Two 2026 studies — a four-thinker-type investigation and an AI tool for analysing online behaviour — renew psychology's push to classify how people think, while their authors urge caution about what the categories can predict.

Two 2026 research efforts are reviving an old question in psychology with new tools: can the way people think and behave be sorted into meaningful types — and does that classification tell us anything useful about how they live?

Four types of thinker

In July, economists at Claremont Graduate University in California and Wuhan University in China published findings suggesting that people's conscious thoughts cluster into four loose groups. The study, reported by The Guardian and available as a preprint, recruited 258 volunteers aged 19 to 71. An app prompted them at eight random times a day for two weeks to record what they were thinking, what they were doing, and how happy they felt.

From more than 23,000 real-time responses, the researchers sorted thinking into four tentative categories:

• The worriers: occupied with world events, politics and social issues about 60 percent more than average, and with personal safety and health about 50 percent more. They also thought about music nearly twice as often as the others.
• The domestics: focused on chores, the home and finances, but also on stories told through films, books and television.
• The bodily thinkers: the youngest group on average, with a mean age of 32 and nearly two-thirds women. They had the highest rate of intrusive thoughts, at 28 percent.
• The work-hard, play-hards: minds split between career and leisure, reporting the highest income and happiness scores of any group.

The headline finding was less about the labels than about a claim attributed to the Roman philosopher Marcus Aurelius — that the quality of a person's life depends on the quality of their thoughts. The researchers found that what people were thinking predicted their happiness better than any other variable they examined.

The authors are notably cautious. The categories are not fixed, and people likely move between them over time. The sample is small and entirely drawn from what researchers call "Weird" populations — Western, educated, industrialised, rich and democratic — so the results may not generalise. The paper has not yet been peer-reviewed. Lead author Joshua Tasoff said a longer study with more participants would be needed to confirm the grouping, and that the labels were chosen to capture the "vibe" of each cluster rather than to serve as clinical categories.

Measuring behaviour at a scale surveys cannot reach

A separate effort, announced by the University of Colorado Boulder in August, takes a different route to a related question: rather than asking people about themselves, it reads what they already write online. Kai Larsen, a professor of information systems at the Leeds School of Business, developed a class of AI models called PsyProxy that translates ordinary written language into measurable psychological concepts such as trust, stress, loneliness and burnout.

Larsen's argument is one of scale. "We're estimating that it's about a million times more of these texts available for analysis than there are surveys we can ever do," he said. Social media posts, product reviews, employee feedback and patient messages, in his view, open a far wider window onto behaviour than questionnaire-based research can.

The tool is presented as a complement to surveys, not a replacement. Larsen is explicit about its limits: language reveals patterns associated with emotional states, not what people actually do. Someone who posts an angry review may keep buying; someone who vents online may never act on it. Online text also carries selection biases — the same self-presentation pressures that make social media look like everyone else is more successful.

The work was scheduled to be presented at the American Psychological Association's annual convention in August.

What typology does and does not explain

The two projects sit within a much longer, contested tradition of sorting people into types — from the four temperaments to the Myers-Briggs Type Indicator, which remains popular in workplaces despite persistent criticism of its scientific footing. Recent research, including a 2025 analysis of MBTI-based profiling using large language models, has underscored how easily such frameworks can be applied without evidence that they predict real outcomes.

The value of the new work may lie less in the categories themselves than in the questions they raise. The thinker-type study suggests that internal experience — not income, age or education — is the strongest correlate of wellbeing found in its data. The PsyProxy project suggests that the sheer volume of language people produce online could make behavioural patterns visible at a scale that was previously impractical. Both are early, both carry caveats, and both point toward the same unresolved problem: describing how people behave is easier than explaining why, or predicting what they will do next.