Same AI, Different You: How Context Shapes Your Questions

Microsoft published an unusually large observational study: 37.5 million consumer Copilot conversations from January to September 2025. I found no primary basis for calling it the largest chatbot-behavior study ever conducted. The same AI appeared to serve different roles depending on device and time.

In this Copilot sample, desktop conversations skewed toward work while mobile conversations consistently elevated health. “Work tool” and “health confidant” are Microsoft's interpretation of those observed topic patterns, not fixed user identities.

The Big Picture

The study tracked what people discuss, when they discuss it, and on which device. Enterprise accounts were excluded. This is pure consumer behavior.

Three patterns emerged:

Health dominates mobile. Not sometimes. Always. Health and fitness ranked #1 on mobile every hour, every day, for nine months. People reach for their phones with personal health questions regardless of time.

Work dominates desktop. Between 8am and 5pm, "Work and Career" becomes the top desktop topic, overtaking technology queries. The switch happens at business hours, then reverses after 5pm.

Late nights turn philosophical. "Religion and Philosophy" climbs during early morning hours. People ask big questions when everyone else is asleep.

Three Modes of AI Usage

Evidence note: the reported device and topic patterns come from Microsoft's observational study; the “mindset” labels in the interactive are editorial interpretations.

What People Actually Ask About

Topic Evolution Over Time

Evidence note: monthly topic orders are a transcription of the report's published figures, not a reusable Microsoft data export.

The rise of advice-seeking matters. People treat AI not just as a search engine but as a source of guidance on personal matters.

Weekly and Seasonal Rhythms

AI usage follows predictable human patterns.

Programming vs Gaming

Programming conversations spike on weekdays. Gaming rises on weekends. You can identify weekends just by the programming-to-gaming ratio.

Evidence note: this visual communicates the weekday/weekend direction shown in the report. Its repeated bar lengths are schematic rather than source measurements.

The Valentine's Day Effect

February showed interesting patterns. Personal growth conversations rose in the days before the 14th. Relationship queries spiked on the day itself.

Summer brought a drop in education and language learning. Entertainment filled the gap.

The 24-Hour Clock

Travel queries peak during commute hours. Philosophy and religion rise between 2am and 5am—when people are alone with their thoughts.

Evidence note: the clock is a qualitative reconstruction of the report's hourly figure. Low/medium/high/peak bands are editorial bins, not Microsoft-provided values.

Why This Matters

Microsoft frames Copilot as "a colleague at their desk and a confidant in their pocket." The data supports this.

My product hypothesis is that desktop experiences should test information density and workflow execution, while mobile experiences should test brevity, empathy, and personal guidance. The observational study does not establish that those designs cause better outcomes.

The study raises questions about responsibility. When people ask AI for health advice at 2am or relationship guidance before Valentine's Day, the stakes are higher than a search query. Response quality matters more when people treat AI as a trusted advisor.

What Users Say About Copilot

Reviews are mixed.

Power users praise the Microsoft 365 integration: meeting summaries in Teams, draft emails in Outlook, document analysis in Word. For people embedded in the Microsoft ecosystem, the time savings are real.

Critics point to accuracy problems. Hallucinations. Ignored instructions. Generic responses.

The pattern: Copilot works best for routine tasks and first drafts. It struggles with nuance. Users who treat it as a starting point report the best results.

One G2 reviewer: "It solves the information overload problem. Instead of drowning in email threads or missing details in back-to-back meetings, Copilot summarizes everything instantly."

The Bigger Trend

Microsoft's data is consistent with a broader shift toward personal support, but it does not establish a population-wide trend on its own.

The Shift in AI Usage

OpenAI reported that non-work ChatGPT messages rose from 53% to 73% year-over-year. Anthropic found AI usage spans IT, creative work, and business tasks. Both noted a rise in advice-seeking.

Those adjacent studies use different products, populations, and taxonomies. They are useful context, but they should not be blended into the Copilot result as if they were one comparable adoption series. The narrower conclusion is enough: within Microsoft's consumer sample, device and time were associated with materially different conversation topics.

What This Means for You

Your context shapes how you use AI.

At your desk during work hours, you likely treat AI as a productivity assistant. That frame works for document drafting, data analysis, and technical questions.

On your phone at night, you may seek something closer to personal guidance. Be aware of that shift. AI responses are generated text, not human wisdom.

AI has integrated into daily human rhythms. The question is whether the technology can meet the responsibility that comes with that role.

---

Key Numbers

---

Read the Full Study

---

Sources:

Primary Research:

• Costa-Gomes, B., Chen, S., Hsueh, C., et al. (2025). It's About Time: The Copilot Usage Report 2025. Microsoft AI. Read the full research paper (PDF)

Related AI Usage Studies:

• Chatterji et al. (2025). How People Use ChatGPT. OpenAI. [1.1M ChatGPT messages, May 2024–July 2025]
• Handa et al. (2025). Which Economic Tasks Are Performed with AI? Anthropic. [4M+ Claude.ai conversations]
• Tomlinson et al. (2025). Microsoft Research. [200,000 Copilot by Bing conversations]
• Bick, Blandin, and Deming (2024). [Generative AI diffusion research]
• AI Economy Institute (2025). [10%+ adult population weekly AI engagement]

AI Companion Research:

• Earkick Research (2025). AI Companions in 2025
• Scientific American. What Are AI Chatbot Companions Doing to Our Mental Health?

User Reviews:

• G2 Microsoft Copilot Reviews
• Capterra Microsoft Copilot Reviews
• Trustpilot Copilot Reviews

Press Coverage:

• Thurrott. Microsoft AI Releases Its First-Ever Copilot Usage Study
• GeekWire. Microsoft says its Copilot AI tool is a 'vital companion'