Method
How Book Score works
Book Score checks the three places readers find books: your Amazon page, the AI assistants they ask, and the rest of the web. Everything in a report comes from real answers and real pages, checked the same way for every book. Here is how, and where the limits are.
The free check, and your email
The free check takes about a minute. It shows your two scores and your first three fixes straight away. Give us your email and the whole free report opens, and we send you its link. Follow-up emails come only if you tick the box for them, and every one has a way to stop.
Books that aren't on Amazon
Check a book by its title and author instead of a link. We ask the assistants and read the web exactly as for any book, and leave the Amazon parts out of the score: parts we can't check are never counted against a book.
Asking the assistants
We write questions the way a reader asks for a book like yours, without naming it, plus one that asks about your book by name. ChatGPT and Claude answer with live web search on. Gemini answers from what it already knows: Google's terms don't let tools like ours analyse its web-search answers. The free check asks a few questions once. The full report asks 12 questions twice of each assistant: 72 reader answers.
We query provider APIs. Results can differ from the consumer ChatGPT, Claude and Gemini apps and their personalised answers. Mention share describes this question panel only, not all readers, sales or market share. The displayed 90% Wilson interval assumes independent answers; repeated questions and shared models violate that assumption, so it can understate uncertainty. Change flags are screening signals, not proof that an edit caused a change. Compare the same questions, models, territory and settings, and confirm across repeated checks.
Counting books
A helper reads which books each answer recommends. One book is counted once however it is written: with or without its subtitle, "The" or no "The", authors in either order. A book or author named in the question itself is not counted as a recommendation in that answer.
Whether an answer named your book is decided by a fixed rule, not by the helper. The title has to be your book, not another book that shares its words. It doesn't count when the answer credits it to a different writer ("by…"), when it is only part of a longer title, when it sits inside a web address, or when the subtitle given is another book's. Plain titles such as "Crochet for Beginners" belong to many books, so this matters most for them. A website an answer cites is a source it read, not a recommendation: a citation counts as a mention of your book only when that page names your book.
Places to get onto
Every website an answer cited, plus the sites AI cites most for your genre. "Write to these" lists only sites where we found a way in for books on the site's own pages: a submissions page, an email, a form, with any fee copied exactly as the site states it. A full report checks the strongest unchecked sites while it runs. Shops, big publishers and the health and research sites AI reads for facts are not places to pitch. A site gets a thank-you note instead of a pitch only when we open its page and find your book and your name on it.
Your Amazon page
We read your listing the way a shopper does and check it against Amazon's own rules from KDP's help pages: the description, title words, categories, A+ panels, ratings, what reviews say, the cover at thumbnail size, formats and price. Rules we quote are checked on Amazon's pages, with the date.
The score
Book Score out of 100 adds up the parts we can check: AI (40: up to 30 for how often it recommends the book, full marks from 4 answers in 10, and 10 for describing the book correctly when asked about it by name), the Amazon page (about 30: a few checks only apply to some formats), the web around the book (20) and formats (10). Parts we couldn't check are left out rather than counted against you. The report leads with two numbers: the Book Score, and the AI Score, which is the AI part on its own, out of 100 (its 40 points scaled). Beside them sits how often AI named the book, with its panel sampling range. Estimated points are points available under our scoring rules, not a forecast of sales or improved recommendations. Task times are estimates and exclude third-party response times.
The AI Shelf
Every week we ask the same 12 reader questions per genre, twice each, of all three assistants, and chart the books they name. A list is published only when every assistant answered. See the AI Shelf.
What we never do
We never invent praise, awards or numbers in the fixes we write, never log in to your Amazon or KDP account, and never present a simulated assistant as a real one. A score is a diagnosis, not a promise of sales.