How our AI stays honest
Every AI-generated card, quiz item, and tutor answer is grounded in your own material — with visible source citations you can tap to read the exact passage, an honest confidence signal, and the willingness to say 'that isn't in your material' instead of making something up. We grade what you mean, not the exact words you typed.
AI that shows its work
Most study apps hand you AI answers with no way to check them — and the AI-native ones are hated for being confidently wrong on the exact STEM problems you need right. We took the opposite path. When you generate a study kit from your own notes, PDF, or reading, every card carries a citation back to the exact passage it came from. Tap it and read the source. If the AI can't ground an answer in your material, it tells you — and asks whether you want a general-knowledge answer instead of quietly inventing one.
What honesty looks like, in the product
Not a marketing promise — the trust signals are on the cards themselves.
Tappable source citations
Every grounded card and answer shows the passages it came from. Tap a citation to read the exact source text — no more taking the AI's word for it.
An honest confidence signal
Each answer is labeled Grounded (traced to your material), Inferred (reasoned from it), or Not in your material — so you always know how solid it is.
It refuses to guess
Ask the tutor something outside your material and it says so, then offers a general-knowledge answer as your explicit choice — never a fabricated one dressed up as fact.
Verify against source
Edited your notes? One tap re-checks a card against its cited passage and flags drift — with a corrected answer the source actually supports.
Graded on meaning, not spelling
Type an answer in your own words. A faithful paraphrase, a synonym, or a reordered statement is marked correct — we grade the idea, not the exact string.
It names your misconception
When you're wrong, you get more than a red X: the specific wrong idea you appear to hold, so your next question is the right one.
How grounding works
One upload, and every study tool inherits the trust layer.
You bring the material
Upload a PDF, paste notes, or pick a reading. It's chunked into passages the AI can cite precisely.
The AI generates from it — and cites it
Cards, quizzes, and audio are built from your passages. Each one records the exact chunk it drew on and the sentence that supports it.
You can always check
Citations are tappable, confidence is labeled, and 'Verify against source' re-checks any card on demand. Nothing is a black box.
Straight answers
Does the AI ever just make things up?
When you generate from your own material, cards are built only from passages it can cite — anything it can't ground, it drops. When you ask the tutor something your material doesn't cover, it tells you plainly and asks before answering from general knowledge. Confidently-wrong answers dressed up as fact are exactly what this layer exists to prevent.
What does the confidence label actually mean?
Grounded means every claim traces to a passage we cite. Inferred means it's a reasonable synthesis of your material that isn't stated word-for-word. Not in your material means we won't pretend — you decide whether to go to general knowledge.
I typed the right idea but different words — will it be marked wrong?
No. Grading judges meaning. Paraphrases, synonyms, abbreviations, and reordered answers that convey the required idea are marked correct. Exact-string grading is a failure mode we deliberately don't have.
What if I edit my notes after making cards?
Use 'Verify against source' on a card. It re-checks the card against its cited passage and, if the source has drifted from the card, flags it and suggests a corrected answer.
Study on material you can trust
Generate a cited study kit from your own notes and see every source for yourself.