# Book Summary > Name a book and get an orientation to it: what it is, what it is about, how it is built, what > it argues, the themes running through it, its historical and critical context, who it is for > and what to read next. The facts a language model most reliably invents — author, year, > original language, form, series — are checked against a retrieved source and marked confirmed > or unverified. A book that cannot be identified is reported as not found rather than described. https://book-summary.skillsafe.ai/ ## What it does Useful before starting a book, after finishing one, or while deciding whether to bother. The brief covers: an orientation in plain language, the premise, the argument or what the work is doing, a structural breakdown saying what each stretch of the book is for, thematic analysis, historical / biographical / publication context, reception on publication and since, a description of the prose, a difficulty band, read-it-if and skip-it-if guidance, a reading list with a reason attached to each entry, and an editions note. ## The two things that make it different from asking a model directly **1. The volatile facts are checked, not trusted.** Books are the classic place a language model confabulates: plausible titles, wrong authors, editions that never existed, all stated with the same fluency as a correct answer. Five facts carry that risk and get special handling — author, year of first publication, original language, form, and series. The app runs a Wikipedia search for the title the reader typed, hands the retrieved abstracts to the model, and requires each of those five facts to be attributed either to a specific retrieved record or to the literal string `model-knowledge`. Then — and this is the part that matters — it re-reads the cited abstract and checks whether it actually contains the claimed value. Three outcomes, all displayed: - **confirmed** — the cited abstract contains the claim. - **unverified** — the model's own recall, honestly labelled. Not wrong, just unchecked. - **false citation** — the brief cited a real record for a claim that record does not make. The third is why the check exists. `ss.search` returns a MediaWiki *snippet*, not an article infobox, so a "cite only retrieved records" contract can be satisfied by citing a genuine page about the right book for a fact that page never mentions. That is a confabulation wearing a real URL, and a verification pass that only confirms a record id was cited will wave it through while manufacturing the appearance of rigour. A false citation is therefore ranked **worse** than an honest admission of recall, not better. **2. A book it cannot identify is reported as not found.** Inventing a brief for a book that does not exist is the worst output this app could produce, so the not-found branch returns no orientation, no themes and no structure — only real works with similar titles, and a note on what information would let the search succeed. The app is equally careful in the other direction: an obscure, out-of-print or non-English work that the search missed is still a real book, and gets a brief marked `recalled` rather than a refusal. ## What it will not do This app takes a copyrighted work as its input. Summary, synopsis, thematic analysis, structural breakdown, context, comparison and reading guidance are facts about a book and original commentary on it, and the app produces them generously — including when the reader's stated reason is that they do not want to read the book. That is a normal request and the app answers it properly; the reader's motive is not the test. What the app never produces is the work's text: - no extended quotation, no reproduction of a passage, page, chapter, scene, stanza or lyric; - no "write out the opening", no "give me chapter three"; - no chapter-by-chapter near-paraphrase dense enough to substitute for reading the book. A structural breakdown says what each stretch of the book is *for*; the app measures how thick that breakdown becomes and flags it when it starts to read as a retelling. Brief quotation in service of criticism is ordinary scholarly practice and is permitted within a fixed budget: at most 3 extracts, 25 words each, 50 words in total, each attributed and each carrying a stated critical point. The budget is measured after the fact, in the commentary fields as well as the quotations list, so a passage cannot be hidden outside the count. **Poetry, songs and single short stories get zero quotation.** A poem is short enough that even a modest excerpt is a substantial part of the whole work, so those forms are analysed without being reproduced. A request for the work's own text is refused in the browser before any credit is spent, and the in-scope version of the request is offered as a one-click replacement. ## Contract Signed-in runs, metered against the caller's SkillSafe balance. Input: ```json { "title": "Things Fall Apart", "author": "Chinua Achebe", "form": "novel | non-fiction | poetry | song | play | short-story | memoir | essays | reference | graphic-novel | unknown", "reader": "who is asking and why", "ask": "what you want from the brief", "spoilers": "avoid | ending-ok", "depth": "quick | standard | thorough", "sources": [{"id": "SRC-1", "title": "...", "abstract": "...", "url": "..."}], "page_facts": {"records_retrieved": 3, "form_read_as": "novel", "form_read_from": "the retrieved abstract", "short_form": false}, "ask_flags": [{"matched": "...", "why": "...", "instead": "..."}], "follow_up": {"of": "...", "previous": "...", "goal": "..."} } ``` Output is a single JSON object with `found`, `confidence`, `not_found`, `work` (each volatile fact paired with a `_source`), `orientation`, `premise`, `argument`, `structure`, `themes`, `context`, `reception`, `style`, `difficulty`, `for_you`, `read_next`, `quotations`, `editions_note`, `spoiler_note`, `limits` and `declined`. Full field-by-field documentation and worked examples in eight languages: https://book-summary.skillsafe.ai/api.html ## Sources and honesty Grounding is Wikipedia, via the platform's declared `web.wikipedia` search provider, retrieved at run time and shown to the reader in full alongside a list of what each snippet literally states. The app does not claim Wikipedia is authoritative; it claims only that a fact confirmed by a retrieved abstract has been checked against something, and that a fact which has not been is labelled as such. The brief is a starting point for reading, not a reference work and not a citable scholarly source. For anything that matters, read the book. ## Not affiliated Book Summary is an independent tool published on SkillSafe. It is not affiliated with, endorsed by or connected to any publisher, author estate, bookseller or encyclopaedia.