01Who we use
SQEase uses AI models from several providers, each picked for its task. We change models as better ones are released, so we name the providers rather than the models.
| Provider | Used for |
|---|---|
| OpenAI | Reading the documents you upload, writing questions and flashcards, answering each question again to check its answer, and Lex's explanations. |
| Processing the documents you upload so they can be matched to the SQE1 syllabus, an independent second answer to each question, and checks of legal claims. | |
| Anthropic | Reading photos of handwritten notes, arithmetic checks, and a backup for Lex if another provider is unavailable. |
| Deepgram | Turning your voice into text when you dictate to Lex. We switch off its model-improvement programme, so it does not train on your audio. |
| Tavily | Live web search of UK sources, used when our textbook library cannot settle a point. |
02How each question is checked
Every question written from your notes, and every question added to our shared bank since July 2026, goes through these automated checks before it can appear in your practice:
- Grounding. The question is written from your document, with passages from SQE textbooks for the same subtopic.
- Review. A separate reviewer model works through the question against those passages, reaches its own answer, and looks for a second answer that is as good as the key, a question that is unclear, or wrong options that are too easy to rule out.
- Rule-based checks. These catch duplicate options, an explanation that names a different answer from the key, a "how much" question without figures, and named characters.
- Figures. Every figure in an explanation must come from the question, and a separate model recomputes the arithmetic with a calculator.
- Legal claims. Rates, time limits, thresholds and statutes are checked against our textbook library, then against earlier checked claims, and, where those cannot settle a point, against live UK web sources.
- The answer. Two models from different companies answer the question from scratch, without seeing the intended answer, and both must agree with it. If they disagree, two more models look again. If the answer is still in doubt, a final check against live web sources decides, or the question is held back.
A question that fails a check is fixed with small edits that never change its answer letter, then checked again, or it is held back. Software never changes a question's answer. When you upload a document, the library shows how many of its questions were held back.
03The shared question bank
Most of the questions in our shared bank were written before the checks above existed. They have been audited for specific known problems, including figures that do not match the question, explanations that name the wrong option, incorrect citations and named characters, and questions with a problem are fixed or withdrawn as we find them. They have not all been through every check in section 2.
04Known limits
- Mistakes are possible. The checks reduce errors; they do not remove them. A wrong passage in a source can mislead both the writer and the checks that use it.
- Not every check runs on every question. Legal claims are checked only where a question states a rate, time limit, threshold or statute. If a provider is unavailable, some checks are skipped rather than holding up your questions.
- The examinable law has a cut-off. The SRA tests the law in force at a cut-off date set for each sitting. Check the date for your sitting on the SRA's dates page, and treat any question that conflicts with it with caution.
- Your documents shape your questions. Questions written from your notes follow your notes. If your notes are wrong or out of date, a question can be too.
- No legal advice. Nothing on SQEase constitutes legal advice.
05No guarantee of results
06Your documents & AI training
- The text of a document you upload is used to write your questions, and it is deleted when they are finished. A document you leave without generating questions keeps its text for up to 30 days.
- Your documents are never added to our knowledge base, and we do not train or fine-tune AI models on them.
- Our AI providers operate under contractual undertakings that your input is not used to train their models.
07Liability
- SQEase accepts no liability for errors in AI-generated content.
- Nothing excludes liability that cannot lawfully be excluded.
See our Terms & Conditions §10.
08Reporting an error
Use the Report an issue button on any question. A report on a question from our shared bank re-runs the checks straight away. If a problem is confirmed, the question is withdrawn first, and anything that affects the answer goes to a person. A report on a question written from your own document goes to our team.