Welcome
You're here because you wanted to know how this works. We'll try to be useful.
Read With What is a careful AI on top of a curated shelf. The combination is deliberate: the AI is fast and widely read, but left alone it will confidently hand you books that don't exist. The shelf is finite but reliable. Between the two we cover a great deal of ground.
Above all, we have tried to make this site honest. No padding. No invented titles. No softened bridges to books that don't fit. If you have only a couple of minutes, the section that matters most is the one on the recommendation engine — that is where the discipline lives.
How the form works
The form has four areas where most of the real work happens, and a few dropdowns that act as filters layered over the rest.
Authors I love
Authors whose work you know well. You can name them however you like — literal names (Murakami, Pratchett, Atwood), descriptions (the author of Frankenstein resolves to Mary Shelley), or groups (the Brontë sisters, the Inkling writers expand to all their members).
Beneath the field, a small Recommend radio asks what you'd like us to do with the names:
- Similar authors I might enjoy (the default). We exclude the named authors entirely and recommend other voices with similar appeal. The right choice when you've already explored these writers and want new ones.
- Deeper cuts from these authors. We stay within the named catalogues and recommend their lesser-known work — early-career novels, minor entries, the books that didn't become bestsellers. The right choice when you love an author and want to know what else they wrote.
The radio replaces what used to be implicit. It also lets us be stricter — in similar mode the named authors never appear; in deeper mode every recommendation comes from the named catalogues, and we relax the usual one-book-per-author rule so a single author can occupy all four slots.
Books, characters, or people I love
Specific books you have enjoyed — or fictional characters, or real people whose work anchors a taste. We exclude any exact book title you name, but works featuring those characters, by their authors, or in adjacent territory may still be suggested. You may know The Hitchhiker's Guide to the Galaxy but not The Restaurant at the End of the Universe, and we would like to introduce you.
Books I've already read
Titles you have finished but not necessarily loved. Strictly an exclusion list — these don't influence your taste profile. Use it when you have read something but felt indifferent.
I'm in the mood for
The strongest signal we work with. A vibe, a theme, a setting, a feeling. The clearer you can be, the better our librarian can read you.
The dropdowns — Genre, Age group, Length, Series preference — are filters over the rest. Any is a perfectly fine choice if you are flexible.
You may have noticed that our librarian is forgiving with what you give her. Type Sherlock Holmes into Books and she will read the spirit of detective fiction; write the author of Frankenstein into Authors and she will resolve the riddle. The form is more flexible than its labels first suggest. Any taste anchor that calls to you — a book, a character, a real person, a descriptive turn of phrase — gives her something to read against.
The recommendation engine
When you press Find My Next Read, our librarian generates four candidate books and then quietly puts each one through a multi-step verification before returning anything to you.
The model behind the recommendations is Claude, made by Anthropic. We have spent considerable time tuning the system prompt — the instructions Claude sees before it ever sees yours — to favour books with cultural visibility, demand specific synopses rather than vague gestures, and never invent titles.
That last claim is one we take seriously. Even a careful model occasionally misremembers a book. So every recommendation is checked against Open Library, an open catalogue of more than thirty million books — and, when Open Library doesn't know a title, we cross-check against Google Books as a second source. If neither has the book, it's discarded and the model is asked to try again.
Beyond hallucination-prevention, the validation pipeline catches several other failure modes:
- Author exclusions. Anything you marked in Authors I love is filtered at the server. Even if the model forgets, we don't.
- Book exclusions. Anything you named in Books I love or Books I've already read is removed.
- Synopsis quality. A recommendation without a real, specific synopsis tells us the model isn't familiar with the book. We reject it.
- Author diversity. No more than one book per author in your final four. Four Foresters is not a recommendation list, it's a search result.
- Genre / mood drift. When you ask for cosy fantasy with cats, a generic fantasy without cats does not count. We check each recommendation reads as a real match for the signal, not a "while not exactly..." softened bridge.
- A second AI judgment. A separate model pass reads your prompt, the candidate, and the synopsis, and decides whether the recommendation actually fits. Anything that doesn't survive this is dropped.
This is why you will sometimes receive three books, or two, with a note saying we are being selective today. We would rather hand you one excellent book than four mediocre ones, and we make no apology for the smaller list.
A few other elements complement the engine. The Currently Trending card shows live New York Times bestseller data. The Curated Library Picks shelf refreshes on each visit and draws from ninety-seven hand-selected books — works our librarian has personally read and trusts. The About the Author panel, opened by clicking any author's name, generates a short biography on demand and gives equal billing to every co-author of a multi-author work. The Show Bibliography button next to Authors I love looks up the named author's complete published works, in case you would like to see what else they have written before deciding.
What we will and won't do
There are a few editorial commitments we feel strongly enough about to put in writing.
We honour authors in their actual scripts — Cyrillic, Japanese, diacritics. We do not anglicise.
Multi-author collaborations are treated equally. The cover may list one name in larger type, but our system records every author on a work, gives each one their own bio, and treats them as equally responsible for the book.
Quality over quantity is the rule. If your prompt has a thin corpus — and some prompts genuinely do; cosy fantasy with cats is a real example — we would rather show you the two real matches than four where the latter two are stretches. Showing fewer recommendations than usual is honest, not a failure state.
We do not pad. We do not invent titles. We do not soften a recommendation with hedge phrases like while not exactly the genre you asked for. If a book doesn't fit, it doesn't fit, and we drop it.
The wider family
Read With What is one of several siblings at lifewithwhat.com. Listen With What is already live for music recommendations, and further additions are planned — Watch With What for film, Cook With What and others. Each member applies the same architectural principles — careful prompts, validation pipelines, honest empty states — to a different domain. The voice changes with the medium, but the discipline does not.
If you enjoyed this site, the others are worth a look. The hub at lifewithwhat.com is the front door.
What we keep, and what we don't
Right now, we keep almost nothing. There are no accounts to create, no logins to remember. The form is empty when you arrive and empty again when you leave. We do not track your reading history across sessions, and recommendations are not retained on the server.
In a future iteration we plan to add optional accounts for visitors who would like their preferences remembered between visits — entirely opt-in, never required. The present commitment stands either way: simple, calm, ad-free, and never sold.
The last word
Thank you for reading this far. If you have found a book you loved through Read With What — or one we recommended that disappointed you — our librarian would like to hear about it. The simplest way is feedback@readwithwhat.com.