Evaluate an AKT AI tutor by what it does with your attempted answer, not by the model name on its landing page. A useful tutor should identify the decision you misunderstood, explain it with inspectable support and help you attempt a different question independently. The presence of conversational AI is not, by itself, evidence of better preparation.
This article is published by iatroX and includes iatroX among the tools subject to the same criteria. Public product descriptions were reviewed on 24 September 2026. The exercises below are original inspection tasks, not reported results from hands-on testing of paid accounts.
What the public descriptions establish
AKT Prep, as described on 24 September 2026, advertises an AKT bank, a NICE- and UK-guidance-focused tutor, follow-up explanations and weakness-focused study. Its public demonstration presents answer commitment before teaching and discusses why an incorrect option might have been tempting. These are product-design claims, not proof that every response identifies the learner's actual misconception.
PassMedicine's AKT page, checked on the same date, describes syllabus-focused questions, teaching notes, textbooks, revision and timed modes, mocks and review tools. The reviewed page did not establish an equivalent question-linked conversational tutor. That is a limit of the public evidence, not a claim that every PassMedicine product or future release lacks AI.
Per iatroX product information, September 2026, its Socratic Tutor opens on an attempted question and asks targeted follow-ups. The comparison should therefore distinguish the advertised workflow from what an individual learner can demonstrate during actual use.
Find out what the tutor knows
A tutor may be connected to the question stem, answer options, your selected answer, your reasoning, the provider's explanation and supporting sources. Those are different inputs. A fluent answer does not reveal which inputs were actually available.
Start a trial by asking the tutor to identify the lead-in and summarise your chosen reasoning without answering the question again. Check whether its summary is faithful. A tutor that invents a reason you never gave may still produce useful general teaching, but it has not yet diagnosed your misconception.
Also distinguish access to the stored question explanation from independent source retrieval. Rephrasing a provider's explanation and checking a current guideline are separate operations. Both can help, but the interface should not leave you guessing which occurred.
Probe one: a clinical decision without a copied examination item
Use this original fictional scenario: a learner selects a comprehensive later investigation when the question asks which information must be established before choosing the next step. No diagnosis or treatment is specified because the exercise concerns the structure of the reasoning rather than a clinical protocol.
Ask the tutor: "What did the lead-in require, and which missing fact would change my choice? Please ask me one question before explaining." A useful response should bring attention to the decision-changing information, not simply provide a longer list of possible investigations.
Then change the scenario's timing or setting. Ask whether the original reasoning still applies. The test is not whether the tutor agrees with the expected answer every time. It is whether it recognises which change matters and explains why, without inventing clinical details.
Record the exact behaviour observed. "Asked a relevant follow-up about the missing finding" is a better evaluation note than "seemed intelligent".
Probe two: statistical interpretation
Use an original fictional dataset of 200 participants in each of two groups, with an outcome occurring in 20 participants in one group and 30 in the other. The risks are 10% and 15%; their absolute difference is five percentage points. These figures are invented for arithmetic practice, not reported treatment evidence.
Before requesting a calculation, ask the tutor which denominator belongs to each group and which comparison the question requires. A useful tutor should distinguish an absolute difference from a relative change and preserve the outcome and follow-up context.
Next alter only the group size or event count. The learner should attempt the new calculation unaided. This exposes whether the conversation taught a transferable setup or merely supplied the first numerical answer. Independently check the arithmetic rather than allowing one AI output to verify another.
Probe three: an organisational scenario
Create a fictional practice audit in which the stated objective is to assess completion of a follow-up process, but the available report counts only appointments booked. Ask the tutor whether the measure answers the stated audit question and what information remains missing.
This tests whether it notices the difference between a process starting and being completed. It should not fill the gap with an invented NHS rule, contract requirement or legal obligation. Where an actual organisational question depends on a current policy, the response should identify the relevant jurisdiction and source rather than manufacture certainty.
The point is to sample different reasoning tasks. A tutor that handles a familiar clinical explanation well may still require careful inspection when numbers, professional rules or incomplete administrative data are involved.
A practical inspection checklist
| Dimension | Observable behaviour to inspect | Insufficient evidence |
|---|---|---|
| Context | Accurately identifies the question and your attempted reasoning | A generic disease summary |
| Teaching sequence | Allows commitment, then asks a relevant follow-up | Revealing the answer before you have tried |
| Misconception | Explains the precise difference between your option and the better-supported alternative | Saying only that the chosen answer was wrong |
| Sources | Provides a relevant, accessible source and indicates its applicability | A link that does not support the specific claim |
| Uncertainty | Identifies missing facts or conflicting guidance | Confidently inventing details to complete the story |
| Follow-up | Supports a new independent attempt or a meaningful later review | A long conversation with no return to problem-solving |
This checklist is intentionally qualitative. Turning a small informal trial into a numerical ranking would imply a measurement precision the method does not possess.
Verify the answer rather than the appearance of a citation
Open the cited source and locate the relevant recommendation or explanation. Check the population, care setting, publication or update date and any conditions that change applicability. A source can be reputable and still not support the exact sentence generated by the tutor.
When sources appear to disagree, ask whether the disagreement concerns different patients, different time periods or a genuine difference in recommendations. Do not accept a synthetic compromise merely because it sounds balanced. The tutor should make the unresolved issue visible.
The same standard applies to iatroX. Its source-grounding methodology is a design intended to support traceability, not proof that every output is correct. Learning tools remain subject to critical review.
What happens in the next session?
A useful conversation is not automatically a longitudinal learning system. Inspect whether the platform can revisit the relevant topic, distinguish remembered items from fresh practice and let you review uncertain correct answers as well as mistakes.
Do not assume that the word "adaptive" guarantees all these functions. Ask what is selected, on what evidence and whether you can preserve curriculum breadth. A system that repeatedly serves a weak topic may help locally while leaving another domain unsampled.
For a personal trial, retain a short record of the original misconception, the explanation that resolved it and a later unaided attempt. Those observations provide a more concrete purchasing basis than the number of messages exchanged.
The verdict by learner need
Inspect AKT Prep when you specifically want its advertised question-linked tutor and weakness-focused workflow. Keep PassMedicine when its explanations, bank and review tools already support effective independent study; the absence of a demonstrated conversational feature on a public page does not make those functions obsolete.
Consider iatroX when targeted Socratic follow-up and subsequent practice address a remaining gap. There is no need to add it when an existing tutor already resolves that problem. The strongest AI tutoring choice is the one that helps you return to a better unaided attempt, not the one that produces the longest explanation.
Frequently asked questions
Is an AI tutor automatically better than written explanations?
No. A clear written explanation may be sufficient, while a tutor is useful when responsive questioning helps resolve a specific misconception.
Does NICE-focused marketing guarantee a correct AKT answer?
No. Check that the cited source supports the precise claim and applies to the scenario, regardless of which platform generated the explanation.
Can this article tell me which tutor performed best in testing?
No paid-account comparative test was performed for this article. It provides dated public product descriptions and an original method readers can use to inspect the learning experience.