ASCO Guidelines Assistant

The Guidelines Assistant is a generative model over a corpus that exists without it. The product page describes an interactive AI tool developed by ASCO and Google Cloud that uses generative AI on Google Cloud's Vertex AI platform and Gemini models to answer oncology questions from ASCO's published clinical practice guidelines, with citations and follow up questions.

The model is the engine of this module and the surrounding platform carries value without it. The guideline library is built through systematic review and expert panels, is published and read independently, and is reached through the same website and member app the assistant sits in. Remove the model and the assistant is gone, while the guidelines remain exactly as authoritative as before.

The model was not developed by the society: it is a third party model family on a third party platform, which the society states openly.

The oversight structure is described and one part is missing. What the tool may do is bounded and published: it draws solely from ASCO's clinical practice guidelines, answers with citations and sources, and is updated as guidelines are published or revised. What it should do outside that bound is published too: the society's chief executive, in the society's AI in Oncology article of 25 September 2025, describes the assistant as designed to say it does not have an answer when the guidelines do not cover a question, rather than answer from elsewhere.

The product page and the AI tools notice set the clinician's role: the assistant is a voluntary educational resource for oncology professionals that does not account for variation among patients and does not substitute for the independent judgment of a licensed healthcare professional, and interactions are not actively monitored by health professionals.

What is missing is how the tool handles two guidelines that address the same situation differently, or a recommendation superseded by a newer one, and whether an answer names the guideline version it used. Nothing routes a case to a person, which the notice states plainly. No absolute accuracy claim is made; the product page says drawing only on the guidelines removes accuracy concerns associated with publicly available AI tools, a comparative claim rather than a promise that every answer is right.

The model family and platform are named, without the version and update discipline behind them. The product page states that the assistant runs on Google Cloud's Vertex AI platform with Gemini models and was developed by ASCO and Google Cloud; the launch announcement and the society's AI in Oncology article say the same.

The corpus is unusually well specified because it is a public artifact: the tool draws solely from ASCO's published clinical practice guidelines, updated as guidelines are published or revised, so a clinician can check the whole scope of what the system could have drawn on. The society's chief executive has described the design as constrained to the guideline documents and built to decline when they do not cover a question.

No Gemini model version is named, no retrieval or ranking method is described, no release notes say when the model or the index changed, and nothing describes how conflicting or superseded guidance is handled.

The model provider is named, and the rest of the party list and its terms are not. The product page states that the assistant was developed by ASCO and Google Cloud and runs on Google Cloud's Vertex AI platform using Gemini models, so a question typed into the tool reaches Google as the model and platform provider.

The AI tools notice adds that information entered, including prompts and uploaded files, may be shared with ASCO and third party vendors associated with providing the tools, without naming those vendors beyond Google. The general privacy policy names Google Analytics and Adobe Experience Cloud for website analytics and requires vendors that receive personal information to safeguard it. No subprocessor list is published, and no retention or training terms govern prompts once they reach the model provider.

The chain carries a clinician's question and feedback, not a patient record: the notice prohibits entering protected health information or patient information.

No evaluation of the assistant is published, and no performance claim a reader can check. The society's own AI in Oncology article on the assistant (25 September 2025) reports no usage figures, accuracy rate or evaluation results, and describes improvement through member feedback. The tool was demonstrated at the society's 2025 annual meeting, which is exposure rather than measurement. A PubMed search on 28 September 2026 for the assistant by name and for evaluations of ASCO guideline chat tools returned no study of this product. The product page's statement that drawing only on the guidelines removes accuracy concerns associated with publicly available AI tools carries no figure or method. No named institutional deployment exists, because access is an individual member benefit.

The underlying content has an exceptionally strong evidence process: systematic reviews, expert panels and a published guideline methodology. None of that is evidence about the retrieval and generation layer built on top of it. A rigorous guideline does not establish that a model answering questions about it returns the right passage or summarizes it without distortion.

The corpus is the one place where evaluation would be cheapest: the guidelines are public, finite and structured, the correct answer to a guideline question is usually determinate, and the society has the panel members who wrote them.

The society states plainly what happens to what a user types, and what it states is a disclosure rather than a protection. The AI tools notice says that information entered into ASCO AI tools, including prompts and uploaded files, may be shared with ASCO and third party vendors associated with providing the tools, that the tools are not a HIPAA protected environment, that users must never enter protected health information or patient information, and that the Guidelines Assistant is not a secure environment for individually identifiable or sensitive personal information and is not part of a medical record. The product page adds that user feedback will be used to refine the tool.

Nothing states how long prompts and feedback are retained, whether prompt content is used to train or tune any model, or what deidentification applies. The general privacy policy (last modified 11 September 2025), which covers the society's websites and mobile applications, retains personal information for as long as necessary for its business purposes and allows its use to improve services. The precise statement that inputs are shared and unprotected is candor about the practice; nothing in it protects the data, such as a bar on training use or a retention window.

No safety engineering is published beyond the corpus bound and the stated design to decline when the guidelines do not cover a question: no guardrails, no hallucination mitigation and no process for handling a safety event.

The society takes a plain scope position that the product does not handle protected health information, explains it, and has a service privacy document behind it. The AI tools notice states that ASCO AI tools are not a HIPAA protected environment, that users must never enter protected health information or patient information, that the Guidelines Assistant is not a secure environment for individually identifiable information and is not part of a medical record, and that anything entered may be shared with ASCO and third party vendors providing the tools. The assistant answers from published guidelines, takes nothing but a clinician's question, and has no record integration, so the stated scope covers what the product takes in.

The society's privacy policy (last modified 11 September 2025) covers its websites and mobile applications, which is how the assistant is reached, and a separate mobile app privacy policy is listed on the legal page. No business associate status is stated and none would be offered, as there is no protected health information in scope for an agreement to cover.

The practical exposure is the question typed in a hurry about a real case. The prohibition is published in the notice; the product page repeats only the educational disclaimer.

Controls are asserted with nothing independent behind them. The society's privacy policy says it uses industry standard technology and security measures, requires vendors that receive personal information to safeguard it, and has a specialized security team that regularly reviews and improves its measures. No SOC 2 report, ISO 27001 certificate, HITRUST certification, trust center, penetration testing statement or vulnerability disclosure program is published for the society or the assistant: the legal page lists privacy, mobile app privacy, terms of use, accessibility, conflict of interest and related policies, and none of them names an audit.

The assistant runs on Google Cloud's Vertex AI platform. Google's certifications belong to the host and do not attach to the society's own operation of the tool. The AI tools notice itself tells users the assistant is not a secure environment for individually identifiable or sensitive personal information.

A published statement of intended purpose, with no regulatory reasoning behind it. The AI tools notice states that ASCO AI tools are not intended to be used for the diagnosis, prevention, monitoring, prediction, prognosis, treatment or alleviation of disease, and describes the Guidelines Assistant as a tool designed to help clinicians navigate the society's clinical practice guidelines. The product page describes it as a voluntary educational resource for oncology professionals that does not account for variation among patients and does not substitute for the independent judgment of a licensed healthcare professional. Together these say what the product is for and for whom, on the product page and in a dedicated notice rather than only in terms of use.

The output is retrieved published guidance, with citations to documents available in full to any clinician, and the refusal to individualize to a patient marks the line between a reference tool and patient specific decision support. No clearance exists or is sought, and the society publishes no analysis of why the product sits outside device regulation.

The framing pulls two ways. The product page invites clinicians to make faster, evidence based clinical decisions, and the notice says the guidelines assist clinicians in forming treatment recommendations, while the same notice says the tools are not intended for treatment. The decision stays with the clinician in both framings.

Responsible AI is addressed in policy and disclosure language, with no evaluation of the assistant behind it. The society publishes six guiding principles for AI in oncology (2024), a policy position for the field rather than an evaluation of this tool, and its AI tools notice states that AI generated outputs may not be correct, that outputs are not a substitute for professional judgment, and that interactions with the assistant are not actively monitored by health professionals.

Two adjacent structures govern the content rather than the model. The guidelines follow the ASCO Guidelines Methodology Manual and the society's conflict of interest policy for guideline panels, which is real assurance about the corpus. And a feedback loop is disclosed: the product page says user feedback will be used to refine the tool and the development of guidelines over time. That creates a path from what clinicians ask a chat interface back toward the standard of care, and nothing published says how that input is weighted, who reviews it, or how it is kept apart from the formal evidence review that guideline development rests on.

No evaluation of retrieval accuracy, no error taxonomy, no bias assessment and no subgroup analysis is published. The bias question specific to this product is untouched: guideline recommendations are shaped by the evidence available to the panel, and a retrieval tool inherits every population gap in that evidence without surfacing it.

Mechanisms exist that let a clinician challenge an answer, and the society disclaims everything behind the output. Every response carries citations and sources, and the tool draws solely from published guidelines anyone can read, so a clinician can check not only where a statement came from but the whole body of material the system could have drawn on. Users can give feedback, and the society's chief executive has described members pointing out places where answers fell short.

The AI tools notice states that outputs are provided as is with no warranty, express or implied, that users are responsible for ensuring outputs are accurate and appropriate for their use, and that the society assumes no responsibility for any injury, liability or damage arising from use of the tools or their outputs, or for errors or omissions. The same notice says interactions are not actively monitored by health professionals.

No accuracy figure, error rate or correction commitment is published, and nothing says whether the answer names the guideline version it used or how superseded guidance is retired. Feedback is a door into the process, not a route that ends in a correction the society is bound to make.

No integration exists and none is claimed. The assistant is a web application reached through the society's site and a member mobile app. No record system, marketplace listing, application programming interface, health data exchange standard or in context launch from a record is described, and the AI tools notice states that the assistant is not part of a medical record and must never receive patient information.

That is a coherent design for a member benefit that takes no patient data, and nothing on the product page implies otherwise. It is also where the distribution model shows its cost: the stated purpose is helping oncologists make faster decisions, and those clinicians work inside an oncology record system all day, so a tool they must leave that system to reach competes with the record's own content rather than sitting inside it.

A single hosted option with location implied rather than committed. The assistant is reached through the society's website and a dedicated web application, and through the ASCO member app on both mobile platforms, behind an ASCO account login. It runs on Google Cloud's Vertex AI platform with Gemini models, which says whose infrastructure processes a query and not where.

No region, data residency position or processing path is published, and the AI tools notice says prompts may be shared with ASCO and third party vendors associated with providing the tools without saying where they sit. The society has a substantial international membership, with published dues for members in upper middle, lower middle and low income countries, and nothing says whether the assistant's processing or availability differs by country. No deployment for an institution rather than an individual is described.

Published tiers with figures, a stated unit of charge and a route to start without a sales conversation. The assistant carries no separate charge: the product page says it is available to ASCO members through asco.org and the member app and requires an ASCO account. The price of access is membership, and the membership dues page publishes annual dues by role and country: $695 for physicians and researchers or scientists, $348 for early career and retired members, $174 for affiliated health professionals, $104 for members in upper middle income countries, and $0 for students, fellows and residents, emeritus members and members in low and lower middle income countries, with 15 percent off for auto renewal and multi year discounts of 10 and 25 percent. An online application sets the rate; no sales contact is needed.

Two limits sit on the record rather than on the grade. There is no route at any price for a non member, and no institutional or enterprise license is published, so a cancer center wanting the tool across its staff has no published option. And membership is not available to people employed or primarily consulting in sales, marketing, financial planning, business development or media.

Coverage is stated clearly and nothing validates it yet. The assistant covers the society's full library of clinical practice guidelines, which is broad within oncology and narrow by design outside it, for oncology professionals who are society members, reached through the website and member app. The launch material and the society's AI in Oncology article describe the intended user as the busy oncologist, including community and rural practice where volume is high and subspecialty depth thin.

No evaluation is published for any tumor type, question class or practice setting, so the scope is a statement rather than a validated range.

The sharper scoping limit sits inside the specialty. Oncology practice draws on more than one society's guidance, on compendia and on trial protocols, and treatment decisions often sit at the boundary between them. The society's chief executive has said the assistant is designed to decline where its guidelines do not speak; how it behaves at the boundary, where a guideline speaks partly, is not described.