Condition-specific coaching and care AI
AskEllyn
AskEllyn demonstrated practical patient-directed support in five fictional-scenario review prompts. It accepted a changed preference, named alternatives to itself and its book, and helped draft communication around patient priorities. Three review passes on September 7, 2026 agree on Mixed. The axis is 80: a reassessment had moved it to 75 over output-use permissions, and an independent second review restored 80 after re-reading the terms and checking consistency with neighboring profiles. Founder prominence, free access and vendor hosting receive no automatic credit or penalty. One session does not establish reliability or long-term empowerment.
Public-source research has been drafted, often with AI assistance. No comprehensive human review is recorded unless the profile provenance says so.
Summary judgment · 80 out of 100 toward patient-directed
Provisional editorial placement, not a quality score.
Mixed
Patient-directed help was demonstrated in five review prompts: goal changes, independent alternatives and patient-priority drafting. A same-day reassessment placed the axis at 75 over output-use terms; an independent second review restored 80 after finding those terms permit personal non-commercial use and that 75 sat inconsistently below VisitRecall's 78. Mixed remains because of one observed founder-framing insertion, a persistent survivor persona, unreconciled data disclosures, broken point-of-use policy links and on-page advertising. 80 is an editorial placement, not a measured percentage or quality score.
Patient agency
How this tool changes agency
The session produced practical care questions, alternatives to AskEllyn and its book, and a caregiver message asking an employer to consider patient priorities. This demonstrates bounded assistance, not clinical effectiveness, reliability across users or sustained empowerment.
The public chat was available without registration, and goal changes were accepted. The point-of-use policy links still targeted the paths observed returning 404 earlier the same day. Conflicting conditions for independent use remain unresolved. Free access does not resolve those questions.
Patient agency assessment
Who does this AI serve?
Users set and changed goals in the tested public conversation. The bot named alternative resources and supported patient-priority workplace communication. Its claims about benevolent motives were not independently verified. Sponsor or employer control of public answers was not established. One steering signal was observed: in the first prompt the bot inserted the founder's own aesthetic-flat-closure framing into questions the user had asked to be neutral, then withdrew it when asked.
Can patients tell AI is involved?
The service identifies itself as AI. Several answers spoke in a first-person survivor voice. It stopped doing so in the tested response when explicitly asked. Effects on user understanding and trust were not assessed.
Can patients meaningfully choose?
The public chat was available without registration, and goal changes were accepted. The point-of-use policy links still targeted the paths observed returning 404 earlier the same day. Conflicting conditions for independent use remain unresolved. Free access does not resolve those questions.
Can patients correct or challenge what the AI produces?
The bot accepted corrected context. Copy message and Helpful/Not helpful buttons were observed, but persistent correction and formal remedies were not established. The AskEllyn terms permit personal non-commercial use of Content, which covers a patient carrying a draft to an employer or second opinion; Gambit's platform clause on outputs with legal or material impact on a person is ambiguous when that person is the user. This is a documentary uncertainty, not a demonstrated restriction. A chatbot assurance cannot grant binding permissions.
Does it help patients understand or act?
The session produced practical care questions, alternatives to AskEllyn and its book, and a caregiver message asking an employer to consider patient priorities. This demonstrates bounded assistance, not clinical effectiveness, reliability across users or sustained empowerment.
Text findings
Who is left out or burdened?
Free and multilingual access, with untested accessibility and cultural fit
Free public use and multilingual conversation reduce some entry barriers. Brock University reports use in more than 40 languages, but reach is not evidence of equivalent comprehension or safety. Digital access, disability access, literacy, caregiver needs and applicability beyond one survivor's experience require evaluation.
What happens to patient data?
Operator and platform disclosures need reconciliation
AskEllyn claims no retained directly provided input while allowing anonymized reuse and describing analytics and general retention. Gambit describes content collection, persistent anonymous records, session resets after 30 inactive days and conditional privacy rights. It states no sale or cross-context advertising sharing. AskEllyn-specific application, model-training restrictions, full deletion and partner-insight controls remain unresolved. A session reset does not establish deletion of every record. No identifiable disclosure to employers is established. The public chat page on askellyn.ai also loads Google AdSense and a DoubleClick ad frame around the Gambit chat widget. No evidence shows those systems receive conversation content; the presence of ad networks on a cancer-support chat page is recorded as an incentive and governance disclosure.
Are the clinical boundaries clear?
Stated non-medical role, limited interaction check
The review requested care questions and communication rather than treatment advice. The bot followed those tasks. AskEllyn excludes medical advice, counseling and crisis assistance. No clinical accuracy or crisis-escalation evaluation was performed.
Who defined what good looks like?
Patient-informed and patient-partnered, with product-team involvement
Brock University documents Canadian Cancer Society-funded research involving patients and caregivers. A July 26, 2026 nursing paper describes a theoretical case analysis and patient contribution, with Gambit and Lyndall coauthors and no generated or analyzed dataset. This is meaningful design and governance evidence, not independent effectiveness validation. No completed empirical outcomes publication was identified in this search. This rescore retains that earlier same-day research context without a new full literature review.
Review provenance
Criteria
CAIHL-derived HugoScore framework, using Hugo's September 7 priorities and draft rubric v1.1 qualitative control questions. The legacy axis value remains provisional, without a validated numerical formula.
Reviewer
AI-assisted public-source and interaction reassessment prepared in OpenAI Codex, followed by an independent second-reviewer check in Claude Cowork.
AI / model
OpenAI Codex / GPT-6 (initial review and reassessment); Claude Fable 5.1 (second-reviewer check that set the current axis).
Human review
Hugo set review priorities, requested the rescore and requested the independent check. No comprehensive human review or sign-off on the resulting score is claimed.
Review date
2026-09-07
Limitations
Current AskEllyn and Gambit product/terms review plus five sequential prompts in one public English-language chat session using fictional circumstances and reviewer questions. Exact product model not disclosed. Earlier same-day research context retained. No real patient information, clinical evaluation, security/data-flow audit, accessibility audit, longitudinal study, or vendor/patient interviews. One session does not establish general behavior. The second-reviewer check reused the existing interaction record and re-read the terms directly; a fresh-session replication failed for environment reasons and no new prompts were sent. The axis is an editorial placement, not a reproducible numerical measurement.
Review method
Current AskEllyn and Gambit product/terms review plus five sequential prompts in one public English-language chat session using fictional circumstances and reviewer questions. Exact product model not disclosed. Earlier same-day research context retained. No real patient information, clinical evaluation, security/data-flow audit, accessibility audit, longitudinal study, or vendor/patient interviews. One session does not establish general behavior. A second reviewer independently re-read the same evidence and both sets of terms and re-inspected the public chat page; a fresh-session replication attempt did not load outside the host iframe.
Draft profile · Medium draft for documentation and bounded observations. Low for generalization and sustained outcomes. The axis is an editorial placement.