Quality, verified
Anyone can bolt an AI onto a question bank. We do the part almost nobody does: the questions our AI writes are independently reviewed by a separate examiner-grade AI, fact-checked against real FAA source material — and we publish the results.
Oral soundness has gone from 85.8% to 96.3% across 13 published audits. We publish the history so you can watch it move rather than take our word for it.
A question fails the review above if anything in it was wrong — including an explanation that argues correctly for the right answer and then names the wrong letter, which is the most common thing we find. Those are confusing to read; they cannot teach you a false fact. These are the ones that could:
The oral is our weaker lane, and we would rather you heard that from us: a spoken question carries a full model answer, which is longer than a written explanation and has more room to be wrong.
Written: 11 published audits, 1292 questions, most recent 2026-10-26 — the marked answer independently verified correct in 99.8% of them. Oral: 13 published audits, 1668 questions, most recent 2026-10-26. Covers all three ratings (General · Airframe · Powerplant).
One pass rate hides the difference between a mislabeled option and a wrong answer. Here is every defect found across all 11 published audits — 1292 questions — by kind, most serious first.
Counts in red could have changed which answer you should pick. The rest left the answer key correct and got the supporting text wrong.
A pipeline runs continuously in the background — no student ever waits behind a quality gate.
Our AI writes a practice question and its explanation, grounded in the FAA handbooks and regulations for that ACS code.
A separate examiner-grade AI — not the one that wrote it — fact-checks the answer, the wrong-answer choices, and the explanation against FAA source material.
Anything questionable is flagged. Recurring problems become permanent written constraints on how new questions get generated — so the same mistake cannot come back. We track the pass rate and publish it, openly, right here.
Before a question reaches you it is written from FAA source for its ACS code, and thrown out and rewritten if it comes out malformed or repeats one you were recently asked.
The fact-check is separate, and it comes afterward: a different AI re-reads a sample against FAA source, which is where the numbers above come from. A question that clears it joins a reviewed pool we draw from before writing a new one — and a question that fails is pulled from that pool and not let back in.
Most of what you see is still written fresh for the code you are weakest on — the reviewed pool is what builds up alongside it.
Why not check every question first? A second read would mean waiting through another AI pass before each question appeared, and a gate made fast enough not to annoy you is not a gate worth trusting. We would rather check a sample properly, publish what we find, and email you the correction when something you were already shown turns out to be wrong.
The explanation described schematic diagrams correctly — and then dismissed option A, the marked answer, as though it were the installation-diagram distractor. It meant option B.
Option A is correct and the reasoning behind it was correct; only the letter was wrong. The installation-diagram distractor is B. The answer key never changed — but a student re-reading that explanation would find it arguing against the answer it had just marked right.
Logged in our audit records against ACS knowledge code AM.I.B.K4.
This is the most common defect we find, and the mildest. It cannot cost you a question, because the answer key is untouched. We publish it and we email it anyway: a student whose explanation contradicts itself has no way of knowing which half to trust, and “probably the answer key” is not good enough the week before a written.
When an audit finds a problem in a question you were served, you get an email naming the error and the correction — whether or not it changed your answer, and whether or not you got it right. An instructor reads and approves every correction before it is sent.
You shouldn’t have to wonder which of the things you studied were wrong.
On exam day, "close enough" isn't. We don't think your prep should be either — so we check our own work, in public.
Get started →AMT Schools, Inc. is not affiliated with or endorsed by the Federal Aviation Administration, and is not an FAA-certificated Part 147 school. Our practice questions are written and reviewed by us — they are not FAA test questions, and no exam outcome is guaranteed.
Quality audit methodology and results are published in good faith. Individual questions, prompts, and model details are proprietary.