"Is it safe?" is the wrong question about medical AI. The right question is: "What is it intended for, what are its limits, and how should clinicians use it responsibly?" DR.INFO is useful for illustrating this distinction — because it makes its intended-purpose boundaries unusually explicit.
Why "Safe" Is the Wrong Yes/No Question
Every clinical AI tool exists on a spectrum between "completely unreliable" and "infallible oracle." No tool occupies either extreme. The useful questions are: What does the tool claim to do? What does it claim not to do? What evidence supports its performance? How should clinicians interact with its outputs?
A tool can be useful, cited, and regulatory-compliant — and still produce errors. The question is whether the tool's design, labelling, and clinician workflow create appropriate safeguards against those errors.
What DR.INFO Says About Its Intended Use
DR.INFO's product label provides an unusually clear intended-purpose statement. It describes DR.INFO as an informational and educational software device for qualified healthcare professionals. The label explicitly states three boundaries.
It is not to be used to inform diagnosis, treatment, or medical decision-making. It is not intended for emergency or time-critical decisions. Outputs must be independently reviewed and verified by the responsible healthcare professional.
The label also warns that AI may introduce summarisation errors — an honest acknowledgment that even cited, source-grounded AI can misrepresent the underlying evidence. DR.INFO's terms add that the tool is intended for qualified medical professionals and is not for laypersons, caregivers, or patients.
This level of labelling transparency is above average for the clinical AI category. Many tools make broad claims about clinical utility without specifying intended-purpose boundaries this precisely.
Why Independent Verification Still Matters
DR.INFO's label says outputs must be independently reviewed and verified. This is not a weakness — it is the responsible position for any clinical AI tool in 2026.
Even the best-performing clinical AI systems produce errors. The Nature Medicine study on ChatGPT Health found 52% emergency undertriage. OpenEvidence, despite training exclusively on peer-reviewed literature, can surface outdated or context-inappropriate evidence. DR.INFO's own pilot study measured perceived usefulness, not objective diagnostic accuracy.
The verification step is not optional. Click the citation. Read the source passage. Confirm the AI's summary matches the source. Apply clinical context. Make your decision based on the verified information, not on the AI summary alone. This workflow takes seconds when citations are specific — and is the minimum standard for safe clinical AI use.
How to Use Cited Medical AI Responsibly
Four principles apply to DR.INFO and to every clinical AI search tool.
Check the citation. Does the source actually say what the AI claims? Inline citations with page markers (as DR.INFO provides) make this verification faster than tools with vague or absent citations.
Assess the source. Is the cited source authoritative for your clinical context? An FDA drug label is authoritative for US prescribing but may not reflect UK-specific licensing. A NICE guideline is authoritative for UK practice but may not be cited by a European tool.
Apply clinical context. The AI answer is a starting point, not a conclusion. Patient-specific factors — comorbidities, medications, patient preferences, local protocols — are not captured by the AI query.
Retain clinical responsibility. The decision is yours. The AI assisted with information retrieval. Documentation should reflect that the clinical decision was made by the clinician using verified authoritative sources.
What UK Clinicians Should Look For
For UK clinicians, the additional requirement is UK relevance. A tool can be well-cited and CE-marked and still not be optimised for UK practice — citing international guidelines rather than NICE, referencing US drug names rather than BNF entries, or missing UK-specific screening programmes.
The safest use of clinician AI is as a checkable clinical starting point: ask the question, inspect the source, apply context, and retain clinical responsibility.
iatroX is built around this workflow — cited UK clinical answers, calculators, exam preparation. UKCA-marked, MHRA-registered. Free.