One of the biggest problems with book recommendations is that they're usually too obvious. I don't need another list of books to read after The Lord of the Rings, or someone telling me to try Brandon Sanderson because I like fantasy. I wanted recommendations based on the strange mixture of books I actually enjoy — ones that felt personal, not algorithmic.
So I gave ChatGPT my bookshelf. Metaphorically, at least.
Rather than asking for recommendations straight away, I told ChatGPT to learn my reading taste first. I started listing favorite authors and books, then asked it to quiz me about others I'd forgotten. It wanted to know what I'd enjoyed about particular novels, whether I'd read similar authors, and even the rough timeline of when I'd discovered them, building a picture of the books that had shaped me.
I also gave it some ground rules. It should avoid obvious recommendations unless there was a compelling reason to include them. Every suggestion had to be explained in relation to something I'd already read, even if that connection was simply, "This is nothing like your usual books, but I think you'll love it."
After about half an hour, ChatGPT stopped asking questions and started analyzing me instead.
"Your shelves suggest that you like speculative fiction with a sense of play," it said. "You are drawn to books with elaborate worlds, but you do not seem especially impressed by complexity for its own sake. Humor matters, although you tend to prefer humor that reveals something about the characters or the society around them."
It wasn't a perfect summary, but it was close enough to make me think this experiment might actually work.
Literary profiling
The obvious appeal of feeding ChatGPT a full reading history is that it can spot patterns across hundreds of books at once. I could have described my taste as fantasy, science fiction and comedy, but that would have been far too broad to produce anything useful. ChatGPT noticed that I repeatedly chose books about bureaucratic absurdity, unreliable institutions, strange cities and reluctant heroes who would much rather be somewhere else.
It also noticed my fondness for stories that treat big ideas lightly without treating them as trivial. That led it toward Martha Wells’ Murderbot Diaries, which pair sharp comedy with questions about identity, autonomy and the exhausting burden of dealing with humans. I had already read them, which was mildly disappointing but also reassuring. The system had identified exactly the sort of thing I wanted.
When I told it Murderbot was already familiar territory, it adjusted rather than simply replacing one title with another popular series.
“You appear to like characters who stand slightly outside their own societies and comment on the absurdity around them,” it replied. “I will move away from well-known sarcastic narrators and look for books where the humor comes from social observation, institutional failure or characters trying to remain sensible in deeply unreasonable worlds.”
That shift produced better surprises like The Gone-Away World by Nick Harkaway and The City of Dreaming Books by Walter Moers for its combination of literary obsession, elaborate worldbuilding and gleeful weirdness. It suggested The Dragon Waiting by John M. Ford because I seemed to enjoy alternate histories that trusted the reader to keep up. It also pointed me toward Diana Wynne Jones’ adult novels, noting her lighter touch and sharp understanding of human foolishness.
The recommendations became more convincing when ChatGPT explained what each book might lack. One novel had the humor but less warmth. Another had brilliant worldbuilding but moved slowly. A third matched my interest in satire but was considerably darker than most of the books I had marked as favorites.
Library AI
The experiment improved once I began disagreeing with it. One recommendation leaned too heavily into grim fantasy, a genre I can enjoy in small doses but rarely seek out for relaxation. Another featured a long military campaign, which is usually the point where my attention begins quietly packing a suitcase. Each correction sharpened the next round.
One of its most intriguing suggestions was QualityLand by Marc-Uwe Kling, a satirical science fiction novel. The recommendation came with a warning that the satire was broader and more direct than some of my favorites but that the subject matter fit my interest in technology and systems going wrong in very organized ways.
There were still misses. ChatGPT occasionally became too eager to prove it had discovered a pattern, linking two books because they both contained libraries or because their protagonists were technically immortal. At one point it recommended something almost entirely because it featured a sarcastic demon, which felt less like literary analysis and more like the work of an intern who had skimmed the dust jacket.
Even so, the overall experience was far better than typing “funny fantasy books” into a search bar. And I now have a pretty good reading list for the next few years. My bookshelf had always contained this information. ChatGPT simply read the evidence more patiently than I had.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
➡️ Read our full guide to the best business laptops
1. Best overall:
Dell 14 Premium
2. Best on a budget:
Acer Aspire 5
3. Best MacBook:
Apple MacBook Pro 14-inch (M4)
