ClaireMed

Claire's value is the interview and the draft, and both come from the model. It conducts the patient interview itself, choosing each follow up question from the patient's previous answer, then turns the answers into a structured history and drafts the note in the physician's preferred format. Take the model out and there is no intake, no history and no draft left to sell.

The forms work rests on the same capability: supported clinician forms, including Ontario WSIB and immunization documentation, are prepared from the intake rather than filled by hand. The vendor labels messages, prescriptions, referrals and post visit follow through as in development, so the product today is intake, history, draft and forms.

Claire drafts and the physician decides. The vendor states it does not diagnose, prescribe or sign, never finalizes a note on its own, and presents every draft for the physician to review, edit and sign before it enters the record. The terms make the same point in the contract, placing review and approval of every note, order, referral, coding suggestion and patient message on the clinician. The browser extension follows the same pattern, showing fill suggestions for review before they are applied to a page.

The missing piece sits at the front of the workflow. Claire interviews patients on its own before the visit, and nothing published says what happens when a patient reports something urgent during that interview, such as whether the intake stops, flags the answer or alerts the clinic. For the roadmap workflows the vendor describes pre approved actions only, with exceptions escalated to a person; none of those workflows is live yet.

The approach is described and nothing under it is identified. The vendor explains that the interview adapts each question to the previous answer and that drafts are grounded in the intake Claire conducts rather than generated from a loose prompt. No model, model provider, version or update practice is named, and no accuracy figure is published.

The FAQ's grounding answer reads as reassurance about hallucination, and the terms set the expectation more precisely: they disclaim any warranty that AI outputs will be free of hallucinations, omissions or inaccuracies. Where the two differ, the terms are the binding text.

The hosting layer is named and the model layer is not. Claire runs on Google Cloud Platform under a signed BAA, and the vendor says every third party that handles PHI is bound by a BAA or equivalent agreement and reviewed for security. No subprocessor list is published and no model provider is named, so who receives a patient's intake answers in order to generate the history and draft cannot be established from public material.

The training terms are clear on the company's side. Customer data containing PHI does not train its foundation models or any general model, and only deidentified or aggregated data may be used to improve the service. If an outside model provider is involved, the terms that govern data once it reaches that provider are not published. Payments run through Stripe and website visitor identification through RB2B, and the privacy policy says neither touches clinical data.

The evidence is a vendor survey with no method behind it. The Canada page reports that 87.5 percent of surveyed primary care physicians using Claire agreed it improves the quality of care they deliver, along with under three minutes from the end of intake to a drafted SOAP note and under one minute to connect. No sample size, date, survey instrument or clinic count is given, and a figure of 87.5 percent can come from as few as eight respondents.

The company names no deployment and publishes no peer reviewed study or independent evaluation. Its own pages also pull in two directions. The cost calculator starts every time saving assumption at zero and states that Claire does not claim a standard amount of time saved, while the FAQ says the documentation backlog that pushes charting into the evening goes away.

One contract term matters for anyone hoping to produce evidence: the terms of service bar customers from benchmarking the service for any public or comparative purpose, which closes one route to independent results.

Retention, training and deidentification are each stated in the terms of service, specifically enough to hold the company to.

On training, section 5.4 says customer data containing PHI or personal health information is not used to train its foundation models or any generally applicable model. Deidentified and aggregated data may be used to improve the service, with deidentification defined by reference to the HIPAA standard and section 47 of Ontario's PHIPA.

On retention, section 5.5 says encounter audio is deleted once the documentation is generated and quality checked, and transcripts, notes and other outputs are kept for the subscription plus a period the customer chooses in the product, from immediate deletion up to the legal medical record minimum. Customers can export or request deletion at any time, and data stays available for export for 30 days after termination before deletion under the BAA.

Two gaps remain. Nothing is published on guardrails against fabricated content or on how a safety event is handled, and the terms disclaim any promise that outputs are free of hallucinations, omissions or inaccuracies. And the terms describe audio recordings of encounters while the product pages present Claire as an intake tool rather than a scribe, so which workflows record audio is not stated.

The contract states business associate status, and the agreement itself is not published. The terms of service say the company acts as a business associate wherever it handles protected health information, incorporate the BAA by reference and make it control any conflict over PHI. The privacy and compliance page says a BAA is signed with every practice before PHI is handled, commits to minimum necessary use, and sets breach notification at no later than 60 days. There is one plan, so no tier question arises.

The Canadian half is documented to the same depth. The company acts as an agent of the health information custodian under PHIPA and a service provider under PIPEDA, and the terms say the BAA is structured to meet the equivalent obligations, with a PHIPA and PIPEDA addendum on request.

The BAA and a security overview are offered on request. Until the agreement text can be read, its scope and its terms for subcontractors cannot be checked.

Controls are described and nothing independent stands behind them. The compliance page lists encryption in transit and at rest, least privilege access with multifactor authentication and regular review, logging and monitoring of all PHI access, a written incident response plan and vendor due diligence. No SOC 2 report, ISO 27001 certificate, HITRUST certification, penetration test or trust portal is published, and the site's HIPAA, PHIPA and PIPEDA badges are compliance claims rather than audits.

The service runs on Google Cloud Platform, whose certifications cover the hosting provider rather than the company's own operations. A security overview is offered on request rather than published.

Claire is positioned as a documentation tool rather than a device, and what ships fits that position. The terms of service open by stating that Claire is not a medical device, does not render a diagnosis and does not replace clinical judgment, and the FAQ says the same in plain words. The live product is intake, a structured history and a draft note the physician reviews and signs.

Two details sit around that position. The terms define the service more broadly than the product pages, including order and referral management, inbox triage and coding suggestions, and the roadmap adds prescriptions and referrals. Those would change what a clinician relies on; the vendor says they will run only inside clinician approved workflows. And no regulator is named: no Health Canada position is stated, and the Canada page points clinics to CPSO and Ontario privacy commissioner guidance on AI scribes as things to review, without describing how Claire meets them.

Governance appears as commitments with no evaluation behind them. The about page says every capability is designed around clinical safety and privacy, the enterprise page lays out a pilot model with a clinical owner, structured feedback and agreed stop criteria, and the privacy policy commits, under Quebec's Law 25, to tell people about any decision made by automated processing alone, adding that Claire makes none today.

No fairness statement, subgroup performance or evaluation method is published. That gap weighs more for a patient facing intake than for a tool only clinicians use. The quality of the history depends on how well the interview works for patients of different ages, languages, literacy levels and conditions, and no results by group are published. The languages the intake supports are not stated.

Review before signing is the mechanism, and nothing stands behind the output itself. The terms of service provide the service as is, disclaim any warranty that AI outputs will be free of hallucinations, omissions or inaccuracies, and cap liability at twelve months of fees. The company's indemnity covers United States patent, copyright and trade secret claims only, while the customer indemnifies the company for claims arising from its clinical decisions. The one warranty is that the service performs materially as documented, with a fix or a pro rata refund as the remedy.

Clinical responsibility sits with the clinician by design. Every note, order, referral and message must be reviewed and approved before it is finalized, and the FAQ names that review step as the safeguard. Disputes go to arbitration in Ontario, while the same section names Delaware courts for anything outside arbitration, a mismatch a buyer's counsel will want resolved.

Claire is designed to sit beside the record rather than connect to it. The vendor calls it EMR agnostic and says it fits alongside Accuro, Oscar and TELUS PS Suite without becoming the system of record, and the FAQ says the physician finalizes the drafted SOAP note and places it in the chart. No integration with any named system, marketplace listing or interface standard is documented.

The one connecting mechanism is the Claire-Note Chrome Extension, described in the privacy material. When a clinician starts a scan or fill, it reads the visible patient context and field labels on the current EMR page and suggests entries, which the clinician reviews before they are applied. That gives a write path into browser based EMRs through the clinician's screen, with no system named for it and nothing to verify how it performs. Since the extension works on browser based forms, a clinic on a desktop EMR moves the note by hand.

Claire is web based, runs on Google Cloud Platform under a signed BAA, and states its location loosely. The privacy policy says the company is headquartered in Canada and processes personal information in the United States and Canada, and that data may be stored outside the province or country where a patient lives, including in the United States.

No residency commitment is made. The compliance page says residency options are available on request, and the FAQ answers the storage question by asking practices with a requirement to raise it in a demo. Tenant isolation is not described. For an Ontario custodian that needs records kept in Canada, residency is a term to put in the contract rather than something the published material settles.

The price is published in full and the route to start runs through a demo. A dedicated pricing page sets one plan at $499 USD per provider per month or $4,999 USD per provider per year, and spells out that both figures are in US dollars and charged per provider rather than per clinic. A buyer can size the cost of a practice before any conversation.

What a buyer cannot do is start without talking to the company. Every call to action leads to a demo request, no trial or self serve signup is published, and the pricing page asks buyers to confirm taxes, billing terms and setup before purchase. The terms add that fees are non refundable except as stated, may rise at renewal with 30 days' notice, and renew automatically.

Coverage is stated clearly and nothing validates it yet. The FAQ says Claire is built for primary care first, with the same approach adapting to other ambulatory settings on request. The how it works page names the appointment types supported today, from diabetes follow ups and anesthesia pre op questions to neurology assessments, and the forms it prepares are Canadian and partly provincial: Ontario WSIB forms and immunization documentation.

The only evidence of use is the primary care physician survey, which carries no sample or method. Languages are not stated anywhere, which matters more here than for an in room scribe, because the patient is the one answering the intake. Markets are Canada and the United States.