Data behind Google AI Overview Study, a HasData measurement of when Google shows an AI Overview and whom it cites. The core panel is 1,100 queries, 100 per intent bucket across 11 buckets, and an AI Overview appeared on 66.5% of them (732 of 1,100). A second pass re-ran the 1,100 queries with organic results captured and added a 100-query travel bucket, 1,200 rows in all, so citation position can be checked against ranking.
The year-over-year chart above is the study's frame, since per-intent trigger rates moved hard between the 2025 run and this panel. The rows behind the 2026 dots are in data/, and the 2025 rates come from the study's earlier run, published in the same article.
The data/ folder holds the panels, aggregates/ the derived tables, method/ the collection and analysis scripts.
| File | Rows | What it holds |
|---|---|---|
data/aio-dataset.csv |
1,100 | The flat panel, one query per row with every SERP feature flag and the AI Overview source list |
data/aio_panel.json |
1,100 | The same panel with the full field set, including the featured-snippet domain |
data/organic_panel.json |
1,200 | The second pass, the 1,100 core queries plus a new travel bucket, organic results captured per query |
data/site_study.json |
113 domains | The niche-site slice of the study |
data/vertical_citations.json |
1 | Citation counts by content vertical, one aggregate object |
aggregates/*.json |
4 files | Derived tables, position odds, top-1 domains, panel summaries |
The two panels join on query, since all 1,100 core queries reappear in the second pass. site_study.json is keyed by domain, and vertical_citations.json is one aggregate object.
aio-dataset.csv has 16 columns. intent names the bucket (symptom, tech, howto, diy, informational, navigational, local, events, purchase, comparison, recipe), query is the search phrase, ai_overview is a 0/1 flag, sources carries the cited domains separated by semicolons inside one CSV cell, and the remaining columns are 0/1 flags for the other SERP features (featured snippet, PAA, knowledge graph, local results, recipes, shopping, inline shopping, immersive products, videos, discussions, events, perspectives).
aio_panel.json carries the same panel as JSON. The bucket field is bucket, the Overview flag is aio, aio_sources is a proper array, and two fields are extra, fs_domain (the featured-snippet holder) and keys (the raw response's top-level keys). organic_panel.json drops those two and adds organic, the captured organic results, and that's the field that lets citation position be compared with ranking.
Queries were sent through the HasData SERP API from one pipeline, US desktop settings, and each response was parsed for the AI Overview block, its source list, and every other SERP feature. The collection scripts (aio_panel.py, organic_panel.py, site_study.py) read the API key from the HASDATA_API_KEY environment variable and write to method/out/. vertical_citations.py makes no API calls, it aggregates per-article audit data collected outside this repo, so its output ships here as data only. The analyze_*.py scripts turn the panels into the aggregate tables.
The study's findings and charts are in the article, and every headline number traces to a row set in these files. charts/aio-position-odds.svg plots the second pass's citation odds by organic position, built from aggregates/position_odds.json.
The dataset is released under CC BY 4.0. You can copy, share, and adapt it, including commercially, as long as you credit HasData with a link to hasdata.com or the study.
Cite the study when you use its findings.
HasData. (2026). Google AI Overview Data by Search Intent (2025 vs 2026). HasData. https://hasdata.com/blog/google-ai-overview-study
@misc{hasdata_google_ai_overview_study,
author = {{HasData}},
title = {Google AI Overview Data by Search Intent (2025 vs 2026)},
year = {2026},
howpublished = {\url{https://hasdata.com/blog/google-ai-overview-study}}
}Every GitHub release of this repository is also archived on Zenodo, for work that has to point at the exact files it used. The DOI 10.5281/zenodo.23101610 always resolves to the latest release, and the Zenodo record lists a separate DOI for each version. GitHub's "Cite this repository" button in the sidebar gives the study reference in APA and BibTeX.
The data comes from search result pages collected for research. Whether and how such collection is appropriate depends on jurisdiction and use, and nothing in this repository is legal advice. Is Web Scraping Legal? covers how we think about the question.
- Google AI Overview Study, the study this data belongs to
- SERP History, a related look at how result pages change over time