearly universal among power users at 98.9% and reached 95.7% among mainstream users. Even 68.8% of light users had tried it.
Within discovery, researching and comparing options led at 24.9% of users. Its three-month average grew 6% from late 2025, the only one of 10 activities to show meaningful growth. Eight activities declined, including return or complaint messages, down 30%.
Consumers are concentrating AI at the front of the purchase decision, where a broad field must be narrowed. Product discovery is becoming a research tool with a clear job: help people compare options, understand features and decide what is worth buying.
Gemini’s lead starts with Android
Gemini reached 63.2% of AI discovery users in July, up 17% from June. ChatGPT reached 56.5%, up 1%. July was the first month Gemini led for product discovery. Microsoft Copilot remained third at 25.7%.
The device split explains much of the change. Among Android discovery users, 76.3% used Gemini and 44.8% used ChatGPT. On iPhone, 68.8% used ChatGPT and 49.4% used Gemini.
The data suggest that the base layer of consumer technology can steer AI choice. An assistant tied to a mobile operating system and search ecosystem is easier to find and try. Distribution can decide which model gets the first opportunity to answer a shopping request.
Consumers want three different kinds of AI help
Asked to choose one AI surface for product research, 42% selected AI inside search, 34.6% chose standalone platforms and 23.4% chose merchant assistants. Search leads overall, but its advantage fades as AI use intensifies.
Power users favored standalone platforms by more than two to one over search, 57.1% to 28.2%. Millennials also put standalone AI first at 42%, while baby boomers and seniors favored search at 47.6%. Android owners preferred search, while iPhone owners split almost evenly.
Each surface wins with a different promise. Search removes friction: 41.7% cited free access with no separate account and 38.5% cited trust in the search engine. Standalone platforms win on neutral intelligence, trusted answers and follow-up conversation. Merchant assistants win near the transaction with exact inventory, current prices and checkout connections.
The market can support all three. Search is the easy starting point. Standalone AI is a research destination for people who want more control. Merchant AI connects an answer to an available product.
Implications and Opportunities
AI companies can optimize for repeated value. The strongest products will learn which jobs users return to, make those jobs easier and support deeper follow-up as decisions grow more complex.
Search companies enter with distribution, habit and trust. Standalone platforms can distinguish themselves through neutrality and stronger conversation. Merchants can supply facts general platforms may lack, including live inventory, accurate prices, loyalty benefits and a path to checkout.
Across the digital economy, clear specifications, current availability and reliable pricing need to travel across all three surfaces. The company providing the best ground truth has a better chance of appearing in the answer and completing the sale.
Conclusion
Consumer AI use is becoming more deliberate. The middle tiers are growing, power users are cutting weaker activities and product research is emerging as a clear job for AI. The next stage of adoption will be defined by which experiences earn a permanent place in the consumer’s shorter routine.
Read More
PYMNTS Intelligence is the leading provider of information on the consumer trends driving innovation in consumer finance, digital payments and financial inclusion. To stay up to date, subscribe to our newsletters and read our in-depth reports.
Methodology
“The AI Power User Gets Pickier” draws on the monthly Consumer AI Benchmark Report. This edition uses 11 independent survey waves fielded from September 2025 through July 2026. Samples ranged from 1,969 to 3,288 U.S. consumers, including 2,301 in July, and were balanced to match the U.S. adult population by age, gender, education and income.
Consumers were assigned to four weighted activity personas based on the breadth and frequency of their personal AI use. Product-discovery bases ranged from 484 to 956 respondents per month, including 521 in July. Platform use included 518 discovery users in July. Preference-reason bases were 798 for standalone AI, 963 for search and 540 for merchant assistants. Percent changes are relative differences, not percentage-point changes.
