Rad AI
Rad AI is a generative AI platform for radiology that drafts report impressions, builds full structured reports from free dictation and automates patient follow-up on incidental findings, sold to practices and health systems by demo.
What is Rad AI?
Rad AI is an American software company, founded in 2018 and based in San Francisco, that builds generative AI specifically for radiology. It was co-founded by Dr Jeff Chang — the youngest radiologist in United States history, who spent ten years reading emergency studies overnight before moving into machine learning — and Doktor Gurson, a serial technology entrepreneur. The company sells three complementary products rather than a single tool.
Rad AI Impressions writes the impression section of a report directly from the findings a radiologist has just dictated, in that radiologist's own phrasing and their practice's conventions, and returns it in 0.5 to 3 seconds. Rad AI Reporting goes further and produces the complete structured report from free dictation, replacing an existing reporting platform; the company claims up to 50% less dictation time and up to 90% fewer spoken words, with a feature called Omni Unchanged that pulls stable findings from a prior study in one phrase. Rad AI Continuity handles what happens after the report: it detects follow-up recommendations for significant incidental findings, classifies them by category and anatomy, sets the due date from the recommended interval and consensus guidelines, then automates outreach to the patient by SMS, post and EHR message and to the referring provider by EHR message or fax, all tracked on a central dashboard with an audit trail.
The pitch to hospitals is that nothing changes for the radiologist: the existing microphone, PACS, RIS, EHR, templates and macros stay in place, and the automation is described as zero-click. For IT, the platform is cloud-native and standards-based, using OIDM and HL7 FHIR, with a lightweight client, SSO and no servers or virtual machines to provision.
Adoption is genuinely wide. Rad AI says it works with more than 40% of all US health systems and nine of the ten largest US radiology practices, with solutions supporting providers responsible for close to half of American imaging volume. Headline figures include roughly a billion fewer words dictated, 60+ minutes saved per shift and 84% of users reporting reduced burnout. Access is by sales demo only; no pricing is published anywhere on the site.
What it does
- Generate the impression section of a radiology report automatically from dictated findings
- Produce a complete structured report from free dictation, in the radiologist's own language
- Insert consensus guideline recommendations automatically, such as Fleischner, incidental thyroid nodules and AAA
- Detect and classify follow-up recommendations found in radiology reports
- Trigger and track follow-up communication to both the patient and the referring provider
- Flag clinically significant errors in dictated findings, in roughly 5% of reports
- Pull stable findings from a prior report with a single spoken phrase
When to use Rad AI / When not to
A quick filter to help you decide if Rad AI is the right fit.
When to use Rad AI
- Radiologists reading high volumes, especially on night, emergency or teleradiology shifts
- Radiology practices and health systems trying to cut turnaround times and physician burnout
- Organisations planning a migration away from PowerScribe 360 before its announced end of life
- Follow-up and care coordination teams responsible for tracking significant incidental findings
- Hospital IT directors who need a lightweight rollout with no servers or virtual machines
When not to use Rad AI
- Individual users and consumers: everything goes through a sales demo, with no self-service sign-up
- Anyone wanting to test before committing, since no free plan and no free trial are advertised
- Specialties outside radiology, as all three products act on the imaging report
- Teams outside North America, where deployments are American apart from one Canadian partnership
- Buyers looking for image-based detection, because Rad AI works on report text rather than on pixels
How to use Rad AI
A typical end-to-end flow, from setup to results.
- Request a demonstration through the form on the Rad AI website, as there is no self-service sign-up
- Agree a contract with the sales team, since scope and pricing are negotiated rather than published
- Connect the platform to the existing PACS, RIS and EHR systems, using OIDM and HL7 FHIR standards
- Roll out the lightweight client to radiologist workstations, with SSO and optional auto-update
- Keep existing microphones, templates, macros and free-dictation habits unchanged
- With Impressions, dictate the study indication and findings using the usual voice recognition software
- Let the impression generate in 0.5 to 3 seconds, then review and sign off the final report
- With Reporting, dictate findings and let the platform assemble the full structured report before signature
- For Continuity, connect EHR or RIS HL7 feeds so follow-up recommendations are detected and scheduled
- Monitor follow-up progress, communications and the audit trail from the central dashboard
Pros & Cons
Pros
- Concrete, repeatedly quantified gains: 60+ minutes per shift, 35% to 90% fewer words dictated
- No workflow change required, as existing microphone, PACS, RIS, EHR and templates are preserved
- Deployment described by CIOs as their easiest, with no server or virtual machine to provision
- SOC 2 Type II HIPAA+ certification, rolling 12-month third-party audits and over 130 daily tests
- A de-identification pipeline built specifically for radiology reports
- Exceptionally broad market validation across US health systems and large radiology practices
- End-to-end coverage from the dictated report through to closing the patient follow-up loop
Cons
- No public pricing at all and no pricing page, so every evaluation starts with a sales demo
- No free plan and no free trial are advertised anywhere
- No terms and conditions are readable: the footer link is dead and /terms returns a 404
- The privacy policy is a generic template centred on the website, not on clinical data handling
- No DPA or business associate agreement is mentioned publicly, and no retention period is quantified
- No documented way for a customer to exclude their data from model training
- No mobile app, no public support email or help centre, and availability is effectively North American
Pricing & Plans
Rad AI does not publish any pricing. There is no pricing page on the site and none appears in its sitemap, so no entry-level amount or currency can be stated. No permanent free plan and no free trial are advertised. Commercial terms are negotiated after a demonstration request, which is the only route of access. Prospective buyers should note that the amounts shown on the Rad AI Continuity page are the sliders of a return-on-investment calculator representing net reimbursement per imaging study, and are in no way a price for the product.
- Rad AI Impressions — automated generation of the report impression from dictated findings
- pricing on request
- Rad AI Reporting — full radiology reporting platform built on free dictation
- pricing on request
- Rad AI Continuity — detection
- communication and tracking of patient follow-up
- pricing on request
- the offer is structured by product and contracted after a demo
Data, GDPR & hosting
A consolidated view of how Rad AI handles your data.
GDPR overview
The privacy policy carries a dedicated European Union section which, for EU users, is stated to supersede any conflicting wording elsewhere in the document. It lists five legal bases — consent, performance of a contract, legal obligation, public interest and legitimate interest — and eight rights: withdrawing consent, objecting, access, rectification, restriction, erasure, portability and lodging a complaint with a supervisory authority. Requests are free and answered within one month, and objection to direct marketing is possible at any time without justification. Two gaps matter: no Article 27 EU representative is designated and no data protection officer is named, the only channel being privacy@radai.com. The company and its processing operations sit in the United States.
Who owns the data?
The published policy names Rad AI, at 548 Market St, PMB 49792, San Francisco, CA 94104-5401, as owner and data controller, reachable at privacy@radai.com. It covers website data — contact forms, newsletter, analytics — and states that data may be handled by internal staff and by appointed external processors such as hosting, IT and mail providers, whose updated list can be requested at any time. Users remain responsible for any third-party personal data they publish or share through the site. Crucially, the document says nothing explicit about ownership of the clinical reports the products process for hospitals; that question is settled in customer contracts, which are not published.
Reuse rights
Nothing in the published documents grants an end user the right to reuse data without asking. The policy is written from the controller's side: it lists why Rad AI collects website data — replying to enquiries, analytics, newsletters, CRM and lead management — and names the processors involved, all of them processing in the United States. There is no licence clause, no reuse permission and no terms and conditions page at all, since the footer link is dead and /terms returns a 404. On the product side, Rad AI describes proprietary models trained for radiology on one of the world's largest report datasets, plus a de-identification pipeline specialised to radiology reports, without ever saying whether customer data feeds that training or how a client could refuse it.
Data retention & training
Hosting summary
The published privacy policy states a place of processing of the United States for every named third-party service, and says data is processed at the owner's operating offices and wherever the parties involved are located. The processors named are Google LLC, Amplitude Inc., HubSpot Inc., Basin, Webflow Inc. and Cloudflare Inc., and an updated list of processors can be requested from the owner at any time. Cloudflare is flagged as distributing content across countries, which the policy admits makes the exact location of some data difficult to determine. Two limits should be kept in mind: this describes the data collected through the marketing website — forms, newsletter, analytics — and no hosting jurisdiction is published for the clinical report data the products process for hospitals. No broad region such as the EU or APAC is named, and no residency option is advertised.
Where Rad AI works
Country-level availability.
Available in
Things to keep in mind
Risks and trade-offs to weigh before adopting Rad AI.
- No terms and conditions are accessible before contact: the footer link points nowhere and /terms returns a 404, so the contractual relationship is invisible until sales engagement
- No DPA or business associate agreement is published, even though the software processes patient report data — this must be demanded in writing during procurement
- The site never says whether customer data trains its models, and offers no opt-out, so a practice cannot assess that exposure from public information alone
- Every efficiency and quality figure is vendor-reported, with no independent peer-reviewed study cited on the site; treat them as marketing claims until validated locally
- Automation bias is a real clinical risk: an impression that arrives in under three seconds invites a lighter review, and the radiologist remains responsible for what they sign
- Long-term reliance on generated impressions may erode the habit of synthesising findings, a skill trainees in particular still need to build
- The compliance standards listed on the security page are shown as one block without saying which are held by Rad AI and which by its hosting provider
Setup & Integrations
Technical difficulty
Low for the radiologist, moderate for the institution. Nothing changes at the workstation: the same microphone, templates, macros and dictation habits are kept, and the client is lightweight, supports SSO and can be rolled out in minutes with no server or virtual machine. The real work sits with IT, which must connect the platform to existing PACS, RIS and EHR systems using OIDM and HL7 FHIR, and, for Continuity, wire up HL7 feeds. This is an institutional integration project rather than a self-service install, so expect procurement, security review and clinical change management.
Deployment
Behind Rad AI
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Rad AI.
Frequently asked questions
What does Rad AI actually do?
How much does Rad AI cost?
Is there a free trial or a free plan?
Do radiologists have to change the way they work?
How much time does it save?
Is it secure and HIPAA compliant?
Is Rad AI GDPR compliant?
Is customer data used to train the models?
What does deployment involve for an IT team?
Is Rad AI available outside the United States?
Should you pick Rad AI?
Rad AI is one of the rare vertical AI products that has clearly crossed from promise into routine use. Working with more than 40% of US health systems and nine of the ten largest American radiology practices is not a pilot footprint, and the three products fit together sensibly: Impressions removes the most repetitive part of dictation, Reporting extends that to the whole document, and Continuity picks up the incidental finding that would otherwise be forgotten once the report is signed. The engineering choices — keeping the existing microphone, PACS, RIS and templates, shipping a lightweight client with no servers to provision — explain why adoption has been fast where other clinical AI stalls.
The reservations are commercial and contractual rather than technical. Nothing about the terms of the relationship is public: there is no price, no rate card, no readable terms and conditions, and no published data processing or business associate agreement, which is a striking gap for a vendor whose software reads patient reports. The privacy policy is a generic generated template describing the marketing website, not clinical data handling, with no quantified retention period and no stated position on whether customer data trains the models. Every performance figure quoted comes from Rad AI itself.
For a US or Canadian radiology practice or health system with the ability to run a proper procurement, this is a serious and well-funded option that deserves evaluation, and the questions above are exactly the ones to put in writing during that process. For anyone outside North America, for an individual radiologist, or for a buyer who needs to compare costs before booking a call, this is not currently an accessible product.
- Choosing a selection results in a full page refresh.
- Opens in a new window.