AI counseling tool helped students open up but struggled with crisis clues

A study of Chinese university students found that an AI agent encouraged disclosure, while trained counselors remained essential for assessing crisis risk.

Editorial illustration — not from the study.

A randomized study of 388 university students in China found that an AI-based counseling agent improved students’ views of the counseling relationship and their willingness to disclose information, compared with a control condition. A stronger perceived counseling alliance partly explained the greater willingness to disclose.

The agent also improved recognition of general risks and slightly improved sensitivity to high-risk cues. Its performance was more limited when students’ scenarios involved crisis-level disclosures, supporting a role as an initial support tool rather than a system that independently assesses and responds to danger.

What the study found

Students assigned to the LLM-based counseling condition reported significantly higher perceived counseling alliance and willingness to disclose than those in the control condition. Perceived counseling alliance partly mediated the relationship between the counseling condition and disclosure willingness.

In scenario-based tests, the agent improved general risk recognition and slightly increased sensitivity to high-risk cues. However, its performance was more limited for crisis-level disclosures. The results therefore indicate stronger benefits for engagement and emotional expression than for independent crisis detection.

Why the role matters

University counseling services can face more demand than professionals can handle. The findings suggest that an AI agent could help lower barriers to seeking help and make it easier for students to express what they are experiencing. At the same time, the limited performance with crisis-level disclosures supports a human–AI approach in which trained counselors retain responsibility for risk assessment and intervention.

Evidence and limits

The study used a randomized between-subjects experiment with 388 valid responses from university students in Chinese university counseling settings. Its risk-recognition results came from scenario-based analyses, and the abstract does not report detailed effect sizes or how closely the scenarios matched real counseling encounters. The agent performed less well with crisis-level disclosures, so the findings do not support using it as an autonomous crisis detector or replacement for trained counselors.

EU carbon taxes were linked to lower energy intensity and higher human development

A quasi-experimental study of 24 European Union member states found that adopting carbon taxes was associated with lower energy intensity, greater renewable energy use and higher Human Development Index scores. The gains were strongest in countries that had maintained the policies for more than two decades.

6 hours ago·1 views

Society & Economics

Who asks the SEC for records may signal where a stock is headed

A study of Freedom of Information Act requests to the Securities and Exchange Commission finds that the type of requester is linked to different signals about a public company’s future stock returns. The analysis also finds evidence that institutional investors and short sellers trade on information obtained through these requests.

11 hours ago·1 views

Society & Economics

Digital markets widen Gond painting’s reach—but not equally for all artists

Gond painting has moved from household walls and floors into exhibitions, digital marketplaces and online platforms. Based on 96 interviews and qualitative fieldwork, researchers found that technology expands markets and strengthens the art form’s public status, but unequal access to digital resources continues to exclude some artists.

1 day ago·6 views