It is a reasonable question, and one we get asked often: if a general AI assistant can already hold a spoken conversation, why pay for a PLAB 2 platform? For some of your preparation, you should not — a general assistant is cheaper, more flexible, and very good at explaining guidelines, drilling differentials and talking through management at midnight.
The difference shows up the moment you want to be examined rather than helped. Role-play is the easy half; the marking scheme, the hidden examiner notes, the eight-minute clock and the record of how you scored last week are the half that actually moves your grade.
| Feature | ZWIP (online AI practice) | ChatGPT voice mode |
|---|---|---|
| Spoken role-play | Real-time voice you can interrupt, with a patient who reacts to how you ask | Genuinely good — this is what it is built for |
| Marking scheme | A full examiner rubric applied to every consultation | None — you get an opinion generated fresh each time |
| Hidden examiner notes | Marking points withheld from you and handed to the marker afterwards | Impossible: anything the marker knows, you typed in |
| Scores by domain | Data Gathering, Clinical Management and Interpersonal marked separately | Unstructured prose, different every time you ask |
| Exam timing | Reading time then eight minutes, held strictly | The conversation ends when you decide it does |
| Station library | A large authored bank across all 9 station types — you meet cases you did not invent | You write the case, so you already know the twist |
| Investigations | Released only when you actually ask for them | Whatever you decide to tell it |
| Progress over time | Domain trends, pass rate and coaching across every consultation | No score history to trend |
| Marking calibrated | Tested across all 36 stations before being trusted; re-tested when it changes | No calibration, and no way to check it |
| Tone | Marks you the way an examiner would — a grade that can go down | Trained to be agreeable, which flatters borderline performances |
| Knowledge and revision | Not what we do — use a general assistant for this | Excellent: explanations, differentials, guideline summaries |
| Cost | A low monthly membership | Cheaper, and free at lower usage tiers |
The verdict
A general AI assistant is a good study aid and a poor examiner. It will rehearse a conversation with you all day, but it cannot hold a standard you have not shown it, cannot hide the marking scheme it is marking you against, and cannot tell you that your history-taking has been slipping for a fortnight.
We take the marking seriously enough to measure it. Before our examiner rubric marks anyone’s work it is run repeatedly over a full set of real consultations and checked for consistency: identical runs agree to within a tenth of a mark, and across 72 repeated pairs no pass-or-fail decision changed. That is the difference between being marked and being agreed with — and you can read how the whole thing is built on our technology page.
Keep comparing
Try it for yourself
Register free and practise a full station with an AI patient — 15 voice minutes included, no card required.
Start free