AI disclosure
Bespoke Intermedia LLC. Effective September 1, 2026. Updated October 18, 2026.
How the practice questions are made
The practice questions and their explanations are drafted with AI assistance (Anthropic's Claude) from the official exam outline and primary sources. They are not copied from real exams. The assessment framework is designed by Julian Calvo. The steps after drafting are:
- Our process checks each question against the sources it cites. No person reviews every question.
- A blind, independent AI review answers each question without seeing the key. A question where the review disagrees with the key is rechecked and fixed or dropped.
- No licensed professional (a nurse, a pharmacist, an insurance adjuster, an attorney and so on) has reviewed a question bank unless that bank's page names the reviewer, the credential and the date. Where it says nothing, there was no such review.
Each question names the source it was written from, and you can read it after you answer. If a question is wrong, use the flag on it and say what is wrong; flagged questions are read by a person.
Where else AI is used
Older sections of the site (career guides, glossary, news and similar pages) were also produced with AI assistance. How each kind of page is made is on the editorial policy page. Some signed-in features call a language model to explain a scenario or answer a question you type; the text those features return is model output, and the screen says so where it appears. Model output can be wrong. Check it against the source before you rely on it.
DecipherU Match
Match turns your answers into a list of licenses with arithmetic over public occupation data. A model does not decide the list. Match can use a model for one optional step: putting the facts behind a result card into two to four sentences. That step is off by default. When it is on, the facts behind the card (the license, the fit band, the reasons shown on the card and any flags) go to the model provider, never your name or email, and any sentence that states something the facts do not support is dropped before you see it. If too little survives, a fixed template is used instead.
Ask why
When the analysis for an exam is paid for and this feature is switched on, a button under an explanation lets you ask a follow-up about that one question, such as why an option is wrong or for the explanation in simpler words. The model is told to answer only from that question's own explanation, the sources stored with it and the exam's domain summary, and a separate check rejects any answer with a number, law or rule that the material does not contain. When the material does not cover your question, you get the original explanation back instead of an answer.
What is sent to the model provider (Anthropic, through our AI gateway, with no fallback to another provider): the text you typed, up to three of your earlier questions and our answers from the same visit, and that item's explanation, sources and domain summary. Your name, email and account id are not part of the request body. We do not store what you typed or what comes back; we keep only a count of questions asked and refused, with a reason code, and the count events are recorded only if you accepted analytics. Answers are model output, can be wrong, and carry a link to report a problem.
How AI text is labeled
Text from a model that you request, such as an explanation of a scenario, is marked as model output where it appears. The practice questions are described here once instead of on every screen.
What the model sees
When a signed-in feature calls the model, the text you typed and the question in front of you are sent through our AI gateway to the model provider, today Anthropic's Claude. Anthropic does not train on that data under its commercial terms. If we add another provider we list it with the other processors on the what we keep page before using it.
What no model decides
Your readiness reading is arithmetic over your answers, not a model's opinion. No model decides what you see, what you are told about your chances, or anything about you that leaves the site.