ChatGPT just revealed (exactly) what people prompt.
It's a gold mine.
Here are my key takeaways from OpenAI’s 60-page study “How People Use ChatGPT.”
The Gold Rush
In 1849, people chased shiny rocks in rivers.
In 2025, the gold is data, or what millions actually prompt on ChatGPT every day.
By July 2025, 700 million people were using ChatGPT weekly, or 2.5 billion messages a day (29,000/sec). That’s 10% of the world’s adults.
In comparison, Google has 200,000 queries per second. ChatGPT isn’t so far off. And the trend keeps going in that direction:
TL;DR: Google Search lost 40 million daily users in 2 years.
ChatGPT gained 100 million daily users in 2 years.
It's happening. ChatGPT changes how we search.
ChatGPT’s DAU/MAU ratio (desktop) doubled from 9-10% in early 2023 to nearly 20% by August 2025. Google Search’s ratio dropped from 27-28% to 25% in the same period.
So why do they use ChatGPT? To code? For therapy?
Data says no.
And by data, I mean a 60-page study from OpenAI itself that is live since this week (here’s the link).
ChatGPT is a coach, a co-writer, a search-plus engine for everyday life.
Understanding how people use it is understanding the future trends of the new gold rush. Time to mine some gold:
What People Actually Prompt
Work vs. Life flipped.
In June 2024, 53% of usage was non-work. By June 2025, it was ~73%.
Translation: ChatGPT is now a life tool first, a work tool second (even though work use is still growing in absolute terms).
The Big Three (≈80% of all chats).
Practical Guidance (how-to, tutoring, creative ideation)
Seeking Information (search-like queries)
Writing (drafting + editing).
Together, they dominate the feed. Over the last year, Guidance held ~29%, Writing slid 36% → 24%, Info surged 14% → 24%.
Coding is small. Editing is huge.
Coding is only ~4% of messages. I was surprised.
Meanwhile, in Writing, two-thirds of requests are “make my text better” (editing/translation/argument) rather than “write from scratch.”
That’s your tell: collaboration beats one-click generation.
Advisor > Autopilot.
Three ways people use ChatGPT:
Asking (50%) = get options, explanations, advice to help you decide.
Doing (40%) = ask it to produce an output (email, code, plan).
Expressing (10%) = casual chat.
What the data shows:
Asking is growing and gets higher satisfaction.
At work, people lean Doing (especially Writing) because it drafts fast. But the best results come when you treat ChatGPT like a thinking partner first.
What to do:
Use Ask → Decide → Do. Get options, pick one, then have it drafted. You’ll ship faster without shipping junk.
Quick example:
Ask: “Give me 3 angles to announce a product update to SMBs. For each: who it appeals to, 1 risk, 1 must-include detail.”
Decide: “Choose the best angle given: 150-word limit, low churn, upsell CTA.”
Do: “Draft the email using that angle. Keep it 120–150 words, add a P.S. for a webinar, and a 3-item verify list.”
If you skip “Ask” and go straight to “Do,” you’ll often get a plausible draft that’s mis-aimed—pretty words, wrong decision.
Don’t outsource judgment.
Don’t outsource taste.
Who’s Using It (and Why You Should Care)
The gender gap closed. ChatGPT mainly started male. By mid-2025, women used ChatGPT weekly more than men.
ChatGPT is young. Nearly half of adult messages come from users under 26.
Growth is global. Usage is accelerating in low- and middle-income countries.
Pros use it to ask, not just do. Highly educated professionals are more likely to use ChatGPT for work and to ask for guidance at work, not just push tasks.
People don’t find value in automation, but in decision support. That’s where productivity jumps, especially in knowledge jobs. One estimate pegs consumer surplus from generative AI at ~$97B/year in the US; inside firms, the big gains show up when people use ChatGPT as an advisor or research assistant.
Don’t automate everything.
Just make better decisions thanks to it.
Misconceptions to Drop
“It’s mostly coding.” Nope, ~4% of total messages.
“It’s AI companionship.” Tiny slice: ~2% combined.
“It’s only for work.” ~70% of consumer queries are non-work today.
Opportunities
Guidance-as-a-service. Coaching, tutoring, plan-building. Guidance stayed ~29% even as the user base exploded.
Writing partner. At work, Writing is king (~40% of work messages). Your edge: draft fast, then use GPT to finetune tone, structure, and clarity.
Search-plus. Info seeking rose 14% → 24%. Use GPT to synthesize, not just fetch. Ask for trade-offs, next steps, and verification cues.
Ride new capabilities early. When image generation landed (Apr 2025), multimedia jumped from ~2% → >7% and stayed high. New features shift habits. Be early.
What to Do — with copy-paste prompts
1) Ask → Decide → Do (default loop)
You are my decision coach to make drafts.
Context:
- Goal: [GOAL]
- Audience: [AUDIENCE]
- Constraints: [TIME/BUDGET/TOOLS]
- Deliverable after we decide: [DELIVERABLE TYPE & LENGTH]
Instructions (in order):
1) OPTIONS: List 3 options with pros/cons + key risks (≤6 bullets total).
2) RECOMMEND: Pick ONE based on [CRITERION]; give a 2-line why.
3) DRAFT: Produce the deliverable for the chosen option.
Output rules:
- Start with a 1-line plan (“compare → recommend → draft”).
- Short bullets; list assumptions if inputs are missing.
- Stop with the draft + a 3-item verify list.
2) 2-Pass Edit (structure → tone)
You are my editor.
Input: [PASTE DRAFT]
Tone sample (100 words): [PASTE or skip]
Pass 1 — Structure & Clarity (no tone changes):
- Reorganize logic, cut redundancy, fix unclear lines.
- Output: revised text + 3 bullets on biggest fixes.
Pass 2 — Tone & Polish:
- Apply the tone sample (or neutral professional). Keep my voice.
- Output: final version.
Rules: Preserve meaning. No new facts. Keep length within ±10%.
3) Search-Plus Brief (5 bullets + counterpoints + next steps)
Topic: [TOPIC] for [AUDIENCE]
Produce:
- 5-bullet executive brief (what matters & why).
- 2 counterpoints/limitations.
- Verify list: 3 claims to double-check + suggested sources/types.
- Next steps: beginner vs advanced (3 bullets each).
Style: concise, decision-ready. No fluff.
4) Coach me with constraints (personalized plan)
You are my personal coach.
Goal: [GOAL]
Constraints: [TIME/WEEK], [BUDGET], [TOOLS], [STARTING LEVEL]
Milestone/deadline: [DATE]
Make a 4-week plan with:
- Weekly outcomes; 3 concrete tasks/week (30–60 min each); a “fallback if behind”.
- A 10-minute weekly review checklist.
- Metrics to track progress.
End with: next 48-hour actions (3 bullets) + one motivational reminder.
5) Fact-spotter (verify before you ship)
You are my fact-spotter.
Input: [PASTE TEXT]
Tasks:
1) Flag all factual/quantitative claims needing verification (bullets).
2) Suggest where/how to verify each (source types or examples).
3) Rewrite the text with citation placeholders like [VERIFY-1], [VERIFY-2].
Rules: Don’t invent sources. If a claim is weak, propose safer wording.
Don’t (and what to do instead)
Don’t dump a vague mega-prompt and accept the first output. → Use Ask → Decide → Do.
Don’t outsource the blank page. → Draft something, then run 2-Pass Edit.
Don’t collect tabs when you need a decision. → Run Search-Plus Brief.
Don’t hide goal/audience/constraints. → State them in the prompt (see Coach).
Don’t publish important claims unverified. → Use Fact-spotter first.
Keep Your Team’s Edge
Train for asking. Teach your team to frame Asking prompts (comparisons, trade-offs, criteria) before delegating Doing tasks.
Instrument the obvious. If you build products, double-down where users already live: guidance, writing, search-plus.
Global & young. Localize and price for emerging markets; program for Gen-Z workflows (tutoring, coaching, job prep).
The New Gold Standard
If knowledge is power, usage knowledge is a superpower.
The 700M-user reality check says ChatGPT is not a slot machine.
Do that, and you’ll get the productivity boost without paying the cognitive debt I ranted about last time.
My 3-step recipe this week:
Ask for options (& trade-offs).
Decide. You have taste & judgement.
Do, hand GPT the context to draft, then you edit.
archive of part articles: https://docs.google.com/document/d/1pWuMCBVQo1zKcgKltX_BZxAr31KgxmOlp3Vzvmc5Hxc/edit?tab=t.0#heading=h.b81azxnvmq1z
What Ruben (that’s me) Will Do
I just recorded a Facecam video of how I feel about this study, and how it will change my media business (content creation) and my software (EasyGen).
only accessible to paid members