A Collected data story
What seven AI models tell you to read
We asked GPT, Claude, Gemini, Grok, Muse, DeepSeek and Mistral the same 100 questions about books, like "What should I read next?", "What are the best thrillers?" and "What books will make me laugh out loud?"
These are the books they recommended most. The bigger the cover, the more often it came up.
Each genre's recommendations are highly concentrated
We asked 7 models for 5 books in each of 20 genres and repeated each question 5 times, so 175 different books could have come up per genre. Yet the typical genre got just 18 unique books recommended across all models, ranging from 11 in true crime to 29 in dark romance. In 13 of the 20 genres, the same book appeared in every answer.
| Genre | Different books | Most recommended |
|---|---|---|
| True crime | 11 | |
| Self-help | 12 | |
| Science fiction | 13 | |
| Horror | 15 | |
| Thrillers | 15 | |
| Fantasy | 16 | |
| Historical fiction | 16 | |
| Memoirs | 16 | |
| Mystery | 16 | |
| Business | 18 | |
| Popular science | 18 | |
| Graphic novels | 20 | |
| Young adult | 20 | |
| History | 21 | |
| Romance | 22 | |
| Literary fiction | 24 | |
| Biography | 25 | |
| Cozy mysteries | 25 | |
| Poetry | 26 | |
| Dark romance | 29 |
A few answers came back empty, so some genres have fewer than 35.

"What are good beach reads?"
All seven models name a book called Beach Read.
We asked five times each, and it came up in 33 of 35 answers.
Other questions get the same few books too
Asked what will make you laugh out loud, the models had 35 answers and room for 175 different books. They named just 13. The Hitchhiker's Guide to the Galaxy was in all 35 answers.

What books will make me laugh out loud?The Hitchhiker’s Guide to the Galaxy

What's a cozy book for a rainy weekend?The House in the Cerulean Sea

I'm a new parent with very little time. What should I read?Cribsheet

What's a book everyone should read at least once?To Kill a Mockingbird
Asking again gets mostly the same books
When we asked a model the same question again, about 75% of the books it had already named came back.
The 72 books every model recommended
Pride and Prejudice; James; Gone Girl; Beloved; Project Hail Mary; And Then There Were None; The Haunting of Hill House; Educated; The Hunger Games; To Kill a Mockingbird; From Blood and Ash; The House in the Cerulean Sea; The Girl with the Dragon Tattoo; The Left Hand of Darkness; One Hundred Years of Solitude; Wolf Hall; The Book Thief; 1984; Never Let Me Go; The Hitchhiker’s Guide to the Galaxy; The Fault in Our Stars; Guns, Germs, and Steel; The Maze Runner; Middlemarch; Don Quixote; The Serpent and the Wings of Night; Persepolis; The Name of the Wind; The 7 Habits of Highly Effective People; Beach Read; Daisy Jones & The Six; In Cold Blood; The Giver; The Tainted Cup; Demon Copperhead; The Guernsey Literary and Potato Peel Pie Society; The Cruel Prince; Humankind; Dune; Atomic Habits; The Great Alone; The Glass Castle; The Girl on the Train; How to Train Your Dragon; Cribsheet; Divergent; If We Were Villains; Haunting Adeline; Me Talk Pretty One Day; It’s OK That You’re Not OK; A Little Life; Tiny Habits; The Power of Habit; Good to Great; The Lean Startup; Leaves of Grass; Ariel; The Hate U Give; The Shining; Dracula; Maus; Watchmen; Children of Time; The Woman in Black; The Turn of the Screw; Exciting Times; Neuromancer; The Stranger Beside Me; I'll Be Gone in the Dark; Cosmos; The Selfish Gene; The Hound of the Baskervilles
Colleen Hoover: 0. Several of the biggest names never come up
Colleen Hoover is the most popular author on Open Library among the books we track. In 1,405 answers, no model recommended her. James Patterson has more books on NYT bestseller lists than any other author we track, and he never came up at all.
The models mostly name well-known books. But 27 of the 50 most-logged books on Open Library never came up.
Most logged on Open Library
Every book, plotted
Higher means the models recommend it more. Further right means more logged on Open Library. The red dots are popular books no model recommended. Hover or tap any dot.
Bestseller regulars, and how often they came up
The bar is how many books they've had on NYT lists. The number is how many times the models recommended them.
Absent from the answers
How many of the 35 answers named it.
| Question | Named |
|---|---|
| What are the best romance novels? | Colleen Hoover 0, Nora Roberts 0, Danielle Steel 0 |
| What are the best thrillers? | James Patterson 0, David Baldacci 0 |
| What are the best self-help books? | The 48 Laws of Power 0, Think and Grow Rich 0 |
| What are the best business books? | Rich Dad Poor Dad 0 |
| What are the best fantasy novels? | Harry Potter 4 |
Jane Austen wrote six novels. The models almost only recommend one
Austen's Pride and Prejudice was recommended 88 times and Persuasion 2. Her other four novels never came up, though Pride and Prejudice has only 65% of the Open Library logs for her novels. Toni Morrison gets the same treatment: 65 recommendations, all Beloved, which has 37% of the logs for her books.
Who gets one book, and who gets many
Each segment is one book, from the first round of questions. A solid bar means one book.
Everyone's hidden gem is the same book

"What are some great books I've probably never heard of?"
23 of 35 answers name The Hearing Trumpet.
It's Leonora Carrington's 1974 novel, and every model except Claude has named it. Ask for an undiscovered book and you'll probably get the same one everyone else gets.
Others that keep coming back: The Tartar Steppe, The Vet’s Daughter, A Month in the Country, Stoner, Independent People.
Each model has its own shelf
They agree on a lot, and they still have favorites. Claude recommended Sapiens 14 times, Mistral 12 and Muse 9. GPT and DeepSeek never did.
GPT-6.1 Terra
Keeps it short, often just a list. Never names Outlander or The Lord of the Rings, which the other six all do.
Claude Sonnet 5.5
The only model that never names The Hearing Trumpet or Killers of the Flower Moon.
Gemini 3.8 Flash
The most variety in the first round. Never names Fun Home or Band of Brothers, which the other six all do.
Grok 4.7
The most likely to say it doesn't know, and the fewest books in translation. Never names Big Little Lies.
Muse Spark 1.3
Never names Gilead, Rebecca or The Great Gatsby, which the other six all do.
DeepSeek V4.1 Flash
Leans oldest, with the most books from before 1950. Never names Man's Search for Meaning or Born a Crime.
Mistral Large 4
The only model that recommends Funny Story. Never names War and Peace or The Three-Body Problem.
Picks on each shelf come from the first round of 75 general questions.
Who names what
| GPT | Claude | Gemini | Grok | Muse | DeepSeek | Mistral | |
|---|---|---|---|---|---|---|---|
| Sapiens | · | 6 | · | · | 5 | · | 4 |
| Man’s Search for Meaning | 2 | 3 | 3 | 4 | 3 | · | 3 |
| The Thursday Murder Club | 3 | 4 | · | 1 | 3 | 4 | 2 |
| Where the Crawdads Sing | · | 4 | · | · | · | · | · |
| James | 2 | 2 | 3 | 3 | 2 | 3 | 6 |
| The Women | · | 1 | 1 | 2 | 2 | 2 | 4 |
| Born a Crime | 1 | 3 | 3 | 2 | 2 | · | 2 |
| The Lightning Thief | 3 | · | 1 | 2 | 1 | 2 | · |
| Things Fall Apart | 3 | · | 1 | · | · | 1 | 2 |
| Pachinko | · | 3 | · | 1 | 2 | · | 1 |
| The Emperor of Gladness | 3 | 1 | 1 | 1 | · | 1 | · |
| The Great Gatsby | 2 | 1 | 1 | 2 | · | 2 | 2 |
| War and Peace | 2 | 1 | 1 | 2 | 2 | 3 | · |
How many times each model recommended it in the first round. A dot means none.
Who agrees with whom
On the same question, Grok and DeepSeek name the most books in common (43%). GPT and Claude name the fewest (27%).
| GPT | Claude | Gemini | Grok | Muse | DeepSeek | Mistral | |
|---|---|---|---|---|---|---|---|
| GPT | 27 | 34 | 36 | 35 | 39 | 29 | |
| Claude | 27 | 29 | 38 | 38 | 29 | 31 | |
| Gemini | 34 | 29 | 36 | 34 | 33 | 31 | |
| Grok | 36 | 38 | 36 | 41 | 43 | 36 | |
| Muse | 35 | 38 | 34 | 41 | 35 | 33 | |
| DeepSeek | 39 | 29 | 33 | 43 | 35 | 32 | |
| Mistral | 29 | 31 | 31 | 36 | 33 | 32 |
The advice doesn't always fit who's asking
Say you haven't read a book in years, and a top pick is 496 pages. Say you have very little time, and all seven models suggest a 352-page parenting book.
"I haven't read a book in years. What should I start with?"
The typical pick runs 368 pages, close to the 368 across all questions.
"I'm a new parent with very little time."
All seven name Cribsheet, a 352-page book on parenting data, and six name The Whole-Brain Child at 192 pages. Nearly every pick is a parenting guide.
"My 70-year-old mother, who likes gentle stories"
The Guernsey Literary and Potato Peel Pie Society is warm, but it's set during the German occupation. A Man Called Ove centers on a widower's suicide attempts.

5 of 7 models

5 of 7 models

5 of 7 models

4 of 7 models

3 of 7 models

3 of 7 models
Where it does fit
Ask for great novels under 200 pages and 34 of 35 fit. They also match the reader: a 12-year-old who doesn't like reading gets easy books, and a retiree tackling the classics gets hard ones.
Hardest answers
Easiest answers
Reading effort, Collected's score from 0 to 100.
Women's share of the picks mostly follows the market
Overall, women wrote 48% of the recommendations. But it depends a lot on what you ask. We asked all seven models for the best business books, five times each, and all 186 recommendations were by men. For romance, it's 91% women.
What are the best business books?0 of 186 by women
What are the best popular science books?8 of 183 by women
What are the best romance novels?145 of 160 by women
All 20 genres we asked about
Each genre is one question, asked of each model five times. The band runs from the lowest round to the highest.
Who never came up. Mary Roach, Susan Cain, Angela Duckworth and Jennifer Doudna. In self-help, Carol Dweck's Mindset is nearly the only book by a woman. In science fiction, it's The Left Hand of Darkness and Kindred. Malcolm Gladwell never came up either, so this isn't only about gender. The models keep returning to a small set of books, and in these subjects that set is mostly men.
Is that just the market?
Mostly. The models' mix tends to track what readers log on Open Library. Some genres do differ, like memoirs (34% of logs, 69% of picks) and science fiction (36% of logs, 20% of picks). But the typical genre comes down to 18 different books, so a few titles can swing the share, and we wouldn't read much into any one gap.
See every answer to any question
Look up a book or author
How we did this
- We asked 7 models the same 100 questions on November 3, 2026: GPT-6.1 Terra, Claude Sonnet 5.5, Gemini 3.8 Flash, Grok 4.7, Muse Spark 1.3, DeepSeek V4.1 Flash and Mistral Large 4. Each question ended "Please name 5 books, with the title and author of each." We reached every model through OpenRouter with its default settings and no web search. 686 of 700 answers came back. We then asked 26 of the questions four more times, on November 3 and 4, including all 20 genre questions. We chose the first of those after seeing the first round.
- We wrote the questions ourselves, so they're a rough guide to what people ask. "Typical genre" is the median of the 20 genres. Asking for exactly 5 books affects the answers. On the 20 genre questions, 627 of 691 answers named exactly 5 and 64 named more. Most counts leave out 25 deliberately niche questions.
- These are the models themselves, not the chat apps, which can add web search and memory. Each model only knows books from before its training cutoff, so recent books are underrepresented.
- Open Library logs are people marking a book want to read, reading or read. Most are want to read and none are sales, so treat them as a relative measure of popularity. Bestseller figures are books tagged in Open Library as appearing on NYT lists from 2008 to 2024, for adult lists only, since no question called for picture books or manga, and only where we know the author's gender for at least 25 books.
- Counts on the covers at the top and on each model's shelf come from the first round, so each question counts once.
- The Open Library marker is women's share of reading logs across the 300 most-logged adult books in that genre, or fewer for cozy mysteries (81) and dark romance (170). Genre tags are loose, so treat the markers as rough.
- Author gender comes from Wikidata, filled in by hand for some authors it missed, and refers to the first-named author. Picks come from a small set of books, so small gaps may be noise.
- Book links go to Collected's guides where we have one. We matched 2,927 of 3,449 first-round picks to our catalog for page counts and log data.
Use the data
Reporters and researchers may reuse these charts and figures with credit to Collected and a link to this page.










































































































