Datasetpublic

sempite/ai-overview-book-discovery-citations

Who does Google's AI cite when readers ask what to read next? Canonical release: https://doi.org/10.5281/zenodo.22852307 This repository mirrors that deposit. Cite the DOI. The finding 16 reader buying-intent queries, run through Google with AI Overview capture on 13 August 2026. Eleven returned an AI Overview, carrying 95 citations between them across 38 unique domains. Not one went to a website controlled by an author. Category Citations… See the full description on the dataset page: https://huggingface.co/datasets/sempite/ai-overview-book-discovery-citations.

sourceHugging Facecc-by-4.0updated 1d agoView on Hugging Face

0likes18downloads

Dataset Card

Who does Google's AI cite when readers ask what to read next?

Canonical release: https://doi.org/10.5281/zenodo.22852307 This repository mirrors that deposit. Cite the DOI.

The finding

16 reader buying-intent queries, run through Google with AI Overview capture on 13 August 2026. Eleven returned an AI Overview, carrying 95 citations between them across 38 unique domains.

Not one went to a website controlled by an author.

Category	Citations
community_social	31
press_media	16
retailer	13
reader_platform	13
book_blog	10
search_engine	4
publisher	4
library	2
other_commercial	2
author_owned	0

Community platforms carry a third of everything, led by YouTube (11) and Reddit (10). Independent book blogs earn real citations, which makes blogger outreach a viable path. Niche queries surface niche gatekeepers: the Catskills cluster cited hadassahmagazine.org, jewishbookcouncil.org and kveller.com, none of which appear anywhere else in the sample.

Categories were assigned by hand

Every cited domain was read and classified from its page title and URL. The mapping ships as domain_categories.csv so any call can be disputed and the totals re-tallied. Nothing here was classified by a model.

author_owned is deliberately narrow: a publisher's page about an author, or a retailer's author page, is not author-owned. The finding is about domains an author controls.

Configs

Config	Rows	What it is
`citations`	95	one row per citation, with URL, position in the answer and assigned category
`queries`	16	one row per query, with whether an AI Overview appeared
`domain_categories`	38	the hand-assigned domain to category mapping

Also included: data/raw_run.json with every answer in full, the collection and tally code, and METHODS.md with an eight-point limitations section.

Be careful with the headline

Sixteen queries collected on a single day is a probe, not a census. AI Overviews are volatile, the query set was chosen around one author's niche rather than sampled at random, and citations are counted per reference rather than per unique domain (the same domain appears twice in one answer 10 times).

The defensible reading is that author websites are not a meaningful citation surface for these queries. Not that an author site can never be cited.

Citation

bibtex

@dataset{gerrish_book_discovery_2026,
  author    = {Gerrish, Jacob Ezra},
  title     = {Who does Google's AI cite when readers ask what to read next?
               A citation study of 95 sources in AI Overviews for book discovery},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.22852307}
}

Write-up: https://sempite.com/research/ai-answers-book-discovery/

Data CC BY 4.0, code MIT. Produced by SEMPITE.