AI in Healthcare: what works, what doesn't, and what it costs (September 2026)
In 2026, AI in healthcare is mostly used for paperwork and pattern-spotting, not for making diagnoses on its own. Ambient scribes draft visit notes, inbox tools draft replies to patient messages, and FDA-authorized software flags findings on scans for radiologists. Each works best as a first draft or second reader that a clinician checks. Measured time savings are real but modest, and chatbot-assisted diagnosis still underperforms in controlled studies.
State of AI in Healthcare · September 2026
As of September 2026, AI in healthcare is ready for paperwork and for reading images alongside a radiologist, still early for patient messages and prior authorization, and not ready to diagnose patients on its own.
- Drafting visit notes (ambient scribes)
- Flagging findings on medical images
- Second reader for screening mammograms
Backed by trials or large deployments
- Drafting replies to patient messages
- Pre-screening prior authorization
Promising, limited outcome data
- Diagnosis with a general chatbot
- Patient self-triage by chatbot
- Notes used without clinician review
Evidence says avoid or add safeguards
What AI can reliably do in Healthcare today
- 1
Proven Draft clinical visit notes from recorded patient conversations for clinician review and sign-off.
Evidence
AMA, June 2025: The Permanente Medical Group's 7,260 physicians used ambient scribes in 2.58 million encounters (Oct 2023–Dec 2024), saving an estimated 15,791 hours. Source ↗
- 2
Proven Flag findings on medical images using FDA-authorized AI software, most of it in radiology.
Evidence
FDA's public AI-Enabled Medical Device List, with entries through June 2026, lists marketing-authorized AI devices; the large majority were reviewed under the radiology panel. Source ↗
- 3
Proven Triage screening mammograms and highlight suspicious areas, cutting radiologist reading workload without missing more cancers.
Evidence
MASAI randomized trial, The Lancet, Jan 2026, 100,000+ Swedish women: 81% vs 74% of cancers found at screening, 12% fewer interval cancers, similar false positives, 44% less reading workload. Source ↗
- 4
Early Draft replies to patient portal messages that clinicians edit before sending.
Evidence
JAMA Network Open, 2024 (Stanford 2023 pilot): clinicians used 20% of AI drafts; task-load and exhaustion scores fell, but measured reply time did not change. Source ↗
- 5
Early Pre-screen prior authorization requests against coverage criteria, with licensed clinicians making every denial decision.
Evidence
CMS WISeR Model (2026–2031, six states) uses AI/ML vendors for Medicare prior-auth reviews; all non-payment recommendations are decided by licensed clinicians. Outcome data not yet published. Source ↗
What it can't do, or where the risk is
- 1
Can't Improve physicians' diagnostic accuracy just by giving them access to a general-purpose chatbot.
Evidence
JAMA Network Open, 2024 RCT, 50 physicians: 76% diagnostic reasoning score with GPT-4 vs 74% without (not significant); GPT-4 alone scored 92%. Source ↗
- 2
Can't Safely guide patients' self-diagnosis and triage through general-purpose chatbots.
Evidence
Nature Medicine, 2026 RCT, 1,298 UK adults: LLMs alone identified conditions in 94.9% of scenarios, but people using them did so in under 34.5%, worse than controls. Source ↗
- 3
Can't Produce notes free of fabricated or missing details; every AI-drafted note still needs clinician review.
Evidence
npj Digital Medicine, 2025: across 12,999 AI-generated note sentences, 1.47% hallucination and 3.45% omission rates; 44% of hallucinations were judged major. Source ↗
- 4
Can't Handle identifiable patient data under HIPAA without a signed business associate agreement (BAA) with the AI vendor.
Evidence
HHS guidance (reviewed July 2026): covered entities may disclose PHI to a business associate only with satisfactory written assurances, i.e. a BAA. Source ↗
- 5
Can't Transcribe and interpret all patients equally well; speech recognition and clinical algorithms can carry bias.
Evidence
npj Digital Medicine, 2025 review of AI scribe risks: cites significantly higher speech-recognition error rates for African American speakers, plus omission and automation-bias risks. Source ↗
Where to start
Small clinic or practice
Pilot one ambient scribe under a signed BAA. Clinicians review and sign every note; compare documentation time before and after.
Hospital or health system
Add FDA-authorized imaging AI as a second reader where volume is high, such as screening mammography, and track reading time and recall rates.
Patient
Use a chatbot to prepare questions for your doctor, not to decide what you have or whether to seek care.
Regulation & risk
Laws, rules and official guidance that commonly apply when AI is used in Healthcare.
| Name | Applies to | Official source | When it applies |
|---|---|---|---|
| HIPAA (Privacy & Security Rules) | US | hhs.gov | Handling protected health information (PHI); a business associate agreement (BAA) is required |
| FDA — AI/ML-enabled Software as a Medical Device | US | fda.gov | Software used for diagnosis or treatment decisions |
| FTC Health Breach Notification Rule | US | ftc.gov | Breaches of health-app or wearable data not covered by HIPAA |
| EU MDR (Medical Device Regulation) | EU | health.ec.europa.eu | Software with a medical purpose |
| EU AI Act — high-risk (Annex III: health-related) | EU | digital-strategy.ec.europa.eu | Safety components of medical devices, emergency triage and similar uses |
| GDPR Art. 9 (special category: health data) | EU | commission.europa.eu | Any processing of health data |
| WHO — Ethics & governance of AI for health | International (guidance) | who.int | Reference for procurement and ethics review |
This list is a starting point for your own compliance review, not legal advice. Last reviewed 2026-09-24.
Cost & deployment reality
Verified starting prices for 4 of the tools below: $9.99 to $159 per month (median $85.9).
Ambient scribes are sold per clinician per month. Self-serve plans list $39 (capped notes) to $119 (Freed pricing, 2026); mid-range tools run about $100 or more, and enterprise platforms such as DAX Copilot and Abridge often cost several hundred dollars per clinician per month on multiyear contracts (Medical Economics).
Implementation, training and support may be billed separately. Financial return is thin: a five-site JAMA study found about $167 per clinician per month in extra revenue, partly from higher coding (Advisory Board, 2026), and PHTI (2025) found no clear financial impact. Clinician time to review every note is an ongoing cost.
US deployments need a HIPAA business associate agreement, contract limits on training models with encounter data, and correct de-identification if vendors claim it (Foley, 2025). The 2024 Section 1557 rule requires providers to identify and mitigate discrimination risk from decision-support tools (NHeLP).
Budget scribes often have limited EHR integration, so notes are copied in by hand. A typical failure pattern is unchecked notes: in a 2025 NEJM AI trial, physicians reported occasional clinically significant inaccuracies, mostly omissions and pronoun errors, and one mild patient-safety event (UCLA Health).
Scenarios and tools for Healthcare
Clinical documentation
| Tool | Pricing model | Free tier | Starting price | Compliance claims | Official site | Last verified |
|---|---|---|---|---|---|---|
 S10.AI | Paid | No free plan | $159 / month | 2 : HIPAA, GDPR | Online (2026-09-24) | 2026-09-26 |
Admin & workflow automation
| Tool | Pricing model | Free tier | Starting price | Compliance claims | Official site | Last verified |
|---|---|---|---|---|---|---|
 Keragon - AI healthcare platform | Paid | No free plan | $149 / month | 2 : HIPAA, SOC 2 Type II | Online (2026-09-24) | 2026-09-26 |
Clinical evidence search
| Tool | Pricing model | Free tier | Starting price | Compliance claims | Official site | Last verified |
|---|---|---|---|---|---|---|
 EvidenceHunt - AI-Powered Clinical Evidence Search | Freemium | Yes: Ask unlimited questions and use low-cost models (+ 3 PRO-quality questions); 1 evi… | $22.8 / month | 1 : GDPR | Online (2026-09-26) | 2026-09-26 |
Patient questions & pharmacy
| Tool | Pricing model | Free tier | Starting price | Compliance claims | Official site | Last verified |
|---|---|---|---|---|---|---|
 Medidex Connect | Free | Yes: Ask a Pharmacist for Free (homepage CTA); no paid plans shown. | Not verified | Not verified | Online (2026-09-24) | 2026-09-26 |
Care navigation
| Tool | Pricing model | Free tier | Starting price | Compliance claims | Official site | Last verified |
|---|---|---|---|---|---|---|
 AI hospital recommendation - AI in healthcare | Not verified | Not verified | Not verified | Not verified | Online (2026-09-24) | 2026-09-26 |
Consumer imaging apps (not FDA-authorized)
| Tool | Pricing model | Free tier | Starting price | Compliance claims | Official site | Last verified |
|---|---|---|---|---|---|---|
 X-ray Interpreter - AI-generated report instantly | Usage-based | Yes: Enjoy free preview reports when your balance is zero. One credit equals one analysis. | Not verified | Not verified | Online (2026-09-24) | 2026-09-26 |
Consumer health & wellness
| Tool | Pricing model | Free tier | Starting price | Compliance claims | Official site | Last verified |
|---|---|---|---|---|---|---|
 Catalyst Healthspan Audit | Free | Yes: Healthspan Audit — 'Free 3-min Score' (page title). | Not verified | Not verified | Online (2026-09-24) | 2026-09-26 |
 BRI Calculator | Free | Yes: Free Body Roundness Index (BRI) Calculator — 'Free, Fast, and Revealing.' | Not verified | Not verified | Online (2026-09-24) | 2026-09-26 |
 Riley | Freemium | Yes: Riley Basic: Always free (for solo parenting or simpler setups). | $9.99 / month | Not verified | Online (2026-09-24) | 2026-09-26 |
Vendors & implementers
No vendors listed yet.
Recent changes: Healthcare tools
Last verified: 2026-09-25 · auto-checked 2026-09-26 · Reviewed by Toolsfine editorial
- Catalyst Healthspan Audit · AI alternative note · — → Not an AI tool; a lead-generation qui… ·
- Catalyst Healthspan Audit · editor verdict · — → Catalyst Performance Healthspan Audit… ·
- Catalyst Healthspan Audit · pricing page URL · — → https://catalystperformance.sg/health… ·
- Catalyst Healthspan Audit · free tier · — → {"exists":true,"limits":"Healthspan A… ·
- Catalyst Healthspan Audit · pricing model · — → Free ·
- BRI Calculator · AI alternative note · — → Not an AI tool; the BRI formula is pu… ·
- BRI Calculator · editor verdict · — → BRI Calculator: free web calculator f… ·
- BRI Calculator · pricing page URL · — → https://bri-calculator.com/ ·