We analyzed roughly 300,000 conversations, 100,000 each from ChatGPT, Claude and Gemini, to ask a simple question: when do people actually use AI assistants, and does the answer depend on where they live, how old they are, how much they earn and what they are trying to do?
Key findings
- AI usage tracks office hours: 76–78% of conversations happen on weekdays and work prompts peak around 11am local time.
- Claude is 71% work-focused, versus about 40% for ChatGPT and 43% for Gemini.
- Europe shows the sharpest professional/personal split; North America blends personal use across the whole day.
- Age and income predict timing: older and higher-income users cluster on weekday mornings; under-29s peak on Sunday evenings.
- Each platform has a stable specialty (ChatGPT: writing; Claude: code; Gemini: multimedia) that shifts toward creative work at weekends.
Each assistant keeps a different clock
Work conversations peak around 11am local time on every platform and decline through the evening. Claude is overwhelmingly a work tool (71% work conversations), while ChatGPT and Gemini sit around 40–43%. ChatGPT shows the sharpest daily swing between peak work and peak non-work hours (3.59 pp spread); Claude is the flattest (2.11 pp).
The workday is a European story
Pooling platforms within regions, Europe shows the sharpest work/personal divergence during the afternoon, while North America's work and non-work lines run almost parallel all day. Every region except North America dips at lunchtime, most visibly in Asia. Latin America peaks earliest and its evening rise leans personal rather than professional.
| Region | Afternoon work/personal gap | Lunchtime dip | Evening lean |
|---|---|---|---|
| Europe | Sharpest | Yes | Personal |
| North America | Nearly none | No | Blended |
| Asia | Moderate | Most pronounced | Mixed |
| Latin America | Moderate · earliest peak | Yes | Personal |
Age and income influence when people are online
The 30–49 group concentrates most densely in weekday midday hours. Users 65+ cluster in late mornings and drop off in the early evening. Under-29s peak on Sunday afternoons and evenings. By income, the $100–200k bracket shows the tightest weekday work-hour clustering; users under $25k spread across nearly every hour after 5am; the highest earners ($200k+) show a notable Monday-morning spike.
| Group | Mon–Fri AM | Mon–Fri midday | Mon–Fri eve | Sat | Sun eve |
|---|---|---|---|---|---|
| Under 29 | · | · | · | · | peak |
| 30–49 | · | peak | · | · | · |
| 65+ | peak | · | · | · | · |
| Under $25k | · | · | · | · | · |
| $100–200k | · | peak | · | · | · |
| $200k+ | Mon peak | · | · | · | · |
Each assistant has a specialty, and the weekend rewrites it
ChatGPT's top weekday topic is writing (28% of conversations), Claude's is programming (35%), Gemini's is multimedia (29%). At the weekend, writing drops hardest on ChatGPT (28% → 21%) while Claude's programming share rises slightly (35% → 37%). Fiction, cooking, tutoring and creative ideation all rise across platforms at the weekend; editing and critiquing text slides from around rank 3 to rank 7 on Saturdays on ChatGPT, and math and data analysis fall on Claude.
What this means for your brand
- Time your targeting to your audience. Affluent and older users are on weekday mornings; young consumers on Sunday nights.
- Expect the weekend topic shift. Analysis gives way to creative work, so content for hobbies, cooking and fiction has its moment.
- Align with platform specialties. Writing on ChatGPT, code on Claude, multimedia on Gemini.
Getting started
Flowgen tracks brand visibility across every major AI platform, by region, persona and time. Talk to the team to see when your buyers are asking about you.
Methodology
Dataset of 300,000 conversations (100,000 per platform) from 195,907 unique users in 27 countries, July 14, 2025 to June 15, 2026. Demographic analysis covers ~265,000 conversations with known age and income. Topics were classified with an LLM-as-judge against a 24-category taxonomy validated on the WildChat dataset. Results are rake-weighted to demographics and reported with 95% confidence intervals.