Openai, Humanity, and Ipsos on AI use

AI usage statistics for Openai ChatGpt, Claude of Humanity, and IPSOS show how to use AI every day, from prompts to adoption, amid low confidence.

Three heavyweight studies have just landed, each of which has actually pulled back the curtains on what AI really means. Together, these reports from Openai, Anthropology, and Ipsos offer something rarely obtained in the AI hype cycle: actual evidence. What people are saying that people are using these systems, what they are doing with them, how companies are unfolding (and not), and that the public is thinking about this.

OpenAI has released AI usage data from over 1 million sampled CHATGPT conversations between mid-2024 and mid-2025. Humanity has published a rare analysis of Claude AI usage statistics in its human economic index, including its first dive into enterprise API traffic. Ipsos, a global market research firm, surveyed over 23,000 adults in 30 countries in AI Monitor 2025.

Research on IPSOS in particular is useful because it faces the gap between perception and practice. All economics students learn about tidy fiction HOMO ECONOMALUS – A rational man who declares one thing, but then wanders outside the economics textbook, doing another. In academic jargon, this refers to the gap between preferences revealed as described.

It's a strangely familiar gap when you look at how the world employs AI. People report one thing in their research, but Openai and the human use log suggest that they do something else. To see this more clearly, it is worth looking at the main findings of the three reports.

The common reality of the effectiveness of “killer apps” and AI

Openai's ChatGpt conversation dataset tells us something calm. People are not using AI to plan colonies for the moon or lift austerity. They want to write help, practical guidance and quick information searches. These three categories alone make up almost 80% of ChatGpt traffic. Computer programming accounts for just 4%. The reflexes from the treatment barely reach 2%.

Even in a professional setting, “writing” is king, but not something you would imagine from a flashy marketing reel. Two-thirds of these queries are those who ask the system to hone what they have already written.

Anthropic's Claude draws a similar picture, but the users are distorted differently. Coding is dominant (36%), but education and science are rising rapidly to 12.4% and 7.2%, respectively. Additionally, Claude users delegate the entire task more frequently and pass on directives such as “You Do It” rather than engage in step-by-step prompts.

There is a long tail of exotic use cases on both platforms, but adoption is concentrated in the obvious sweet spot. The model works well and is a low barrier task. Sci-fi remains mostly in marketing slides.

Work vs. Play: The split reality of AI use

This is where things get messy. Openai reports that ChatGpt work use has dropped from 40% to 28% over the past year, with personal tinkering jumping almost three-quarters. Ipsos research confirms this broad perception. In many countries, AI feels more like a personal assistant than an enterprise backbone.

But humanity tells a different story. Its enterprise API data suggests a rapid increase in US workplace usage. 40% of employees have used AI since only 20% in 2023. API logs show a high degree of centralized automation deployment.

So which is that? The truth may lie in the labor department. The chat interface is for casual users and side projects. APIs are where serious business happens.

As the Claude Report itself warns, “Whether or not today's narrow, automation-heavy adoption will evolve towards a wider deployment could determine the future economic impact of AI.” In other words, the adoption curve may be about less growth and decline in growth, and what usage will dominate.

AI usage trusts paradoxes

Ipsos' global research shows harsh numbers ambiguity. Over half of respondents (54%) said they trusted the government to responsibly regulate AI. Only 48% said they trusted businesses to keep their data safe. The division is narrow, but it tells it.

Openai CEO Sam Altman himself appeared to embody the paradox at the Paris AI Summit. “Safety is essential to our work,” he told the audience. “We have to make these systems really safe for people, or people just don't use them. That's the same thing and we work very hard on it.”

Then, with almost the same breath: “That's not the main thing we've actually heard. The main concern is, “Can we make this cheaper, can we get even better and get more advanced?”

Safety has been mentioned, but it has not remained. The big themes are scale, cost, and ability. The paradox is that people are distrustful of AI companies, but the usage data shows that they continue to reward them with their daily trust.

AI fine print barriers, artificial and ipsos

Why isn't corporate recruitment completely mainstream? Anthropic's Claude Report is dull. Achieving productivity increases is more dependent on frontier capabilities than the troublesome details of deployment. Beneficial adoption of AI often requires costly restructuring of the process, retraining workers, and other sinking investments. In other words, AI is not plug and play. This is a reengineering project that involves rethinking how you run your business.

The same report highlights another bottleneck: context. AI delivers in complex, high-stakes settings, so you need rich, well-structured information tailored to your tasks. Many companies still can't offer it. Providing the right context often involves modernizing expensive data and changing organizations, which are slower and more expensive to deploy effectively than the hype suggests.

On individual aspects, Ipsos refers to a different kind of barrier: demographics. Adoptions are skewed towards young male, highly educated users. It first brings benefits and shapes the person who is excluded. And the details are ironic. The most common personal use cases, guidance and information seeking is also the most vulnerable to misinformation and hallucinations.

Between AI use and hesitation

General, Openai, Mankind, and Ipsos sketch clear pictures. AI is primarily used in normal cases. Get information, edit emails, and edit codes. Openai logs suggest that ChatGpt work usage is declining. Ipsos has discovered that AI is considered an individual helper. However, human enterprise data shows that 40% of US employees are already using AI in the workplace. What appears to be a contradiction is simply a layer. Personal play on the surface, invisible integration beneath.

However, another AI paradox is clear. Adoptions are rising rapidly, but faith in builders has not risen rapidly. Perhaps the real risk is not whether people will abandon AI, but whether they will normalize their dependence on systems they claim to be untrustworthy. The danger, and perhaps the opportunity, is that the future of AI is determined not by what we say, but by what we are doing at the prompt.

Source link