Access squeeze: insurer playbook and AI limits

UnitedHealth’s mental-health arm built an internal system that flagged patients who got more than 30 outpatient therapy sessions in eight months, then pushed reviewers to decide whether that care was “unwarranted.” Company documents reviewed by ProPublica said the program could save up to $52 million. (propublica.org) The people most likely to trip that wire were patients with severe depression, eating disorders, autism, and other conditions that often need long treatment arcs, not quick fixes. ProPublica reported that therapists described suddenly becoming targets after years of routine coverage. (propublica.org) This was not just about one denied claim at a time. The documents described “outlier management,” which works like a cost audit for therapy: find clinicians or patients above an internal threshold, then tighten scrutiny until visits fall. (propublica.org) ProPublica’s earlier reporting found another version of the same logic: some insurers cut coverage when a patient starts improving, arguing that progress means less care is needed. Therapists told reporters that improvement can be the point when treatment is finally working and stopping can trigger relapse. (propublica.org) Federal investigators have already found broad problems in this market. A Department of Labor report cited by ProPublica found widespread noncompliance with the federal mental-health parity law, which requires mental-health benefits to be covered comparably to medical and surgical care. (propublica.org) Now put that next to what is happening on the consumer side. A KOAA report published on April 9, 2026, described people using artificial intelligence chatbots for low-level mental-health support, including one Colorado Springs woman who said ChatGPT helped push her toward treatment after years of alcoholism and an eating disorder. (koaa.com) The same KOAA report drew a bright line: experts said artificial intelligence can be useful for journaling, reflection, or organizing thoughts, but it should not replace a licensed professional for severe symptoms or crisis situations. In other words, it can act more like a notebook or a search helper than an emergency room. (koaa.com) That warning matches what major mental-health groups are saying. The National Alliance on Mental Illness says artificial intelligence does not represent clinical expertise and is building benchmarks to test whether these tools stay within “safe informational boundaries” instead of acting like therapy. (nami.org) The American Psychological Association went further in a 2025 health advisory, saying generative artificial intelligence chatbots and wellness apps still lack enough evidence and regulation to ensure user safety for emotional support or treatment. (apa.org) So the system is pinching from both ends. Insurers are using internal cost controls to narrow paid access to human therapy, while patients priced out or shut out are being told that artificial intelligence may help with basic support but is not built to carry serious illness. (propublica.org) (apa.org) The result is a gap with a very specific shape: the people with mild needs may get a digital stopgap, but the people with the longest, hardest cases still need trained clinicians, covered visits, and time. That is exactly where the reporting says the pressure is greatest. (propublica.org) (koaa.com)