Counsel Health CAIHL draft report

HugoScore CAIHL Draft Report: Counsel Health

Status: Draft for human review Last reviewed: 2026-06-12 Review method: Public-source review of the official homepage and FAQ, privacy policy, informed-consent document, vendor HealthBench research blog, funding press coverage, and Oura partnership announcement. No hands-on testing or independent validation was done. Service: Counsel Health Vendor: Counsel, Inc. (care delivered by Counsel Medical Group, P.A.) Category: AI-enabled virtual primary care

Summary

Counsel Health pairs a free medical AI chat with one-click escalation to its own physicians, who can diagnose, prescribe (no controlled substances), order labs, and refer. Consumers can sign up directly, optionally connect their patient portal records, and pay $29 per 7-day visit or $199/year for unlimited physician access, with clinical hours of 8am to 9pm, 7 days a week. The same platform is sold to employers and health plans as a "responsible AI front door," backed by a $25M Series A led by a16z and GV announced in late 2025, with 100,000+ members via partnerships. Oura announced that Counsel-enabled AI care will be available through Oura Labs starting in June 2026 for eligible members in 43 U.S. states. Counsel is unavailable in seven states plus DC and serves adults only.

The 2026-06-12 privacy-policy review changes the data-governance picture. The FAQ says Counsel never sells data and never uses it to train AI models. The full privacy policy, last modified June 23, 2024, says Counsel and its LLM partners may use user data to generate responses, improve AI models and responses, and personalize AI technology. It also permits targeted-advertising flows that may be considered sales of non-health personal information and allows de-identified data to be used or disclosed for any purpose. This should be framed as a tension between current FAQ commitments and older legal policy text until Counsel clarifies which governs current practice.

From a CAIHL perspective, Counsel is a genuine hybrid. The consumer product is patient-chosen and gives unusually cheap, fast access to physicians who carry the conversation to resolution. The enterprise identity is payer-aligned by design: the pitch to health plans promises reduced avoidable spend, audit-ready guardrails, and steering members to in-network resources. Both things can be true at once, but the question of whose priorities govern a given member's experience depends on which channel they arrived through, and that difference is not visible in the product marketing.

Evidence Reviewed

• Counsel Health homepage and FAQ (product model, pricing, states, data commitments): https://www.counselhealth.com/
• Counsel privacy policy, rendered 2026-06-12, last modified June 23, 2024: https://www.counselhealth.com/privacy-policy
• Informed consent (medical group structure, AI scope, record sharing, lab subcontractors): https://www.counselhealth.com/informed-consent
• Vendor research blog on HealthBench emergency-escalation evaluation: https://www.counselhealth.com/blog/how-counsel-leveraged-healthbench-to-assess-emergency-escalation
• Series A press release via Business Wire (confirmed via search and Healthcare IT Today reprint): https://www.businesswire.com/news/home/20251016535098/en/Counsel-Health-Raises-$25M-to-Launch-Physician-Supervised-AI-Front-Door-for-Healthcare
• Healthcare IT Today reprint of the funding announcement (metrics, investors): https://www.healthcareittoday.com/2025/11/07/counsel-health-raises-25m-to-launch-physician-supervised-ai-front-door-for-healthcare/
• Oura Ring 5 Business Wire announcement, including Counsel integration and 43-state Oura Labs availability starting in June 2026: https://www.businesswire.com/news/home/20260528686853/en/URA-Introduces-The-Worlds-Smallest-Smart-Ring-Oura-Ring-5

CAIHL Profile

• Who does this AI serve? Patients and payers/employers together. The consumer product delivers cheap, fast physician access. The enterprise pitch promises payers reduced avoidable spend and steering of members to in-network resources.
• Can patients tell AI is involved? Yes. Counsel AI is clearly labeled, the FAQ explains when AI answers versus when a physician joins, and physicians enter conversations visibly by name.
• Can patients meaningfully choose? Partial. Direct consumers opt in freely with a free tier and no insurance requirement, but the service is unavailable in seven states plus DC, employer and plan members receive it as a configured benefit, Oura announced a June 2026 Oura Labs channel in 43 states, and record connection is optional consent.
• Can patients correct or challenge what the AI produces? Partial. Any conversation can be escalated to a licensed physician, and the informed consent documents a complaint policy plus state medical board routes, but no specific workflow for correcting AI answers or record errors is publicly described.
• Does it help patients understand or act? Yes. Answers, prescriptions and refills, lab ordering, second opinions, referrals, appointment help, and unlimited follow-ups within a visit are concretely action-oriented.

Agency Interpretation

Counsel's clearest agency value is collapsing the distance between a health question and a physician who can act on it. Twenty-nine dollars for a week-long physician conversation, with prescriptions, labs, and referrals included, is below most copays, and testimonials emphasize exactly the situations where agency usually fails: holiday refills, unbookable specialists, and dismissed concerns.

The data-governance interpretation is now more mixed. Counsel's FAQ makes strong promises not to sell data or train AI models on member data, but the older privacy policy permits LLM-partner model improvement, targeted-advertising flows that may count as sales of non-health data, and broad use of de-identified data. Protected health information appears to sit under HIPAA covered-entity handling, but the boundary around non-PHI data and AI chat data needs clarification.

The unresolved tension is the dual master. Counsel tells consumers it is "the doctor that's with you for life" while telling payers it reduces avoidable utilization and steers members in-network. Steering can serve patients or serve plan economics, and nothing public lets a member see which logic applied to their answer. The evidence base has the same vendor-defined shape as its peers: an internal HealthBench-derived triage evaluation on 103 filtered scenarios, and press-release metrics like 96% issue resolution and $381 saved per engaged member, with no independent or peer-reviewed outcome evidence located.

Key Unknowns

• Which data-governance statement controls current practice: the FAQ's "never sell, never train" commitment or the June 23, 2024 privacy-policy language permitting LLM-partner model improvement, targeted-advertising flows, and broad de-identified data use.
• Whether employer and health-plan deployments report any member-level information back to the sponsor, even in aggregate.
• How physician oversight works operationally: whether physicians review AI answers that members never escalate, and what "physicians safeguard every interaction" means in practice.
• Which underlying AI models Counsel uses and how its "proprietary AI" is built and updated.
• How escalation thresholds differ, if at all, between consumer and payer-channel members.
• Correction workflows for wrong AI answers or wrong record data.
• Independent validation of the 96% resolution and $381 savings claims, and any clinical outcome evidence.
• The separate Health Information Privacy Policy that governs PHI and controls if it conflicts with the general privacy policy.

Publication Recommendation

Ready for human review as a draft profile, with the FAQ-versus-privacy-policy tension flagged prominently. Confidence should not rise above medium until Counsel clarifies current data practice, the enterprise data-flow question is answered, and some independent evidence exists. A hands-on test of the free tier and an escalated visit would also clarify visibility and escalation mechanics.