5 Benchmarks for Evaluating AI Tools

Young people aren’t waiting for organizations to determine how AI should be used in mental health care. They’re already using it. 

Recent research shows one in eight adolescents uses AI for mental health advice, often turning to these tools to fill gaps in access, availability, and timeliness of support. At the same time, children’s hospitals, schools, and community organizations are facing workforce shortages and growing demand for mental health services.  

“Youth are already engaging with AI platforms, and organizations are trying to understand how that use can help inform care delivery,” said Heidi Baskfield, vice president of mental and population health at the Children’s Hospital Association. “It’s happening at a speed our systems are not prepared to keep up with.” 

CHA worked with children’s hospitals to develop AI procurement benchmarks for youth mental health tools to help hospitals, schools, and community partners evaluate vendors and make more informed decisions.

Why benchmarks matter

Organizations need a framework to help determine which AI tools are safe, effective, and appropriate for children and adolescents.

These systems deliver intervention directly to kids, supporting coping and resilience and, in some cases, addressing depression, anxiety, or suicidal thinking.

But that direct interaction with young people experiencing mental health symptoms and crisis situations makes AI tools fundamentally different from other platforms like general wellness apps.

A poorly designed or unsafe tool can amplify stigma, miss signs of crisis, and provide harmful responses and guidance. On the other hand, well-chosen systems can extend access to evidence-based support in communities where youth mental health providers are scarce.

Our benchmarks outline core criteria to ensure digital mental health tools meet high standards of evidence, safety, privacy, and accessibility.

1. Demonstrated impact

Many AI products highlight high engagement and user satisfaction. But engagement is not the same as impact.

Our guidelines emphasize the need for independent evaluation or clinical study evidence that shows measurable improvements in youth mental health outcomes. This includes evidence of reduced symptoms, improved coping skills, and increased help-seeking. 

2. Safety and risk mitigation

Safety is the most critical concern. AI tools may respond appropriately to explicit crisis statements but struggle with subtle or indirect distress signals.

Clear escalation protocols are essential. Systems must be designed to link users with trained professionals or crisis services when risk is detected. Crisis hotline numbers alone are not sufficient as the sole safety mechanism.

Without these safeguards, tools intended to support youth may trigger unsafe content or create unintended harm.

3. Privacy and data protection

AI systems rely on data to function and improve. Policies and practices should ensure only essential information is collected, stored, and shared.

Organizations should understand whether tools meet both health (HIPAA) and education (FERPA) requirements. Systems must have protections in place for both compliance with the law and proper ethical use of personal, identifiable information.

4. System integration

Organizations must consider how AI integrates with clinical systems, school infrastructure, and care pathways. Systems will not deliver sustained value if they cannot connect to what is already in place.

Jason Williams, Arkansas Children’s vice president and chief mental and behavioral health officer, says a tool will fail if it doesn’t fit into existing workflows.

“You could spend a lot of time and energy looking at the outcomes of a tool that doesn’t actually work with your system,” Williams said.

5. Access and equity by design

AI tools have the potential to expand access, but only if they are designed for diverse populations.

Platforms should address gaps in ADA compliance, languages, device flexibility, and broadband access.

Blending expertise

Procurement decisions for AI tools should not be made by administrators or IT departments alone.

Schools often serve as the first point of contact. Mental health professionals must be included in the evaluation and procurement process for any student-facing mental health tool.

Veronica Lake, executive director of student services at Kentwood Public Schools in Michigan, emphasized the importance of combining human expertise and technological insights to identify trends and support decision-making.

“Access is as limited as our willingness to collaborate,” said Lake, who provided input on the guidance. “The mental health crisis touches all of us. We have stronger data and progress when we partner together.”

A complex ecosystem

Not all tools carry the same level of risk or evidence.

Thomas Fernandez, MD, Yale New Haven Children’s Hospital child and adolescent psychiatrist and researcher, said AI tools span a wide range of use cases and vary in purpose and design.

“When we use AI-based mental health tools, we’re not referring to a single technology,” he said. “We’re referring to a heterogeneous ecosystem of digital tools.”

Fernandez explained this ecosystem includes clinically validated digital therapeutics, wellness and self-help applications, companion bots, and tools that support educators and clinicians.

The bottom line

AI is already part of how children and teens seek mental health support.

Our procurement benchmarks give organizations a path forward to implement digital mental health tools that maximize safety, privacy, and effectiveness.

Learn more about CHA’s AI procurement benchmarks for school-based mental health tools.

This work was presented in the webinar, "AI Guidelines for Youth Mental Health Tools," and the “Contracts First: Thoughtful AI Mental Health Procurement” session at CHA’s 2026 Together for Kids Pediatric Quality Conference.