Last reviewed by Robert Prime — October 2026
Update, October 2026: this study ran on our first AI tool, which Book Score has now replaced. The findings below are the July 2026 snapshot. Our live measurement is the weekly AI Shelf, which charts the books ChatGPT, Claude and Gemini recommend in 16 genres, and Book Score checks your own book against the same kind of questions.
Readers have started asking AI what to read next. The obvious question for any author — does AI recommend my book? — turns out to have a measurable answer, and we built the instrument to measure it: the AI Book Discoverability Index, an index of every source the AI engines cite when recommending books.
This guide walks through what the data shows — and what to do about it.
How the index works
We ask the four citation-transparent engines — ChatGPT, Claude, Gemini and Perplexity — the questions real readers ask: “best [genre] books”, “books like X”, “what should I read about…”. Every source they cite gets logged. As of 10 June 2026 the index held 91,000+ live citations across 916 audited books in 16 genres, and it grows with every audit (now 141,282+ live citations across 981 audited books as of July 2026). Numbers trace to live citations, not model memory — and three rules apply: simulated-assistant runs are excluded, a domain must be cited for at least two different books to rank, and any domain sourced ≥70% from one title is excluded.
Finding 1: A quarter of books are invisible
27% of audited books are never named by any engine, for any reader question. The median AI Shelf Score is 17/100, and only 3% of books are recommended reliably by name. AI doesn’t have a long tail — it has a short shelf, and most books aren’t on it.
Finding 2: Reddit barely matters for books
Across the wider web, Reddit is among AI’s most-cited domains. For book recommendations it’s under 1% — rank #11 of 5,881 sources at the time of writing in the index. The engines build book answers from Goodreads, Wikipedia, YouTube and specialist best-of lists instead. If you’ve been told to “win Reddit” for AI visibility, the data says: for books, put that effort into the sources AI actually cites.
Finding 3: Retailers aren’t where AI learns
Every bookshop trails the reference and curation sources. AI cites where it learns about books — review sites, best-of lists, reference pages — far more than where readers buy them. A strong Amazon listing matters for conversion; it barely registers for AI discovery.
Finding 4: Every genre has a kingmaker
Each genre has niche sites that punch far above their weight — the hidden gatekeepers AI trusts for that shelf. In personal finance, a credit union’s “best money books” list is cited across 54 different audited books. In true crime it’s CrimeReads; in young adult, ReadBrightly. The index’s per-genre lists name the top sources for every genre — that’s an author’s submission map.
What to do with this
- Check where you stand. The free Book Score check reads your Amazon page and asks ChatGPT, Claude and Gemini the questions readers ask, in about a minute.
- Target your genre’s sources. Work the sites AI cites for your genre, starting with the kingmakers above — those are the places citations come from. The AI Shelf shows which books AI names in your genre each week.
- Fix the basics AI reads. A recognisable author entity (Goodreads, Wikidata), structured data, and a plain-spoken description — covered in our AI book discovery guide.
For the books AI recommends right now, see this week’s AI Shelf.
