Use traditional GPnotebook search when you know roughly which page you need: a named condition, a familiar table, a full topic review. Use AI Answers when your question spans several topics, or when you do not know which of 35,000 articles holds the answer. The two modes are complements, not rivals, and the safest workflow uses both deliberately.
What traditional search does
Classic GPnotebook search is deterministic page retrieval. You enter a condition or phrase, the platform returns the relevant concise clinical pages, and you navigate the archive yourself. You do the interpretation, and you see the authored material exactly as the editorial team wrote it. For two decades this model has fitted consultation tempo precisely because the articles are short, structured, and heavily interlinked.
What AI Answers does
AI Answers accepts a complete natural-language question and synthesises a single structured response from the relevant pages, drawing primarily on GPnotebook's GP-authored archive with supplementary use of authoritative sources such as NICE where needed. It removes the navigation step and part of the interpretation step. What it adds is a generated layer between you and the authored content.
When the classic search is the better tool
Deterministic retrieval wins when you are looking up a named condition and want the whole topic, when you need a familiar table or list you have used before, when you want to review a complete GPnotebook article rather than an extract, and generally whenever you would rather see the authored page than a machine's summary of it. If you already know the destination, search is faster and adds no interpretation layer at all.
When AI Answers earns its place
AI Answers wins when the question spans several topics, when you are asking about referral thresholds or management steps and do not know which page holds them, when you have a drug-switch query that touches more than one article, and whenever you genuinely do not know where in the archive the answer lives. This is the classic synthesis problem, and it is exactly what a conversational layer over a large corpus is for.
The safety trade-off
Be explicit with yourself about what changes between the two modes. Traditional pages require more clinician synthesis but expose the original authored material with nothing added. AI answers reduce navigation and cognitive load but introduce a generated interpretation layer that may occasionally combine pages in ways their authors did not intend. Neither mode is unsafe; they simply place the verification burden in different places.
A safe workflow
A defensible pattern for consultation use has four steps. Read the AI answer in full rather than the first line. Open the linked GPnotebook articles for anything that will change management. Verify high-risk recommendations, referral criteria, dosing in impaired renal or hepatic function, prescribing in pregnancy, against the direct guideline or medicines source rather than a summary of it. Then apply patient-specific clinical judgement, which no retrieval mode supplies.
When neither mode is enough
Some questions need synthesis across multiple UK national sources rather than within one publisher's archive: a management plan that spans a NICE guideline, a CKS topic, and an SmPC, for example. That is the job Ask iatroX was built for: it retrieves directly from NICE, CKS, SmPC/eMC, MHRA, SIGN and NHS content and returns a cited answer anchored to those sources, with the original guidance one click away. Used alongside GPnotebook, it covers the questions that an internal archive, however good, was never designed to answer alone.
Ten questions, ten routes
| Question | Best first route |
|---|---|
| "Remind me of the full topic on polymyalgia rheumatica" | Traditional search |
| "What are the referral criteria for MGUS?" | AI Answers |
| "Show me the hyponatraemia investigation page" | Traditional search |
| "How do I switch a patient from citalopram to sertraline?" | AI Answers |
| "What is the DVLA rule after a first seizure?" | Traditional search, then verify at source |
| "Child with a limp and a fever: what must I not miss?" | AI Answers |
| "The familiar table of causes of raised ALP" | Traditional search |
| "Can I use nitrofurantoin at eGFR 40?" | AI Answers, then verify against the medicines source |
| "Two-week-wait criteria for suspected laryngeal cancer" | Direct national guidance; use either mode only as a signpost |
| "A NICE-vs-summary discrepancy you suspect" | Direct national guidance |
The pattern is consistent: known destination, use search; unknown destination or cross-topic synthesis, use AI Answers; high-stakes threshold, go to the canonical source regardless of which tool signposted it.