Vivian Health

The AI Assistant is the engine of a core module layered on a marketplace that ran for years without it. The employer site says so itself: AI is only as strong as the data and humans behind it, and the assistant is built on top of a network of clinicians and recruiters and a data layer of job requirements, clinician intent and behavior, and placement outcomes. The assistant chats with applicants around the clock, screens them against each job's requirements, prepares them for submission and hands them to recruiters; since June 2026 it also generates candidate lists and personalized outreach for recruiters to review (AI Proposals). A chat agent that answers applicants and a writer that drafts outreach are functions whose working form is a trained model, and the privacy policy names a candidate ranking algorithm behind recruiter search.

The company reported in April 2026 that nearly half of applications on the platform are screened by the assistant, which is operational dependence rather than a feature on a page. The marketplace, salary data, employer reviews and the Vivian IQ pay analytics stand without the model.

The oversight structure is described and the part that decides who is dropped is missing. The company publishes what the assistant screens for by job type (travel, permanent, per diem and local, advanced practice), that it hands screened candidates to recruiters, that for permanent roles requirements are evaluated with flexibility and candidates are not disqualified unless the job explicitly requires it, and that AI Proposals leaves recruiters in control, reviewing, editing and removing candidates before sending.

The October 2025 launch post also says the assistant screens every candidate in real time, surfacing licensed and experienced professionals while filtering out the rest. What happens to a filtered candidate is not published: whether a recruiter can see her, whether the flexibility rule applies to travel, per diem and advanced practice roles as it does to permanent ones, and at what point the screen stops rather than asks a recruiter. A wrong pass costs a recruiter time; a wrong rejection costs a clinician a job and is seen by no one.

The inputs are described and nothing that produces the output is identified. The employer site names the data layer (detailed job requirements, clinician intent and behavior, placement outcomes), the product posts name what the assistant screens for by job type, and AI Proposals says its matches draw on specialty, experience, licensure, job type preferences and behavioral signals such as recent searches, previous screenings and prior applications. The privacy policy mentions a candidate ranking algorithm and an AI powered chat service.

No model, model family or provider is named for the chat, the screening, the ranking or the message drafting, nothing says whether screening is a learned model or rules over structured credential fields, and no version or release practice is published beyond dated product posts.

The data flows are described and no model provider is named. The privacy policy lists the categories of vendors that receive user information (hosting, fraud and abuse prevention, data analysis, push notifications, chat including AI, email and text messaging, including building promotional target lists) and names one, Very Good Security, which captures and tokenizes Social Security numbers. Candidate information also goes to recruiters and employers and to agencies with technical integrations, and advertising partners receive device and usage data to target the company's ads.

The AI chat vendor, the provider behind the assistant's language functions and the hosting provider are not named, and no subprocessor list is published.

Every figure is company measured and most are ceilings rather than averages. The employer site reports candidates responding up to 4 times more often to the assistant, a placement lift of up to 15 percent (footnoted as applying when the assistant is first to engage the applicant) and 20 percent more complete profiles; the April 2026 post reports up to a 10 to 15 percent increase in application to placement conversion and nearly half of applications screened by the assistant; the AI Proposals post reports beta proposals sent nearly 12 times faster with up to 3 times higher response rates, and recruiters visiting the talent pool 37 percent more. A Trinity Health talent acquisition director is quoted as saving more than $550,000 in the first three months, and AccentCare describes a same week hire.

None of these has a denominator, a comparison group or a period, and "up to" framing reports the best result rather than the typical one. No independent evaluation is published.

Identity documents are handled specifically; the data the AI uses is not. The privacy policy (effective 2 July 2026) says driver's license images are shown only to recruiters the candidate approves and kept for 200 days after sharing, and that a full Social Security number, when an integrated employer asks for it, is captured, encrypted and tokenized by Very Good Security before reaching the company's systems, held there for at most 30 days and deletable at any time. It says aggregated usage data is used to study job market trends and to develop new products.

Nothing states how long profiles, chat transcripts and behavioral signals are kept, whether they train the assistant or the ranking algorithm, or whether they are de-identified first; the policy lists AI chat among the vendors that receive user information. Profiles can include Covid 19 vaccination status. Safety engineering, meaning how a wrong or harmful assistant response is caught and handled, is not published. No patient information is handled.

A substantive privacy policy covers the service, and no business associate status or scope position is stated. The privacy policy (effective 2 July 2026) covers clinicians, references and employer users in detail, says the company acts as a service provider or processor under a data processing agreement when it works on an agency's or facility's instructions, and says nothing about protected health information. The platform serves clinicians rather than patients, and nothing describes patient data, but the company does not say so; a plain statement that the service handles no protected health information, with its scope explained, is what a buyer would rely on.

One item is health information in the ordinary sense: profiles can include Covid 19 vaccination status and associated documentation, which is shared with recruiters and employers.

No independent check of the company's security is published. The privacy policy describes controls: driver's license images in a secure database shown only to approved recruiters, and Social Security numbers captured, encrypted and tokenized by Very Good Security before reaching the company. No SOC 2 report, ISO 27001 certificate, penetration testing statement or trust page is named on the consumer site, the employer site, the privacy policy or the terms, and the footers link only to privacy and terms.

The data held makes the question concrete: license, certification, DEA registration and identity records for more than 2 million clinicians, which is useful for credential fraud, exchanged daily with agencies and facilities. Using a specialist vendor for Social Security numbers narrows one risk and says nothing about the rest.

No device claim is made and none applies: the product is a jobs marketplace and recruiting assistant, and nothing it sells touches patient care. What governs it is employment law, including rules on automated employment decision tools, such as New York City's requirement that an employer using one for roles in the city obtain and publish an independent bias audit. The company publishes no bias audit or candidate notice for the assistant, and its privacy policy states that it does not engage in profiling that produces legal or similarly significant effects.

Employment law is named here as what governs the product; it is not a device pathway. Because the company sells to staffing agencies as well as direct employers, which party carries an audit duty depends on who makes the hiring decision, and that should be settled in the contract.

Positions are stated and no evaluation is published. The privacy policy (effective 2 July 2026) says the company does not engage in profiling that results in decisions producing legal or similarly significant effects, and the April 2026 product post says that for permanent roles requirements are evaluated with flexibility and candidates are not disqualified unless the job explicitly requires it. The screening criteria are named by job type and are job related: licensure in the required state, board certification, DEA registration, recent clinical activity, specialty and experience, availability and prior facility relationships.

Nothing describes how the assistant or the ranking algorithm is tested, whether results differ across groups of clinicians, or any bias audit, and AI Proposals ranks candidates partly on behavioral signals (recent searches, previous screenings, prior applications). The launch post says the assistant screens every candidate and filters out the rest, which sits uneasily beside the profiling statement.

Review before use exists on the recruiter side, and nothing stands behind the output. AI Proposals lets recruiters review each proposal, edit messages, remove candidates and regenerate the list before anything is sent, and the assistant hands screened applicants to a recruiter rather than submitting them to an employer itself. The terms put the accuracy of profile content on the healthcare professional, disclaim any guarantee about statements made to employers, require individual arbitration and limit the company's liability.

For the applicant, no route reaches the assistant's output. The privacy policy offers access, correction and deletion of personal information and lists a right to opt out of profiling as not applicable because no significant decisions are made by profiling; nothing lets a clinician see why she was screened out or ranked low, or ask for the screening to be redone. No accuracy or error rate is published for screening or ranking.

No clinical record integration exists: this is a hiring marketplace. For its buyers, the systems of record that matter are applicant tracking and vendor management systems, and integration with them is claimed without naming any. The staffing agency page describes automatic job postings, instant ATS candidate syncing for qualifying plans and direct submissions to select vendor management systems; the facilities page says the company can support ATS integrations to push a candidate's universal profile; and the assistant requires a job feed carrying the VMS name and job ID.

The site names no applicant tracking or vendor management system, and nothing says which direction data flows beyond pushing candidate profiles.

A single hosted service, with location implied rather than committed. The privacy policy names categories of vendors that receive user information (hosting, fraud and abuse prevention, data analysis, push notifications, chat including AI, email and text messaging) and one by name, Very Good Security, which tokenizes Social Security numbers; it names no host, region or residency commitment. Where the company acts for an agency or facility it describes itself as a service provider or processor under a data processing agreement with that customer.

The platform holds license, certification, work history, vaccination status and identity documents for more than 2 million clinicians and exchanges candidate data with agencies, facilities and vendor management systems.

No price is published and the posture is partly discoverable. Both employer paths, direct healthcare employers and staffing agencies, lead to a demo request, and the service is free to clinicians. The agency page says ATS candidate syncing comes with qualifying plans, which shows that plans differ, and the AI Assistant needs an automated job feed with the VMS name and job ID; no plan names, unit of charge or figures are given.

The company's clinician proposition rests on pay transparency, with salary data, employer reviews and rate breakdowns on postings, and none of that is extended to the employer side.

Coverage is named in detail, with results behind part of it. The marketplace spans nursing (RN, CNA, LPN and LVN), allied health (therapy, technologists, social work) and advanced practice providers (nurse practitioners, physician assistants, CRNAs) across travel, permanent, local contract, per diem and locum work, sold to staffing agencies and direct employers. The assistant started with travel jobs in October 2025 and was extended in April 2026 to permanent, local, per diem and advanced practice roles, each with its own published screening workflow.

The published results (the placement and response lifts) come from staffing agencies and mostly predate the extension, so they describe travel recruiting; nothing is published for the job types added in 2026 or for direct employer hiring beyond a Trinity Health savings quote.