Best AI Tools for UKMLA Revision
What the UKMLA actually is
The Medical Licensing Assessment tests the core knowledge, skills and behaviours of doctors who want to practise in the UK. It has two components, the applied knowledge test and the clinical and professional skills assessment, with the CPSA being a performance-based assessment of clinical and professional skills, knowledge and behaviours.
Scope and format sit on the GMC's MLA pages, which is the canonical source. It sets out both components and the routes for UK graduates and international doctors.
UK medical schools set and deliver the CPSA for students in their penultimate or final year, and the GMC quality assures those assessments; the practical consequence is that your own medical school sets it.
For revision, you’re dealing with a knowledge test and a performance assessment that tests behaviours and clinical skills.
UK graduates pass the MLA as part of their degree. International doctors take PLAB, which is MLA-compliant.
Most revision products serve only the knowledge half.
What AI is genuinely good at for the AKT
Zoom out to the wider medical school view with the general AI tool roundup; this page narrows to UKMLA revision and the best AI tools for UKMLA. For the method and sequencing, read how to actually use AI for finals; this page is the shopping list.
Explanation on demand lets you interrogate reasoning late at night. Ask why each distractor is wrong, push for counterexamples, and make it compare lookalike topics until the difference actually lands.
Question generation turns a weak topic into practice items fast. Ask for five short stems, then five trickier variants, and steer it toward the confusions you keep missing. You get reps without hunting for them.
Condensing helps when a chapter is too long for a tired Friday. Paste the section, ask for half a page of bullet points, then ask for it again as prompts.
All three are about knowledge, and nobody is checking any of it for accuracy.
Where AI gets the UKMLA wrong
AI gets UKMLA-style questions wrong often enough to trip you, and the kicker is inconsistency. Same wording can produce different answers on different tries, which matters when you want a single defensible choice, not vibes.
Lai and colleagues put GPT-4 through 191 UKMLA questions, running each single best answer three times. It was consistently correct on 67.5% of items, consistently incorrect on 12.6%, and across attempts it averaged 76.3%.
The worrying bit was volatility, with different answers to the same question in 19.9% of cases. You cannot bank learning on sand.
Borchert and colleagues sat the situational judgement test with ChatGPT and reported 76% overall, 929 of 1217 available marks. It answered entirely correctly 0% of rating questions, 23% of multiple-choice, and 6% of ranking items.
That headline 76% hides how brittle the behaviour gets once you move from recall to prioritisation under rules.
Both studies used 2023 era models and today’s are better, but that does not rescue the point. You cannot tell which 12.6% you are looking at, and the model will not tell you when it is bluffing.
If you revise with a chatbot, keep a sourced reference open beside it and check every claim that affects a decision.
The best AI tools for the UKMLA, with prices
Every price below for the best AI tools for UKMLA revision was checked on 13 August 2026. Prices move, so check at source before you pay.
- iatroX What it is: AI question bank with a Socratic tutor and study planner, UK-focused · Price, checked 13 Aug 2026: £29/month or £99/year, see iatroX publishes its prices · Free tier: Yes, but the UKMLA bank is not in it
- PassMedicine What it is: Finals and UKMLA question bank plus a syllabus textbook, not AI · Price, checked 13 Aug 2026: £15 for 4 months, £20 for 6, £25 for 9, £30 for 12, see PassMedicine's finals package · Free tier: Free demo
- Quesmed What it is: Question bank covering UKMLA, MRCP, MSRA and UCAT, with a knowledge library · Price, checked 13 Aug 2026: No price published on the site, see Quesmed's pricing page · Free tier: Not stated
- AMBOSS What it is: Medical library and question bank with an AI layer · Price, checked 13 Aug 2026: No UK price published, see AMBOSS · Free tier: Free trial
- NICE Clinical Knowledge Summaries What it is: Sourced UK clinical reference, not AI, and the thing to check AI against · Price, checked 13 Aug 2026: Free · Free tier: Free, UK access only
- Geeky Medics What it is: Free OSCE guides, MCQs and flashcards, plus AI virtual patients on its platform. See the free Geeky Medics library · Price, checked 13 Aug 2026: Free tier, paid platform bundles · Free tier: Yes, extensive
- General chatbots What it is: Explanation on demand, question generation, condensing, can be confidently wrong · Price, checked 13 Aug 2026: Free tiers, paid tiers vary · Free tier: Yes
- Diagnosica What it is: An AI patient you talk to, then a marked scorecard · Price, checked 13 Aug 2026: $29/month or $249/yr, UK cards £22/month or £189/yr · Free tier: One case a week
Two of the eight won't show price. Quesmed publishes no price, we checked pricing, FAQ and home page on 13 August 2026 and found none. AMBOSS publishes no pound price, its site resolves to the US edition and the UK path returns a 404. Not publishing a price is a product choice, and not a neutral one.
The free tiers aren't equivalent. iatroX's is real, covering askiatroX plus MRCP Part 1, MRCEM SBA, PSA and PARA with no card, but the UKMLA bank sits behind the paid plan.
Diagnosica is our product, so read that row with that in mind. Its rubrics are calibrated to the published mark sheets for PACES, MRCEM, RCGP SCA, USMLE Step 3, PLAB-2, MCCQE, RACGP AKT and NEET-PG. The UKMLA CPSA is not among them. What it gives a UKMLA candidate is history-taking and reasoning practice against an AI patient, not a CPSA-calibrated score.
Before one more stem, talk to an AI patient in Diagnosica, voice or text, then see how your history and reasoning score.Start a case free.Free tier is one case a week, and it's live and early.
The half of the UKMLA nobody's tool covers
The CPSA is performance based and set by your school, which means the core skill being tested is doing the station itself, not spotting the option. Most UKMLA tools are question banks. They won't make you talk, and they won't mark whether you asked.
Practise the conversation, not advice for real patients. Reading about a presentation is not the same as taking a history from someone who has it, with pauses, queries and answers that make you think on your feet. If nobody will role play with you, try how to rehearse stations alone.
A case file from our own library makes the gap obvious. Martin Rowe, 55, an AI patient, not a real one, opens with, "I've suddenly gone short of breath with chest pain over the last two days."
He volunteers the chest pain, the breathlessness, and that he is worried he is having a heart attack. Nothing else, unless you ask for it.
Two facts sit behind a probe, and are flagged as such in the file. He has coughed small amounts of blood on two occasions, and his left calf has been swollen, tight and tender for about a week.
On direct questioning he adds a long-haul flight home from Australia about five days ago, and keyhole surgery on his left knee about three weeks ago.
Those facts sit behind a question. They do not arrive unless you ask. A multiple choice stem hands you the swollen calf in the second line. A patient does not. That gap is the CPSA. International graduates will often be thinking about the PLAB 2 equivalent, since they sit PLAB rather than the CPSA.
What none of it replaces
Physical examination lives in real rooms with real people. No voice or text product trains it, ours included, and Diagnosica marks itself as not doing it; you learn it with patients, supervisors and placements, and you’re assessed the same way.
Your school's station format is local, and schools set and deliver the CPSA themselves, so what you sit is a question for your faculty. Past papers, your mark sheet and your tutors beat anything online for this, ours included.
Actual patients remain the point, and reps against a simulated patient are practice, not experience, and they don’t carry the noise, agendas or consequence. You still need clinics and wards to grow judgement and to prepare in context, as in preparing for finals.
"It doesn't give advice about real patients, and it's not a diagnostic system. Treat every output as educational." Treat simulations as drills, never as care, and never as a substitute for speaking to patients yourself.
Where to start
- Pick one paid question bank for the knowledge half. Choose on a price you can actually see before you pay, and on UK focus. Stop hopping between providers and grind questions daily.
- Keep a free, sourced reference open beside any chatbot, because it'll be confidently wrong sometimes. Use NICE Clinical Knowledge Summaries for concise phrasing and contraindications. Cross-check anything the bot says before it makes it into your notes.
- Spend time talking out loud to something that answers back, the half the question banks don't touch. A study partner is best, but if you're on your own, Diagnosica gives you an AI patient to speak to, and it does no physical examination practice.
Diagnosica is live and early, and it doesn't give advice about real patients. The free tier is one case a week. It's not a diagnostic system. Run a case free or try the no-signup demo on the landing page, about 3 minutes, marked when the case ends.
Educational use only, not medical advice. AI-generated; verify clinically against primary sources.
Clinical review pending.
