The key numbers
Seven findings
- AI Overviews are nearly universal. 90% of searches got one, with 8.0 sources on average.
- Online casino is the exception. Only 20% of casino searches got an AI Overview, and 1% of the core casino questions.
- Ranking matters, and so do sub-questions. 72% of #1 results were cited, and 92% of answers also cite pages from outside the top 20.
- Each industry trusts a different kind of source. Vendors in software, banks and insurers in finance, government and institutions in health and law, review sites in betting and insurance.
- YouTube is the biggest single source, with 14% of all citations. The community platform varies by country: Reddit in English markets, Facebook and Instagram in Latin America.
- Visibility is local. Most cited pages are in the market’s language, and markets share few sources even for the same question.
- Meaning and freshness beat formatting, in every language. Meaning match (AUC 0.56) and newer dates (0.55) predicted citation in all 10 markets; FAQ schema, TL;DRs and question headings barely registered.
Contents
How often do Google AI Overviews appear?
Short answer: On 90% of searches across all ten countries. Online casino is the one exception, at 20%.
AI Overviews appeared on 90% of searches overall, from 83% in Spain to 95% in Japan. Finance, insurance, loans and legal topics, which are often treated as sensitive, still got AI Overviews on 97–98% of searches.
Online casino is the clear exception at 20%. The shared core casino questions, such as how to play blackjack or what RTP means, got an AI Overview 1% of the time across all ten markets. Sports betting sits in between at 76%, and varies widely by country.
AI Overview rate by industry and country
Share of the 200 searches in each industry and country that showed an AI Overview.
| US | UK | IN | AU | DE | FR | ES | MX | BR | JP | |
|---|---|---|---|---|---|---|---|---|---|---|
| Online casino | 16% | 6% | 20% | 12% | 18% | 15% | 14% | 22% | 24% | 51% |
| Sports betting | 92% | 82% | 76% | 78% | 72% | 70% | 50% | 62% | 81% | 94% |
| Home improvement | 98% | 92% | 99% | 98% | 82% | 75% | 66% | 94% | 98% | 98% |
| E-commerce & products | 98% | 86% | 97% | 96% | 82% | 87% | 80% | 98% | 98% | 100% |
| Crypto | 96% | 91% | 98% | 96% | 94% | 88% | 88% | 92% | 95% | 99% |
| Health | 98% | 93% | 98% | 96% | 89% | 91% | 84% | 96% | 99% | 99% |
| Travel | 98% | 94% | 98% | 97% | 92% | 93% | 86% | 96% | 98% | 98% |
| Automotive | 98% | 98% | 98% | 100% | 92% | 92% | 88% | 96% | 98% | 98% |
| Marketing & SEO | 97% | 96% | 96% | 98% | 94% | 95% | 90% | 96% | 98% | 99% |
| Education & careers | 100% | 96% | 98% | 96% | 85% | 96% | 96% | 98% | 98% | 96% |
| SaaS & B2B software | 98% | 96% | 97% | 97% | 94% | 95% | 94% | 98% | 96% | 97% |
| Legal | 97% | 96% | 100% | 96% | 97% | 97% | 96% | 97% | 99% | 98% |
| Personal finance | 99% | 98% | 98% | 98% | 94% | 98% | 97% | 98% | 98% | 98% |
| Real estate | 98% | 98% | 98% | 100% | 100% | 97% | 92% | 97% | 98% | 99% |
| Trading & forex | 99% | 100% | 99% | 98% | 92% | 98% | 96% | 98% | 98% | 98% |
| Insurance | 98% | 98% | 98% | 97% | 99% | 100% | 94% | 98% | 99% | 98% |
| Loans | 100% | 98% | 100% | 98% | 99% | 98% | 95% | 98% | 97% | 99% |
What it means: For almost every commercial topic, AI Overviews are now the default. Casino content is the one area where Google holds them back, regardless of local law.
Does ranking on Google decide which pages AI Overviews cite?
Short answer: Yes. 72% of #1 results are cited, against 6% at positions 16–20. But 92% of AI Overviews also cite pages that don't rank in the top 20.
Ranking still matters most. 72% of #1 results were cited, falling to 33% at positions 7–10 and 6% at 16–20. Half of all citations (51%) are top-10 results.
Yet 92% of AI Overviews cite at least one page that does not rank in the top 20 for the search. Those pages rank for related sub-questions that Google runs in the background (query fan-out). Japan leans most on the top 10 (65% of citations), the US and UK least (44% and 42%).
Share of organic results cited, by Google position
All countries. Organic results 1–20 for searches with an AI Overview.
Citation rate by position band, per country
Share of organic results in each band that the AI Overview cited.
| #1–3 | #4–10 | #11–20 | |
|---|---|---|---|
| United States | 58% | 40% | 12% |
| United Kingdom | 55% | 36% | 11% |
| India | 60% | 39% | 12% |
| Australia | 62% | 42% | 12% |
| Germany | 70% | 43% | 9% |
| France | 68% | 41% | 7% |
| Spain | 67% | 40% | 6% |
| Mexico | 62% | 36% | 5% |
| Brazil | 63% | 38% | 7% |
| Japan | 68% | 38% | 9% |
What it means: Rank for the main query and for the follow-up questions around it. Both routes lead into the answer.
What types of sources do AI Overviews cite in each industry?
Short answer: It depends on the industry: software vendors in SaaS, banks and insurers in finance, government and institutions in health, education and law, and review sites in betting and insurance.
The type of source Google trusts changes completely with the topic. Software and marketing answers lean on vendors’ own sites. Loans, insurance, personal finance, trading and crypto lean on banks, insurers, brokers and exchanges. Health, education and legal answers lean on government bodies and institutions. Sports betting and insurance are where review and comparison sites carry the most weight.
Markets differ less than industries do, but Japan stands out for brand sites, Brazil and Mexico for video and social, and the UK, France and Australia for government sources.
Mix of cited source types, by industry
Share of all citations in the industry. Domains with fewer than 3 citations are unclassified.
Mix of cited source types, by country
Same categories, all industries combined.
What it means: Know which kind of source wins in your industry. If answers cite banks and regulators, a brand blog competes on different terms than in software, where vendor pages are cited freely.
Which platforms do AI Overviews cite: YouTube, Reddit or Facebook?
Short answer: YouTube is cited everywhere (14% of all citations). Reddit is cited mainly in English-speaking countries; Facebook and Instagram fill that role in Mexico and Brazil.
YouTube accounts for 14% of all citations and appears in 55% of AI Overviews, from 21% of citations in Brazil to 9% in Japan. How-to searches cite video the most.
Reddit is an English-market source: 5.9% of citations in the US, 4.8% in Australia and 4.1% in the UK, but 1.1% in Germany and 0.1% in Japan. In Mexico and Brazil, Facebook and Instagram take that role (11% and 9% of citations). Japan relies on its own Q&A and blog platforms instead.
Share of citations going to each platform, by country
Percent of all citations in the country.
| YouTube | TikTok | Quora | Wikipedia | ||||||
|---|---|---|---|---|---|---|---|---|---|
| United States | 16.1 | 5.9 | 1.8 | 0.7 | 0.2 | 0.9 | 0.4 | 0.4 | 1.9 |
| United Kingdom | 12.6 | 4.1 | 2.1 | 0.7 | 0.4 | 0.5 | 0.4 | 0.5 | 1.1 |
| India | 12.1 | 3.3 | 2.1 | 1.9 | 0.0 | 2.1 | 0.8 | 0.7 | 2.0 |
| Australia | 11.8 | 4.8 | 2.3 | 0.7 | 0.2 | 0.5 | 0.5 | 0.5 | 2.0 |
| Germany | 14.4 | 1.1 | 0.6 | 0.7 | 0.3 | 0.0 | 0.1 | 0.5 | 0.6 |
| France | 13.6 | 0.5 | 2.5 | 1.0 | 0.6 | 0.0 | 0.2 | 0.5 | 0.6 |
| Spain | 16.4 | 0.9 | 1.8 | 2.2 | 1.7 | 0.1 | 0.1 | 0.3 | 0.6 |
| Mexico | 17.4 | 1.3 | 8.2 | 2.8 | 3.7 | 0.1 | 0.1 | 0.4 | 2.2 |
| Brazil | 20.9 | 1.2 | 1.6 | 7.5 | 1.2 | 0.0 | 0.1 | 0.2 | 2.1 |
| Japan | 8.7 | 0.1 | 0.1 | 0.1 | 0.1 | 0.0 | 0.0 | 0.3 | 1.8 |
YouTube and Reddit share by search intent
Percent of citations, all countries, plus the share of searches with an AI Overview.
| YouTube | AIO rate | ||
|---|---|---|---|
| How-to | 27 | 1.9 | 92 |
| Comparison (X vs Y) | 17 | 5.3 | 93 |
| Why | 17 | 6.4 | 94 |
| Definition (what is…) | 15 | 1.6 | 94 |
| Lists (types, signs…) | 16 | 1.6 | 93 |
| Yes / no | 12 | 5.1 | 94 |
| Best / which | 13 | 2.2 | 89 |
| Topic keywords | 12 | 1.0 | 85 |
| Cost & price | 7.8 | 2.1 | 92 |
What it means: Video belongs in every market’s plan. The community platform to show up on depends on the country: Reddit in English markets, Facebook and Instagram in Latin America.
Do AI Overviews cite the same sources in every country?
Short answer: Mostly local. 81–95% of cited pages are in the country's own language, and countries share few sources even for the same question.
In every market, most cited pages are in the local language: from 81% to 95%. The most-cited domains are local institutions, banks and comparison sites rather than global brands.
Even for identical questions, markets share few sources. Same-language pairs overlap most (Spain–Mexico 0.30, US–India 0.28, US–UK 0.23). Japan shares almost nothing with any other market (about 0.03).
Citations in the country’s own language
Share of cited pages whose title is in the country’s language (detected from titles).
Most-cited domains per country
Top 6 by citations, leaving out video, social and forum platforms (covered above).
| Country | Most-cited domains (share of citations) |
|---|---|
| United States | nerdwallet.com 1.5% · investopedia.com 1.3% · fidelity.com 0.9% · experian.com 0.8% · finance.yahoo.com 0.8% · bankrate.com 0.7% |
| United Kingdom | gov.uk 2.2% · moneysavingexpert.com 1.8% · moneysupermarket.com 0.9% · investopedia.com 0.7% · moneyhelper.org.uk 0.6% · en.wikipedia.org 0.5% |
| India | bajajfinserv.in 1.2% · investopedia.com 1.0% · policybazaar.com 1.0% · groww.in 0.8% · en.wikipedia.org 0.7% · paisabazaar.com 0.6% |
| Australia | canstar.com.au 1.6% · finder.com.au 1.0% · moneysmart.gov.au 1.0% · ato.gov.au 0.8% · commbank.com.au 0.8% · investopedia.com 0.7% |
| Germany | finanztip.de 1.7% · sparkasse.de 1.4% · adac.de 1.1% · check24.de 1.0% · verivox.de 0.8% · verbraucherzentrale.de 0.8% |
| France | service-public.gouv.fr 2.3% · economie.gouv.fr 0.9% · meilleurtaux.com 0.9% · finance-heros.fr 0.8% · impots.gouv.fr 0.6% · younited.com 0.5% |
| Spain | bbva.es 1.2% · sede.agenciatributaria.gob.es 1.0% · rankia.com 0.9% · bbva.com 0.7% · idealista.com 0.7% · novatostradingclub.com 0.6% |
| Mexico | bbva.mx 2.3% · gob.mx 1.8% · mercadopago.com.mx 1.0% · condusef.gob.mx 0.8% · bitso.com 0.7% · mercadolibre.com.mx 0.5% |
| Brazil | gov.br 1.6% · serasa.com.br 1.3% · caixa.gov.br 1.1% · mercadopago.com.br 1.0% · jusbrasil.com.br 0.9% · mercadobitcoin.com.br 0.6% |
| Japan | note.com 1.8% · soico.jp 1.0% · diamond.jp 0.8% · my-best.com 0.7% · kabutan.jp 0.6% · biz.moneyforward.com 0.6% |
How much two countries agree on sources for the same question
Average overlap (Jaccard) of cited domains for the 170 shared core questions. 1.00 = identical source lists.
| US | UK | IN | AU | DE | FR | ES | MX | BR | JP | |
|---|---|---|---|---|---|---|---|---|---|---|
| United States | – | 0.23 | 0.28 | 0.21 | 0.06 | 0.06 | 0.08 | 0.09 | 0.08 | 0.04 |
| United Kingdom | 0.23 | – | 0.25 | 0.23 | 0.06 | 0.05 | 0.08 | 0.08 | 0.07 | 0.04 |
| India | 0.28 | 0.25 | – | 0.21 | 0.06 | 0.06 | 0.08 | 0.09 | 0.08 | 0.04 |
| Australia | 0.21 | 0.23 | 0.21 | – | 0.07 | 0.05 | 0.08 | 0.09 | 0.07 | 0.04 |
| Germany | 0.06 | 0.06 | 0.06 | 0.07 | – | 0.06 | 0.08 | 0.07 | 0.07 | 0.03 |
| France | 0.06 | 0.05 | 0.06 | 0.05 | 0.06 | – | 0.08 | 0.08 | 0.08 | 0.03 |
| Spain | 0.08 | 0.08 | 0.08 | 0.08 | 0.08 | 0.08 | – | 0.30 | 0.09 | 0.04 |
| Mexico | 0.09 | 0.08 | 0.09 | 0.09 | 0.07 | 0.08 | 0.30 | – | 0.10 | 0.04 |
| Brazil | 0.08 | 0.07 | 0.08 | 0.07 | 0.07 | 0.08 | 0.09 | 0.10 | – | 0.05 |
| Japan | 0.04 | 0.04 | 0.04 | 0.04 | 0.03 | 0.03 | 0.04 | 0.04 | 0.05 | – |
What it means: AI visibility is won market by market. A strong US page rarely carries over to the UK, let alone to Germany or Japan.
What makes a page get cited, and which passage gets quoted?
Short answer: Pages that match the meaning of the question and carry a recent date. Formatting tricks such as FAQ schema or TL;DR boxes barely register.
For the 3,066 seed searches with an AI Overview we compared 35,401 ranking pages in six languages, each split into sections and embedded with a multilingual model. Ranking position still dominated: in a regression its weight was about 8 times that of the next strongest trait.
Two content traits held in every market. Pages whose best section matched the meaning of the query were cited more (AUC 0.555; significant in 10 of 10 markets), and so were pages with a newer date (AUC 0.550; 10 of 10 markets). Keyword matching mattered less (AUC 0.535).
Formatting traits showed only small associations: FAQ schema 0.517, question headings 0.515, TL;DR blocks 0.509, very short paragraphs 0.506. With 35,000 pages these are measurable, but each is a fraction of the meaning and freshness effects and far smaller than ranking.
Inside cited pages, the 2,602 quoted passages were usually the section that best matched the query (39% vs 16% by chance) and sat early on the page (52% in the first quarter vs 32%). They ran a median of 177 tokens, about 130 English words, and only slightly more often sat under question-style headings (25% vs 23%).
Page traits: cited vs non-cited pages, by country
AUC within each search (0.50 = no difference; above 0.50 = cited pages score higher). Bold = the 95% interval excludes 0.50.
| Meaning match (best section) | Keyword match (BM25) | Newer date | Question headings | FAQ schema | TL;DR block | Very short paragraphs | |
|---|---|---|---|---|---|---|---|
| All countries | 0.56 | 0.53 | 0.55 | 0.51 | 0.52 | 0.51 | 0.51 |
| United States | 0.54 | 0.52 | 0.56 | 0.51 | 0.50 | 0.52 | 0.52 |
| United Kingdom | 0.54 | 0.53 | 0.56 | 0.51 | 0.53 | 0.51 | 0.51 |
| India | 0.53 | 0.50 | 0.54 | 0.51 | 0.52 | 0.50 | 0.52 |
| Australia | 0.55 | 0.54 | 0.53 | 0.52 | 0.51 | 0.51 | 0.49 |
| Germany | 0.57 | 0.55 | 0.54 | 0.51 | 0.52 | 0.50 | 0.52 |
| France | 0.57 | 0.55 | 0.56 | 0.53 | 0.52 | 0.51 | 0.49 |
| Spain | 0.53 | 0.51 | 0.54 | 0.51 | 0.52 | 0.50 | 0.52 |
| Mexico | 0.53 | 0.52 | 0.54 | 0.51 | 0.52 | 0.51 | 0.53 |
| Brazil | 0.54 | 0.52 | 0.55 | 0.50 | 0.52 | 0.50 | 0.50 |
| Japan | 0.60 | 0.57 | 0.57 | 0.53 | 0.52 | 0.52 | 0.49 |
The quoted passage vs a random section of the same page
AI Overview sentences matched to their source passage with a multilingual embedding model.
| Quoted section | Random section | |
|---|---|---|
| Is the page’s closest match to the query | 39% | 16% |
| Is in the first quarter of the page | 52% | 32% |
| Sits under a question-style heading | 25% | 23% |
| Page holds the best-matching passage of all candidates | 20% | 8% |
What it means: The pattern is the same in every language: rank, answer the question early in a section that matches its meaning, and keep the page current. Markup and formatting help a little at most.
How to get cited in AI Overviews: seven lessons from the data
- Rank for the question and its follow-ups. Ranking is still the strongest predictor of being cited. AI Overviews also pull in pages that rank for related sub-questions, so cover the follow-ups too. 72% of #1 results cited vs 6% at #16–20 · 92% of AI Overviews cite a page outside the top 20
- Answer early, in a section that matches the question. The quoted passage is usually the section that best matches the query, near the top of the page, and about 130 words long. Best-matching section quoted 39% vs 16% by chance · first quarter 52% vs 32% · median 177 tokens
- Keep pages current, and show the date. Pages with newer dates were cited more often in every country we measured. Freshness AUC 0.550 · significant in 10 of 10 countries
- Win the source type your industry trusts. Software answers cite vendors; finance answers cite banks, insurers and brokers; health and legal answers cite institutions; betting and insurance answers lean on comparison sites. Get mentioned where your industry's answers come from. Vendor sites 43% of SaaS citations · banks & lenders 49% in loans · review sites 21% in sports betting
- Publish video, and show up on the local community platform. YouTube is the single biggest source in every country. Reddit matters in English-speaking countries; Facebook and Instagram in Mexico and Brazil. YouTube 14% of citations, in 55% of AI Overviews · Reddit 5.9% in the US · Facebook + Instagram 11% in Mexico
- Localize country by country. AI Overviews cite local, same-language sources, and the same question gets different sources in each country. A strong US page rarely carries over. 81–95% local-language citations · US–UK source overlap 0.23
- Don't count on markup tricks. FAQ schema, TL;DR boxes and question-style headings show only small associations with being cited. Use them for readers, not as a ranking lever. FAQ schema AUC 0.517 · TL;DR 0.509 · question headings 0.515
AI Overview sources by industry: 17 profiles
Online casino2,000 searches
Most cited: note.com, en.wikipedia.org, pib.gov.in, psychiatry.org, takarakujichance.jp
Leading source types: publishers & reference (20%) and government & institutions (19%).
Sports betting2,000 searches
Most cited: goal.com, wettbasis.com, en.wikipedia.org, foxsports.com, sportytrader.com
Leading source types: publishers & reference (26%) and review & comparison sites (21%).
Home improvement2,000 searches
Most cited: thespruce.com, en.wikipedia.org, leroymerlin.es, nobroker.in, homedepot.com
Leading source types: brands & retailers (29%) and video (21%).
E-commerce & products2,000 searches
Most cited: shopify.com, amazon.es, nuvemshop.com.br, mercadopago.com.mx, rtings.com
Leading source types: brands & retailers (27%) and video (18%).
Crypto2,000 searches
Most cited: binance.com, finance.yahoo.com, coinbase.com, investopedia.com, kraken.com
Leading source types: banks, insurers & brokers (31%) and publishers & reference (21%).
Health2,000 searches
Most cited: mayoclinic.org, healthline.com, goodrx.com, tuasaude.com, medlineplus.gov
Leading source types: government & institutions (29%) and video (13%).
Travel2,000 searches
Most cited: tsa.gov, travel.state.gov, note.com, bahn.de, gov.uk
Leading source types: brands & retailers (20%) and social & forums (17%).
Automotive2,000 searches
Most cited: adac.de, cardekho.com, rac.co.uk, carwow.co.uk, autohero.com
Leading source types: brands & retailers (26%) and video (19%).
Marketing & SEO2,000 searches
Most cited: mailchimp.com, asa.org.uk, brevo.com, coursera.org, zapier.com
Leading source types: brands & retailers (32%) and video (18%).
Education & careers2,000 searches
Most cited: indeed.com, querobolsa.com.br, coursera.org, gov.uk, canva.com
Leading source types: government & institutions (27%) and video (16%).
SaaS & B2B software2,000 searches
Most cited: microsoft.com, salesforce.com, zoho.com, xero.com, forbes.com
Leading source types: brands & retailers (43%) and video (16%).
Legal2,000 searches
Most cited: gov.uk, service-public.gouv.fr, jusbrasil.com.br, gob.mx, citizensadvice.org.uk
Leading source types: brands & retailers (26%) and government & institutions (20%).
Personal finance2,000 searches
Most cited: investopedia.com, fidelity.com, raisin.com, moneysavingexpert.com, finance.yahoo.com
Leading source types: banks, insurers & brokers (33%) and publishers & reference (17%).
Real estate2,000 searches
Most cited: caixa.gov.br, drklein.de, rocketmortgage.com, idealista.com, hoa.org.uk
Leading source types: brands & retailers (22%) and banks, insurers & brokers (18%).
Trading & forex2,000 searches
Most cited: investopedia.com, ig.com, novatostradingclub.com, xtb.com, cmcmarkets.com
Leading source types: banks, insurers & brokers (32%) and video (21%).
Insurance2,000 searches
Most cited: policybazaar.com, canstar.com.au, bbva.mx, moneysavingexpert.com, moneysupermarket.com
Leading source types: banks, insurers & brokers (35%) and review & comparison sites (20%).
Loans2,000 searches
Most cited: experian.com, serasa.com.br, younited.com, bankrate.com, bbva.mx
Leading source types: banks, insurers & brokers (49%) and review & comparison sites (14%).
How we ran the study
Seeds: 20 searches per industry and market, half shared wording across markets (translated where needed), half written locally. Expansion: each cell grown to 200 searches with Google’s own People-also-ask questions and related searches.
Google results were collected through a search-results API (desktop, local Google domain and language), with AI Overviews and positions 1–20 for the 3,400 seed searches. A repeat run of 1,020 searches showed the same AI Overview presence 98% of the time and kept 83% of cited URLs.
Source types: rules for clear-cut platforms and public bodies, plus a model-assisted classification of the 11,911 most-cited domains (87% of citations).
Content analysis: the 3,066 seed searches with an AI Overview, their cited pages and organic results 1–20; pages split at headings into sections of up to 500 tokens and embedded with intfloat/multilingual-e5-small (first 256 tokens per section, heading path included). AUC compares cited and non-cited pages within the same search; confidence intervals come from bootstrapping searches.
Limitations
- One snapshot. Desktop results from one day; AI Overviews change between runs (0.74 average overlap of cited URLs in a repeat run).
- Partial AI Overview text. The API returns paragraphs but not bullet lists, so passage matching covers attributed paragraphs only.
- Query mix. Expanded searches come from Google’s own suggestions, which favour popular phrasings; a third are bare topic keywords.
- Source types. 13% of citations come from long-tail domains that were not classified.
- Correlation, not causation. We observe which pages are cited; we cannot see Google’s ranking signals or click data.
Data collected 14 October 2026 from live Google search results. Domain names are shown for analysis; all trademarks belong to their owners.