Name a book. Get your bearings in it.
What it is, what it argues, how it is built, the themes running through it, the world it came out of, who it is for and what to read next. Useful before you start, after you finish, or while you are deciding whether to bother.
The facts a model most reliably gets wrong — the author, the year, the original language — are checked against a source this page retrieves for you, and each one is marked confirmed or unverified. A book it cannot identify comes back as not found, not invented.
All three ship with a saved brief, so you can read a whole answer — the confirmed and unverified marks, the retrieved abstracts, the checks this page ran — without signing in or spending a credit.
What this app will and will not do
What it is for
- Summary and synopsis. What the book is, what happens in it, what it argues. Asking for a summary so you do not have to read it is a normal request and this app answers it properly.
- Analysis. Themes, structure, the shape of the argument, how the prose works, what the book is trying to do and whether it manages it.
- Context. The world it came out of, the life it came out of, how it reached print, how it was received then and how that has changed.
- Guidance. Whether to read it, where to start, what it asks of you, and what to read next and why.
What it will not do, whatever you type
- Reproduce the text. No passages, no chapters, no “write out the opening”, no lyrics. Those are the book, and handing them over is not summarising.
- Retell it densely enough to replace it. A structural breakdown says what each stretch of the book is for. This page measures how thick that breakdown gets and says so when it starts to read as a retelling.
- Quote past a budget. At most three extracts, 25 words each, 50 words altogether, each attributed and each carrying a point. Quoted runs hidden inside the commentary are counted too.
- Quote poetry or lyrics at all. A poem is short enough that any excerpt is a substantial part of the whole work. Those get analysis and no reproduction.
A request for the book's own text is refused before the run starts, so you are never charged for a refusal. The in-scope version of what you asked is offered in one click.
Why the facts carry marks
Books are the classic place a language model confabulates: plausible titles, wrong authors, editions that never existed, all delivered with total confidence. So the five facts most likely to be wrong and most likely to be repeated — author, year, original language, form, series — are checked rather than trusted. The page retrieves a source, hands it to the model, and then re-reads the source to see whether it actually contains what the brief claims. A fact confirmed that way is marked confirmed. A fact the model recalled on its own is marked unverified, which is honest and often fine. And a fact cited to a source whose text does not support it is flagged loudly — because a citation pointing at a real page is not evidence for a claim the page does not make, and that failure is worse than no citation at all, not better.