Using an AI tutor for the SQE: what it can and cannot do

AI is genuinely good at some parts of SQE preparation and genuinely bad at others, and the difference matters more here than in most subjects: a confidently wrong answer about a legal threshold teaches you to make a submission that fails.

This page is about which is which.

Where it actually helps

Explaining why the wrong answer is wrong. Question banks tell you the correct option. The useful question is why the one you picked was plausible, and that is a conversation rather than a lookup.

Rehearsing the oral stations. Advocacy and the client interview are marked on things you cannot practise alone — being interrupted by a judge, a client who only mentions the deciding fact when asked. The SRA's own guidance on its samples suggests a peer for three of its four exercises. Not everyone has a peer available at eleven at night.

Marking against criteria. Every SQE2 station has published criteria and published Performance Indicators. Scoring an attempt against a known grid is work a machine can do consistently, and it is work most candidates skip because it is tedious.

Showing you what you are avoiding. Honest progress tracking has to count what you have not attempted. People revise what they are already comfortable with.

Where it misleads

Law it was not given. A general chatbot answers from training data of unknown age, for an unknown jurisdiction, with no way to show its source. SQE2 examines the law of England and Wales as it stood four calendar months before your assessment date. Those are not the same thing, and the failure is silent.

Invented authority. Fabricated citations are the best-documented failure mode of language models, and law is where they do most damage. If a tool cannot show you the passage an answer came from, you cannot check it.

Assessment detail. Much published SQE guidance is wrong about the exam's own shape — the 45-minute advocacy error comes from a superseded specification and has propagated across the web. A model trained on that web inherits it.

Confidence. A model's certainty is unrelated to its correctness, and nothing in the interface tells you which answer is which.

What to check before trusting one

1. Is it grounded in a source it will show you? Retrieval from a named corpus is checkable. Generation from memory is not.
2. Which jurisdiction, and as at when? If it cannot answer, it is not an SQE tool.
3. Does it mark against the SRA's actual criteria, or offer generic feedback?
4. Does its progress measure count what you have avoided, or only what you have done?
5. Can you see the papers it is reasoning from?

Those five apply to any tool, including this one. The right posture toward an AI tutor is the one you would take toward a keen trainee: useful, fast, worth checking.

How LexMastery approaches it

The tutor answers from an indexed corpus of black-letter law for your track, retrieved per question — not from open-web recall. Oral stations run against the real clocks, with a character who behaves like the assessor's role: a judge who interrupts, a client who holds facts back until asked. Attempts are marked against the SRA's published criteria for that station.

The readiness score is deliberately unflattering: it is weighted across the papers and skills on your track, and areas you have not attempted pull it down rather than being ignored. First attempts count for more than retries, so you cannot grind a number upward.

Both tracks are covered — SQE1 and SQE2 for England and Wales, and the UBE/NextGen Bar for the United States. It runs in your browser; there is nothing to install.

It is not a substitute for a peer, a tutor or sitting the SRA's own samples. It is the thing that makes the hundredth repetition possible on a Tuesday night.