A doctor searching for a clinical answer in 2026 is no longer choosing only between Google, UpToDate and a local guideline folder. They may now encounter AI scribes, AI evidence engines, guideline-first assistants, local pathway search tools, medical exam platforms, clinical calculators, and general-purpose AI assistants. These products overlap in marketing language, but they solve different problems. Choosing the right one requires understanding what each category actually does — and what it does not.
The scale of use is also increasing. iatroX has processed hundreds of thousands of clinician queries and interactions, while Heidi Evidence reports nearly 2 million Evidence queries across Heidi markets since launch. These figures show that clinicians are willing to use AI for clinical knowledge tasks, but they do not by themselves answer the deeper question: which tool should be used for which clinical job?
Clinical AI Is Not One Category
The phrase "clinical AI" now covers products that serve fundamentally different workflows. A tool that documents consultations (a scribe) is not the same as a tool that retrieves clinical evidence (a search engine). A tool trained on global peer-reviewed literature is not the same as a tool built around UK national guidelines. A traditional reference product with expert-authored content is not the same as an AI-generated synthesis of multiple sources.
Grouping these together under "clinical AI" obscures the distinctions that matter most for clinical practice — and for clinical safety. The clinician choosing between them needs to understand what job each tool is designed to do.
The Main Clinical Jobs These Tools Solve
Clinical documentation and scribing. Ambient listening during consultations, structured note generation, clinical coding, referral letter drafting. Tools in this category include Heidi Health (which also includes Heidi Evidence as a clinical knowledge feature within the broader care-partner platform), Tortus AI (NHS-integrated, writes back to EMIS and SystmOne), Tandem Health (200,000+ NHS clinicians via Accurx, embedded in Doctor Care Anywhere), Nuance DAX/Dragon Copilot (Microsoft ecosystem), Freed AI, and Abridge. These tools sit inside the consultation — they capture what happens during the clinical encounter.
Clinical evidence retrieval. AI-assisted search across peer-reviewed medical literature, returning cited answers to clinical questions. OpenEvidence is the US benchmark — $12 billion valuation, approximately 18 million consultations per month, trained exclusively on peer-reviewed literature. Telecare Aware reported in April 2026 that OpenEvidence withdrew its app from the UK and EU, citing regulatory uncertainty. Heidi Evidence is the evidence-retrieval feature within Heidi's broader platform, reporting nearly 2 million queries across Heidi markets since launch. These tools sit around clinical uncertainty — answering the questions that arise before, during, or after a clinical decision.
Guideline-first clinical answers. Tools that retrieve and synthesise national clinical guidelines — NICE, CKS, BNF, SIGN in the UK; AWMF/S3 in Germany — rather than global peer-reviewed literature. iatroX retrieves cited clinical answers oriented around UK practice, grounded in UK authoritative sources. Praxis Medicine (Balderton/Creandum-backed, 70M SEK raised, UK-source-positioned) is an emerging entrant in this category. Umbil retrieves from NICE, CKS, SIGN, and BNF with clinical workflow tools (referral letters, SBAR, discharge summaries). These tools solve the "what does NICE say?" question that UK clinicians ask dozens of times per week — a question that global evidence engines are not optimised to answer.
Local and institutional pathway search. Tools that integrate local Trust policies, formularies, and antimicrobial guidelines alongside national guidelines. Medwise AI deploys at NHS Trust level with local policy integration — the HRA has listed a prospective pilot comparing Medwise against manual hospital intranet search. These tools solve the institutional workflow problem: "what does my Trust's guideline say, and how does it relate to the NICE recommendation?"
Traditional expert-authored clinical reference. UpToDate (7,600+ expert authors, 13,000+ topics, $500+/year), BMJ Best Practice (structured clinical decision support with evidence-graded recommendations), and DynaMed (point-of-care evidence summaries from EBSCO). These are not AI-generated — they are expert-curated and editorially reviewed, representing decades of physician authorship. UpToDate Expert AI adds a generative layer on top of curated content; BMJ Best Practice and DynaMed remain primarily editorial products.
Exam preparation and structured learning. Q-banks with adaptive algorithms, spaced repetition, and performance analytics. iatroX includes 15+ adaptive exam Q-banks covering UK, US, Italian, and international exams. AMBOSS combines curated clinical content with AI Mode search and structured learning pathways. Passmedicine covers UK exams. UWorld covers USMLE. These tools sit alongside clinical practice — building and maintaining the knowledge base that underpins clinical decisions.
Clinical calculators and scoring tools. iatroX calculators include 80+ clinical scoring tools with editorial content and guideline references. MDCalc is the established US calculator platform. These tools sit at the point of clinical decision — quantifying risk, severity, and probability to inform management.
General-purpose AI assistants. ChatGPT, Claude, Gemini, Perplexity. Powerful for evidence synthesis, writing, brainstorming, and administrative tasks. Not optimised for clinical use, not jurisdiction-specific, and carrying hallucination risk on clinical content. Useful — but requiring careful verification for any clinical application.
Search Is Different from Documentation
This distinction is fundamental and frequently confused. A scribe answers: "Can you document what just happened?" A search tool answers: "Can you help me find the right information for what is about to happen — or what I am deciding right now?" These are different cognitive tasks, different moments in the clinical workflow, and different product categories.
A clinician may need both. Tortus or Tandem to document the consultation. Ask iatroX to check whether the management plan aligns with the NICE guideline. The scribe writes the note. The knowledge tool verifies the plan. They are complementary, not competing.
Evidence Retrieval Is Different from Guideline Execution
A peer-reviewed literature search retrieves what studies have found. A guideline-first answer retrieves what national bodies recommend based on that evidence — synthesised, contextualised, and adapted for the specific healthcare system. These are not the same output.
A UK GP asking "what is the first-line treatment for hypertension?" needs the NICE NG136 recommendation — not a synthesis of the ACC/AHA guidelines, the ESC guidelines, and a meta-analysis that may not reflect UK practice. A tool optimised for peer-reviewed literature may surface excellent evidence that does not match UK prescribing norms. A guideline-first tool surfaces the recommendation the GP actually needs.
Both types of retrieval are valuable. But clinicians should know which one they are getting.
Citations Are Necessary, but Not Sufficient
Every clinical AI tool should provide citations — source links that the clinician can verify. But citation quality varies enormously. A citation that links to the specific NICE guideline paragraph is more useful than one that cites "NICE guidelines" generically. A citation that links to the exact BNF monograph section is more useful than one that says "per BNF."
Beyond citation presence, clinicians should evaluate citation accuracy (does the source actually say what the AI claims?), citation currency (when was the cited source last updated?), and citation relevance (does the cited source apply to UK practice, or is it from a different jurisdiction?).
Why Geography Matters: UK, US, EU and Local Pathways
Clinical evidence is global. Clinical guidelines are local. The pathophysiology of heart failure is the same in London and Los Angeles. The recommended first-line drug, the treatment threshold, the screening protocol, and the referral pathway may not be. NICE NG106 governs UK heart failure management. ACC/AHA governs US management. ESC governs European management.
A clinical AI tool that cannot distinguish between UK and US guidelines is not optimally safe for UK clinical practice — even if the underlying model is clinically competent. For UK clinicians, guideline localisation is not a feature request. It is a clinical requirement.
How to Evaluate a Clinical AI Tool Before Relying on It
Seven questions to ask before trusting any clinical AI tool with clinical work.
What clinical job does this tool solve? Documentation, evidence search, guideline retrieval, local pathway search, exam preparation, calculation, or general assistance? Match the tool to the task.
What sources does it draw from? Peer-reviewed literature? National guidelines? Local Trust policies? All three? Is the source base disclosed?
Can I verify the output? Are citations specific enough to check in seconds? Or vague references that require manual searching?
Is it relevant to my jurisdiction? Does it cite UK guidelines for UK queries? Or might it default to US or international recommendations?
How does it handle uncertainty? Does it acknowledge limitations? Decline to answer when evidence is insufficient? Or generate confident-sounding responses regardless?
What is the business model? Subscription? Ad-funded (pharma advertising)? Enterprise licensing? Free with paid upgrades? The business model shapes incentives.
Will I return to it tomorrow? The best clinical tool is the one that earns repeated daily use — because it fits the workflow, provides reliable value, and is fast enough for the micro-moments of clinical practice.
Where iatroX Fits
Clinical AI tools should be judged by clinical job-to-be-done. A clinician writing notes may need a scribe. A clinician checking the latest literature may need an evidence engine. A GP applying UK guidance may need a guideline-first answer. A trainee preparing for the AKT, MSRA, PLAB, MRCP, or UKMLA may need a Q-bank. A clinician on a ward round may need a calculator. A clinician reflecting after a case may need CPD capture. The question is not simply "which AI is best?" but "which tool fits the clinical task, healthcare system, source base, and level of verification required?"
iatroX is built around the UK clinical knowledge workflow: guideline-grounded answers via Ask iatroX, clinical brainstorming, 80+ calculators with editorial content and guideline references, 15+ adaptive exam Q-banks, and CPD-style learning. Core clinical information-retrieval and brainstorming workflows are accessible. Exam-preparation products may include paid components depending on the exam and region.
Not a scribe. Not a general chatbot. Not a literature search engine. A UK guideline-grounded clinical knowledge layer — questions, calculators, exam preparation, brainstorming, and learning in one platform.