Inside how 700M people actually use ChatGPT

If you spend any time on tech Twitter, you would think ChatGPT is mostly used by software engineers automating their entire jobs.

OpenAI just dropped the receipts, and the reality looks completely different.

Teaming up with Harvard economist David Deming, OpenAI’s Economic Research team published an NBER working paper analyzing 1.5 million anonymized consumer conversations. Backed by ChatGPT’s massive base of 700 million weekly active users, it is the largest study to date on how everyday people actually interact with consumer AI.

Goodbye coders, hello personal advisors

Despite the hype around AI building entire software stacks, coding and self-expression remain niche hobbies.

Instead, three-quarters of all conversations revolve around practical guidance, information retrieval, and writing. When looking at work tasks specifically, writing takes the top spot by a mile.

The researchers broke down user prompts into three distinct buckets: Asking, Doing, and Expressing.

Asking took the lion's share at 49%. People are not just treating the chatbot as a glorified copy-paste generator; they value it most as a trusted sounding board and advisor. This category also earned the highest user satisfaction ratings.

Doing accounted for 40% of activity, covering task completion like drafting emails, planning trips, or writing code, with about a third of this work-related. Expressing made up the final 11%, capturing personal reflection, creative play, and pure curiosity.

Overall, only about 30% of consumer usage is tied to work. The remaining 70% lives entirely in personal daily life, generating value through decision support that traditional economic metrics like GDP completely miss.

The demographic gap is vanishing

The study also shows that the early tech-bro bubble has officially popped.

Back in January 2024, among users with gender-identifiable names, only 37% were feminine. Fast forward to July 2025, and that figure surged to 52%, outpacing masculine names and matching general adult demographics.

Geographic adoption is flattening out just as fast. By May 2025, ChatGPT adoption growth rates in the lowest-income countries were more than four times faster than in the highest-income nations.

Even better for OpenAI: retention cohorts show that usage deepens over time as models improve, proving that people do not just try it once and abandon it.

Why it matters

This paper effectively pulls AI out of Silicon Valley’s echo chamber and anchors it in the real world.

For everyday users, the takeaway is simple: you do not need to master complex prompt engineering or know how to code to extract real value. If you have only been using ChatGPT to rewrite awkward emails, you are missing out on what half the world is doing—using it as a second brain for everyday dilemmas, life planning, and practical advice.

For the broader AI industry, this shifts the competitive battlefield. While frontier labs race to build autonomous coding agents and complex enterprise workflows, 70% of consumer volume belongs to mundane life questions. Building models that deliver grounded, trustworthy advice in a natural voice might be a much wider consumer moat than shaving two seconds off a software benchmark.

We think OpenAI’s move to publish this academic study is calculated and sharp. By framing ChatGPT as a democratized utility expanding across low-income regions and closing gender divides, OpenAI is building an ironclad narrative that AI access is a basic public right. The hidden risk, however, is clear: when hundreds of millions of people rely on a chatbot for decision support, hallucinated guidance carries far higher real-world consequences than a buggy script.