Readers now find books in three places, and each one decides differently. Amazon ranks books from what you tell it (keywords, categories, your description) and from how the book sells. AI assistants such as ChatGPT, Claude, Gemini and Amazon’s Rufus name books from what is written about them across the web. The open web (Google, Goodreads, reviews, best-of lists) feeds both.
Most authors work on one of the three and never measure the other two. This page explains what decides each, what our research shows, and how to check your own book.
How Amazon decides which books readers see
Amazon’s own help pages say it plainly: relevant keywords, along with factors like sales history and Amazon Best Sellers Rank, decide where a book appears in search (Amazon KDP: keywords). What you control:
- Keywords. KDP gives you up to seven keywords or short phrases. Use the words readers type, not words already in your title or categories. Guides: how to fill all seven keyword boxes · KDP keyword research
- Categories. You choose three, and Amazon says changes can take up to 72 hours to show (Amazon KDP: categories). Guide: choosing the right categories
- Title, description and comp titles. These tell Amazon and the reader what the book is. Guides: title strategy · writing a description that sells · comp titles
- The page that converts. Once a reader lands, the Look Inside, your A+ Content and your reviews decide whether they buy, and every sale feeds the ranking. Guides: Look Inside optimisation · A+ Content modules · getting reviews · Author Central
- Paid visibility. Amazon Ads buy a place in search while organic ranking builds. Guide: Amazon Ads for authors
If the book is live and not selling, start with why is my KDP book not selling?
How AI assistants decide which books to recommend
When a reader asks ChatGPT, Claude or Gemini “what should I read next?”, the answer comes from what the assistant has read about books across the web, not from your Amazon page. We measure this. Our July 2026 study logged 141,282 citations across 981 books in 16 genres, and found:
- 27% of books were never named by any assistant, for any reader question. The median score was 17 out of 100.
- Retailers are not where AI learns. Review sites, reference pages and best-of lists are cited far more than bookshops.
- Reddit barely matters for books: under 1% of citations.
- Every genre has a “kingmaker”, a few niche sites AI trusts for that shelf. Those are an author’s submission list.
The full findings: inside the AI Book Discoverability Index. Every week we now chart the books AI names in each genre on the AI Shelf.
Guides by assistant: does ChatGPT recommend books? · how to get ChatGPT to recommend your book · does Gemini recommend books? · Amazon Rufus · how AI recommends books · AEO for authors
The open web: what both of them read
Google, Goodreads, editorial reviews, best-of lists and your own author website are where both Amazon shoppers and AI assistants pick up signals about a book. Guides: Goodreads for authors · editorial reviews and blurbs · author website SEO for AI search · backlist optimisation
Check your book: Book Score
Book Score is publishing.co.uk’s book discoverability audit. Paste your Amazon link and the free check scores your book out of 100 for how readers find it on Amazon, in ChatGPT, Claude and Gemini, and across the web, and gives you the three fixes worth most, in about a minute. The full report adds 72 real AI answers, the books AI names instead, the sites worth contacting and every fix written.
- Free check, no sign-up
- Full report: £49
- Report and Watch: £99 a year
- Done for you: £495
- Publishers and agencies: from £199 a month + VAT
Prices include UK VAT unless marked. Run the free Book Score check →
Fix it yourself, or have it done
- Learn it: the AI Discoverability for Authors course, £149, or £249 done for you.
- Your Amazon page: A+ Content design, from £89.
- Your home on the web: an author website, built to be found by Google and AI.
- A whole launch: the 14-day book launch, from £499.