Two credible UK clinical AI tools can give you a structured answer to the same question and yet be doing fundamentally different things underneath. GPnotebook AI Answers synthesises a curated internal archive. iatroX retrieves from national guidance and medicines references directly. Neither route is simply better; they make different trade-offs, and the safest clinicians understand which one they are looking at.
The curated-archive approach
In this model, the AI synthesises internally written and edited clinical summaries. GPnotebook AI Answers is the clearest UK example: answers are generated primarily from GPnotebook's own clinician-authored archive of more than 35,000 articles, with supplementary use of authoritative sources such as NICE where needed. The corpus is the product of decades of editorial work, and the answer's provenance points back into that corpus.
The direct-guideline approach
In this model, the AI retrieves from national guidance, medicines references and other authoritative sources as close to the underlying recommendation as possible. Ask iatroX works this way: retrieval runs over NICE, CKS, SmPC/eMC, MHRA, SIGN and NHS content, and the citations under an answer point at those sources themselves rather than at an intermediate summary of them.
What the curated archive buys you
The advantages are real and daily. Curated summaries are concise, consistently formatted, and written in primary care language, which makes them fast to read under consultation pressure. Navigation is easier because the editorial team has already decided what matters for a GP. And the bounded corpus keeps the retrieval surface clean.
What it costs you
The archive model adds an editorial layer between the clinician and the original guidance: you are reading an interpretation of an interpretation. Updates depend on the archive's editorial cadence, so there is always a window in which a summary can lag its source. And compression is not free; some evidential nuance, exceptions, populations, strength of recommendation, is inevitably smoothed out of a concise summary.
What direct retrieval buys you
Proximity to canonical wording is the core advantage. When the answer cites the NICE recommendation or the SmPC directly, verifying an exact referral or treatment criterion means one click, not two. Jurisdiction and publication date are visible on the source itself. And when guidance changes, a system retrieving from the source reflects it on the source's timetable rather than an editor's.
What it costs you
Source documents are long, fragmented, and not written for speed; retrieval quality has to be excellent or the answer drowns in the guideline's own structure. Conflicting guidance between sources needs explicit synthesis rather than an editor's quiet resolution. The engineering burden is higher, which is why fewer products attempt it.
Complementary, not competing
The practical conclusion is not to pick a winner but to assign roles. GPnotebook AI Answers is a strong choice for fast curated explanation of a topic in familiar primary care language. iatroX is the tool for direct UK-source-grounded synthesis when you want the answer anchored to the guidance itself. And for genuinely high-stakes verification, nothing substitutes for NICE or the original medicines reference read in full.
That suggests a three-layer workflow: start with an AI summary for orientation; consult the curated reference for structured explanation; confirm against the canonical source before any high-risk decision.
How to tell which model you are looking at
You do not need vendor documentation to classify a tool; the answer itself tells you. Look at where the citations land: on a publisher's own pages, or on NICE, CKS or an SmPC. Look at whose date is displayed: the summary's last-reviewed date and the guideline's publication date are different facts, and only one of them tells you how current the recommendation is. Look for a source note explaining what the answer relied on, which GPnotebook provides and which any serious tool should. And notice what happens when you ask something the corpus does not cover: a well-behaved archive tool reaches explicitly for external guidance or declines, while a poorly behaved one improvises. Two minutes of this triage on any new tool tells you how much verification work each answer will leave you, which is the number that actually matters in clinic.
Five worked examples
| Scenario | Where each model helps |
|---|---|
| Suspected cancer referral | Summaries orient you fast, but the exact two-week-wait criteria should be read in the national guidance itself |
| Renal dosing decision | A curated summary flags the issue; the dose adjustment belongs to the SmPC/eMC entry |
| Menopause treatment options | Curated explanation is excellent for counselling structure; direct retrieval anchors specific eligibility and risk wording |
| Anticoagulation choice | Cross-source synthesis (guideline plus medicines reference) favours direct retrieval; the summary is a useful primer |
| Paediatric fever | A concise summary covers the pathway; red-flag thresholds warrant the original guidance wording |
The common thread: the higher the stakes and the more precise the threshold, the closer to the canonical source your final check should sit.