---
title: "AI citation sources: which sites do AI engines cite? 2,857 citations from 7 engines in 10 countries"
description: "AI citation sources: 2,857 citations from ChatGPT, Gemini, Perplexity, Copilot, Claude and Google AI in 10 countries, coded by site type. Data, method and CSV."
url: "https://theinternetninja.com/recon/ai-citation-sources-2026/"
updated: "2026-10-02"
---

# AI citation sources: which sites do AI engines cite? 2,857 citations from 7 engines in 10 countries

> TIN asked ChatGPT, Perplexity, Gemini, Copilot, Claude, Google AI Overviews and AI Mode the same 60 questions in 10 countries and coded all 1,395 cited sites into 10 categories. ChatGPT cites government and associations (68%), YouTube is the most-cited site in Google's AI, Gemini cites company sites (58%), and 78% of sites are cited by one engine only.

- Study type: Recon measurement by The Internet Ninja. TIN is the instrument; this is a measurement, not an adjudication of any claim.
- Edition: 2026-10-02
- n = 1395
- Population: Every domain cited in Ahrefs Brand Radar answers to TIN's 600 custom prompts (60 questions x 10 countries) on 7 engines, first complete weekly run, read 2026-10-02 with up to 1,000 domains per engine and country. (from 70 Brand Radar cited-domains reads (recon/ai-citation-sources-2026/raw/2026-10-02/), 8 of them empty; each domain coded into one of 10 categories (categories/domains.csv).)
- Pre-registered: method file `recon/ai-citation-sources-2026/method.md` committed at [`bfa0a7ee2457`](https://github.com/ManeeshBehera/plan-2k27-1M/commit/bfa0a7ee2457c52e73fe4769036ecb187188fa4b) before collection
- Dataset: [data.csv](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv), 1395 rows, one per entity (CC BY 4.0)
- Canonical URL: https://theinternetninja.com/recon/ai-citation-sources-2026/

## Key findings

| # | Finding | Value | n |
| --- | --- | --- | --- |
| 1 | ChatGPT sends 68% of its citations to government and associations; Gemini sends 58% to company sites | 68% | 1395 |
| 2 | Seven engines, seven source profiles across ten categories of site | 7 profiles | 1395 |
| 3 | 1,091 of 1,395 cited sites (78%) appear in only one engine's answers | 78% | 1395 |
| 4 | YouTube is the most-cited site in Google AI Overviews and AI Mode; ChatGPT cites no social or video platform | 0% | 1395 |
| 5 | Company sites are the largest category overall, with 37% of all citations | 37% | 1395 |
| 6 | Questions asked from Japan get 84% of citations from local sites; from India, 8% | 84% vs 8% | 1395 |
| 7 | Japan has the highest or second-highest local share for six of the seven engines | 6 of 7 | 1395 |
| 8 | 57 of the most-cited international sites drew 44 citations across the six non-English markets | 44 | 57 |

We asked ChatGPT, Perplexity, Gemini, Copilot, Claude, Google AI Overviews and Google AI Mode the same 60 everyday questions in ten countries, recorded every site their answers cited, and sorted all 1,395 sites into ten kinds of source [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv). Each engine has its own diet, and they barely overlap.

This is edition 1, built on Ahrefs' first complete weekly run, read on 2026-10-02: 2,857 citations [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv). Claude returned 47 of them and none for Germany, Italy, Sweden or Japan; Google AI Overviews returned none for France, Italy and Sweden, and AI Mode none for France. Those eight cells are left empty, never filled. The second weekly run will be added to this page as a dated update.

## 1. What kind of source does each AI engine cite?

Each engine trusts a different kind of source. ChatGPT leans on government and associations (68% combined), Gemini on company sites (58%), and Google's two AI surfaces cite YouTube more than any other site [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv). Copilot sends 28% of its citations to small single-topic sites, more than any other engine, and Claude sends 40% to comparison publishers, on few citations [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv).

Company sites are the largest single category overall (37% of all citations), led by Gemini [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv). Engines often quote the bank, insurer or software vendor directly rather than a third party describing it.

Social and video is a Google thing. YouTube is the most-cited site in both AI Overviews and AI Mode, where social and video platforms take 14% and 11% of citations; ChatGPT cites them 0% of the time [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv).

| Engine | Citations | Largest category | Most-cited sites |
|---|---|---|---|
| ChatGPT | 299 | Government (50.5%) | consumer.ftc.gov (8), content.naic.org (7), pmc.ncbi.nlm.nih.gov (7), weforum.org (7), consumerfinance.gov (6) |
| Perplexity | 1,099 | Company sites (38.9%) | forbes.com (13), wikihow.com (13), shopify.com (11), britannica.com (10), nerdwallet.com (10) |
| Gemini | 218 | Company sites (58.3%) | reddit.com (6), indeed.com (5), boostcamp.app (4), britannica.com (4), darebee.com (4) |
| Copilot | 523 | Company sites (33.3%) | scienceinsights.org (11), bankrate.com (5), britannica.com (5), ebsco.com (5), en.wikipedia.org (5) |
| Claude | 47 | Comparison (40.4%) | moneygeek.com (5), forbes.com (4), insurify.com (4), nerdwallet.com (4), spotpet.com (4) |
| AI Overviews | 184 | Company sites (39.1%) | youtube.com (17), en.wikipedia.org (8), britannica.com (3), heartfoundation.org.nz (3), isappscience.org (3) |
| AI Mode | 487 | Company sites (38.6%) | youtube.com (28), reddit.com (18), nerdwallet.com (7), canstar.com.au (5), en.wikipedia.org (5) |

## 2. The seven source profiles on one chart

Each shape is one engine; each spoke is a category, scaled from 0 to the largest share on the chart. Hover or click an engine, in the legend or on the chart, to highlight it; hover or tap a point to read its value. The table under the chart holds the exact values [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv).

## 3. Do AI engines cite the same sites?

The engines barely agree on which sites to cite. 1,091 of 1,395 sites (78%) appear in only one engine's answers, and one site, forbes.com, appears in six [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv). The closest pair, AI Overviews and AI Mode, share 8 of their 25 most-cited sites; ChatGPT shares at most 1 with any other engine [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv).

| Engines citing the site | Sites | Share |
|---|---|---|
| 1 | 1,091 | 78.2% |
| 2 | 194 | 13.9% |
| 3 | 72 | 5.2% |
| 4 | 28 | 2.0% |
| 5 | 9 | 0.6% |
| 6 | 1 | 0.1% |

## 4. How concentrated is each engine?

Some engines spread citations wide, others concentrate. Copilot cited 399 different sites and its top 10 hold 9.9% of its citations; for AI Overviews the top 10 hold 26.1% [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv).

| Engine | Different sites cited | Top 10 sites' share of citations |
|---|---|---|
| ChatGPT | 174 | 20.7% |
| Perplexity | 634 | 8.6% |
| Gemini | 164 | 18.3% |
| Copilot | 399 | 9.9% |
| Claude | 24 | 70.2% |
| AI Overviews | 136 | 26.1% |
| AI Mode | 327 | 17.2% |

## 5. Does the country change which sources AI cites?

The same 60 questions were asked everywhere, in English for the US, UK, Australia and India and translated for the other six. The country changes how local the answer is. Japan (84%), Sweden (79%), Italy (70%) and Germany (65%) mostly get sites serving their own country; France (32%) and Spain (25%) do not [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv). Nor do the English-speaking markets outside the US: the UK 30%, Australia 29%, India 8% [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv).

| Country | Citations | Local share | Most-cited sites |
|---|---|---|---|
| United States | 581 | 64.0% | nerdwallet.com (13), youtube.com (11), forbes.com (10) |
| United Kingdom | 497 | 30.0% | youtube.com (12), en.wikipedia.org (7), gov.uk (7) |
| Australia | 412 | 29.1% | reddit.com (8), canstar.com.au (7), youtube.com (7) |
| India | 511 | 8.2% | youtube.com (9), britannica.com (8), forbes.com (7) |
| France | 89 | 31.5% | unesco.org (4), edutechwiki.unige.ch (3), coe.int (2) |
| Germany | 228 | 64.9% | de.wikipedia.org (6), familie-und-tipps.de (5), mystipendium.de (5) |
| Spain | 126 | 24.6% | es.wikipedia.org (4), cedia.org (3), educagob.educacionfpydeportes.gob.es (3) |
| Italy | 154 | 70.1% | it.wikipedia.org (7), assicurazioni.segugio.it (5), wise.com (4) |
| Sweden | 126 | 79.4% | sv.wikipedia.org (7), arbetsformedlingen.se (3), rodakorset.se (3) |
| Japan | 133 | 83.5% | smbctb.co.jp (5), sompo-japan.co.jp (4), americanexpress.com (3) |

## 6. Which engines cite local websites?

Japan has the highest or second-highest local share for six of the seven engines; Claude returned no citations for Japan [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv). India is the outlier: across engines 8% of citations go to Indian sites, so a question asked from India is mostly answered from sites serving other countries [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv).

## 7. Is it just where each platform's users are?

A site might be cited less in a country simply because few people there use it. To test that, TIN measured the search audience by country of 57 of the most-cited international sites (Ahrefs organic search traffic, a proxy for users) and set it against where each site is cited [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv).

In the English-speaking markets audience goes with citations, but only partly: a site with a bigger local audience tends to be cited more there (rank correlation 0.42 in the US, 0.46 in the UK, 0.53 in Australia, 0.35 in India) [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv). In the non-English markets the international sites are barely cited at all, 44 citations across six countries, even where they have a real audience [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv). Questions asked in French, German, Spanish, Italian, Swedish and Japanese rarely bring these international platforms into the answer, whatever their reach there.

YouTube gets 20.37% of its search audience from India, yet India accounts for 9 of its 49 citations; the UK, with 2.79%, accounts for 12 [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv). India shows the opposite pull for some sites. NerdWallet gets 0.71% of its search audience from India, yet it is cited 7 times for questions asked from India, more than in the UK and Australia combined [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv). NerdWallet, a US personal-finance site, is cited for Indian money questions with almost no Indian audience behind it.

| Country | Citations to the measured sites | Audience vs citations, rank correlation |
|---|---|---|
| United States | 184 | 0.42 |
| United Kingdom | 139 | 0.46 |
| Australia | 108 | 0.53 |
| India | 161 | 0.35 |
| France | 7 | 0.03 |
| Germany | 11 | 0.22 |
| Spain | 9 | 0.21 |
| Italy | 11 | 0.31 |
| Sweden | 3 | -0.19 |
| Japan | 3 | 0.30 |

Search audience and citations, site by site, for the ten most-cited measured sites. The first tab is the share of each site's search audience in each country; the second is the citations it received for questions asked from that country. A dash means the country is outside the site's top 40 audience countries.

| Site | US | GB | AU | IN | FR | DE | ES | IT | SE | JP |
|---|---|---|---|---|---|---|---|---|---|---|
| youtube.com | 22.47% | 2.79% | 0.89% | 20.37% | 2.16% | 2.17% | 1.53% | 1.53% | – | 3.65% |
| reddit.com | 53.51% | 4.90% | 1.90% | 6.26% | 1.96% | 2.00% | 1.05% | 1.31% | 0.38% | 1.80% |
| forbes.com | 55.50% | 9.98% | 2.30% | 8.36% | 0.41% | 0.77% | 0.42% | 0.52% | 0.16% | – |
| britannica.com | 47.64% | 6.49% | 2.76% | 16.75% | 0.13% | 0.27% | 0.11% | 0.18% | 0.19% | – |
| nerdwallet.com | 84.66% | 0.64% | 0.11% | 0.71% | 0.50% | 0.27% | 0.43% | 0.31% | 0.06% | 0.09% |
| en.wikipedia.org | 42.98% | 10.83% | 3.61% | 13.75% | 0.38% | 0.77% | 0.23% | 0.49% | 0.40% | – |
| wise.com | 7.13% | 2.37% | 1.60% | 6.07% | 0.86% | 2.34% | 1.72% | 0.87% | – | 2.67% |
| investopedia.com | 42.50% | 4.61% | 1.73% | 17.77% | 0.44% | 0.80% | 0.64% | 0.44% | – | 0.23% |
| wikihow.com | 36.79% | 4.65% | 1.72% | 13.00% | 1.48% | 1.40% | 0.72% | 0.25% | – | – |
| isappscience.org | 35.71% | 2.87% | 0.91% | 6.66% | 2.01% | 0.21% | 4.20% | 2.38% | – | – |

| Site | US | GB | AU | IN | FR | DE | ES | IT | SE | JP |
|---|---|---|---|---|---|---|---|---|---|---|
| youtube.com | 11 | 12 | 7 | 9 | 0 | 5 | 2 | 1 | 0 | 2 |
| reddit.com | 10 | 7 | 8 | 4 | 0 | 5 | 0 | 0 | 0 | 0 |
| forbes.com | 10 | 4 | 4 | 7 | 0 | 0 | 0 | 1 | 0 | 0 |
| britannica.com | 6 | 4 | 5 | 8 | 1 | 0 | 0 | 0 | 0 | 0 |
| nerdwallet.com | 13 | 3 | 1 | 7 | 0 | 0 | 0 | 0 | 0 | 0 |
| en.wikipedia.org | 6 | 7 | 3 | 5 | 0 | 0 | 0 | 1 | 0 | 0 |
| wise.com | 2 | 4 | 4 | 3 | 0 | 0 | 0 | 4 | 1 | 0 |
| investopedia.com | 3 | 5 | 2 | 6 | 0 | 0 | 0 | 0 | 0 | 0 |
| wikihow.com | 4 | 5 | 4 | 3 | 0 | 0 | 0 | 0 | 0 | 0 |
| isappscience.org | 3 | 4 | 4 | 4 | 0 | 0 | 0 | 0 | 0 | 0 |

## 8. The most-cited sites across all engines

| Site | Category | Citations | Engines citing it |
|---|---|---|---|
| youtube.com | Social and video | 49 | 3 of 7 |
| reddit.com | Social and video | 34 | 4 of 7 |
| forbes.com | News | 26 | 6 of 7 |
| britannica.com | Reference | 24 | 5 of 7 |
| nerdwallet.com | Comparison | 24 | 5 of 7 |
| en.wikipedia.org | Reference | 22 | 4 of 7 |
| wise.com | Company sites | 18 | 5 of 7 |
| investopedia.com | Reference | 16 | 4 of 7 |
| wikihow.com | Reference | 16 | 2 of 7 |
| isappscience.org | Associations | 15 | 5 of 7 |
| who.int | Government | 15 | 5 of 7 |
| pawlicy.com | Comparison | 14 | 4 of 7 |
| coursera.org | Company sites | 13 | 4 of 7 |
| shopify.com | Company sites | 13 | 2 of 7 |
| unesco.org | Government | 13 | 4 of 7 |

## What this means if you want AI engines to cite you

These are associations in one snapshot, not causes. They show where each engine's answers point today, and so where presence matters.

- For ChatGPT, be cited by the institutions in your field. With 68% of its citations going to public bodies and associations, a business more often reaches ChatGPT's answers through the regulator, trade body or standards group that covers it than through its own blog [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv).
- For Google's AI answers, publish where people watch and discuss. YouTube is the single most-cited site in both AI Overviews and AI Mode, and Reddit is second in AI Mode [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv).
- For Gemini and Perplexity, your own site can be the source. Company sites are their largest category (58% and 39%), so a company's own pages are the kind of source they quote most [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv).
- Comparison sites stand between brands and buyers, especially in Claude (40% of its citations), Gemini (13%) and Copilot (11%) [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv). Being listed fairly on them is part of AI visibility.
- Outside English, translate and localise. In Japan, Sweden, Italy and Germany most citations go to sites serving that country.
- Do not optimise for "AI" as one channel. With 78% of cited sites unique to one engine, a source strategy that works for ChatGPT says little about Gemini or Google [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv).

Related on TIN: [what controlled experiments show about generative engine optimization](https://theinternetninja.com/blog/generative-engine-optimization-what-the-controlled-evidence-shows-2026/), [what the retrieval record shows about AI search optimization](https://theinternetninja.com/blog/ai-search-optimization-what-the-retrieval-record-shows-2026/), and [which AI and automation agencies AI engines cite](https://theinternetninja.com/recon/agency-ai-citations-2026/).

## How this was measured

1. **Questions.** 60 real searched questions: the 4 highest-volume questions in each of 15 everyday topics (health, money, travel, food, technology, career, education, home, car, shopping, law, fitness, pets, parenting, business) from Ahrefs' US database, after removing brand, person, one-country and date-dependent questions by a rule fixed before selection.
2. **Prompts.** The same 60 in each of 10 countries: in English for the US, UK, Australia and India, translated for France, Germany, Spain, Italy, Sweden and Japan. 600 prompts.
3. **Engines.** ChatGPT, Perplexity, Gemini, Microsoft Copilot, Claude, Google AI Overviews and Google AI Mode, run by Ahrefs Brand Radar as custom prompts. Every cited domain read per engine and country, up to 1,000, on 2026-10-02.
4. **Categories.** Every one of the 1,395 cited domains was coded into one of 10 categories, and the country it mainly serves, by two coders working blind to each other and to the citation counts. They agreed on 88.5% of categories and 84.1% of markets; a third coder chose between their two answers for the rest [source](https://theinternetninja.com/recon/ai-citation-sources-2026/data.csv). Wikipedia, government and university domains are coded by rule. 1,187 domains were coded on a first look at the run; the 208 that only the full pull added were coded the same way before any figure was computed.
5. **Audience check.** For 60 of the most-cited sites serving an international or foreign audience (ranked on the first 60 domains per engine and country, before the full pull; amendment 2), Ahrefs organic search traffic by country (top 40 countries) and in total, read 2026-10-01 (28,620 units). Three were dropped because Ahrefs returned implausibly low totals (gov.uk, nhs.uk, scienceinsights.org), leaving 57. Correlation is Spearman's rank correlation across those sites, per country.
6. **Pre-registration.** The method, the category scheme and the audience layer were committed before the data they govern was read. The prompts, coding and data are in the TIN repository.

## Method record

Amendments, each landed before any outcome data:

- `recon/ai-citation-sources-2026/amendment-1.md` at [`ad12d67207cc`](https://github.com/ManeeshBehera/plan-2k27-1M/commit/ad12d67207cc5cec063341e37965aa398ffb5888)
- `recon/ai-citation-sources-2026/amendment-2.md` at [`0247e46db734`](https://github.com/ManeeshBehera/plan-2k27-1M/commit/0247e46db734b8d94d3a1df61b1a3e2766a0d8e0)
- `recon/ai-citation-sources-2026/amendment-3.md` at [`3e65a8e13eab`](https://github.com/ManeeshBehera/plan-2k27-1M/commit/3e65a8e13eabee29342fec35aa7314723e6a4942)

| Metric | Unit | Tool | Computation |
| --- | --- | --- | --- |
| Citations | count | Ahrefs brand-radar-cited-domains, report 01a0f680, prompts=custom, select=domain,responses,pages, limit=1000 | Answers in the cell that cited the domain (Ahrefs `responses`), summed over the engines and countries named by the column. |
| Category share | % | recon/ai-citation-sources-2026/analyze.py | A category's citations divided by all citations in the engine, country or cell (amendment 1, C5). |
| Local share | % | recon/ai-citation-sources-2026/analyze.py | Citations to domains that mainly serve the country the question was filed under (coded market, or that country's Wikipedia language edition), divided by all citations in the cell (amendment 1, C4). |
| Top-25 overlap | sites of 25 | recon/ai-citation-sources-2026/analyze.py | How many of one engine's 25 most-cited domains are also in another engine's 25 most-cited, all countries combined; ties broken by domain name. |
| Top-10 concentration | % | recon/ai-citation-sources-2026/analyze.py | The 10 most-cited domains' share of an engine's citations. |
| Audience vs citations, rank correlation | Spearman rho | recon/ai-citation-sources-2026/analyze.py, Ahrefs site-explorer-metrics-by-country (Ahrefs estimate of organic search traffic) | Per country, Spearman's rank correlation between each measured site's share of its search audience in that country (an Ahrefs estimate; 0 outside its top 40 countries) and its citations there (amendment 2). |

Design: correlational. The findings report what the measured entities have in common, never that a trait produced an outcome.

## What this study cannot say

- 60 questions are a sample of everyday searches, not of everything people ask AI. Business-to-business questions are not covered.
- Edition 1 is one weekly run (amendment 3), a snapshot, not a trend. Claude returned 47 citations in total and none for Germany, Italy, Sweden or Japan; Google AI Overviews returned none for France, Italy and Sweden; Google AI Mode none for France. Those cells are empty, never filled.
- Categories are a judgment made by a fixed rule from the domain name, by two blind coders and a reconciler; a different scheme would group some sites differently. The domain-level data is published so readers can regroup it. The coders saw the domain only, not the home page title amendment 1 describes.
- The category scheme was set after a partial look at the run's top domains, before the pre-registered pull.
- Search traffic stands in for users: it misses app and logged-in use, which matter most for YouTube and Reddit, and Ahrefs' country split is modelled. Only 57 of 1,395 sites are measured.
- The 60 audience sites were ranked on the first 60 domains per engine and country, before the full pull; ranked on the full lists, the set would differ somewhat.
- "Local" means a site mainly serving the country, as coded; Ahrefs files each prompt under a country but the engines may not localise every answer.

## Questions this study answers

**Which websites does ChatGPT cite most?**

In TIN's study ChatGPT sent 51% of its 299 citations to government and intergovernmental sites and 17% to associations and nonprofits, such as consumer.ftc.gov, the NAIC, PubMed Central and the World Economic Forum. It cited no social or video platform at all.

**Does Google AI Overviews cite YouTube and Reddit?**

YouTube, yes: it is the most-cited site in both Google AI Overviews and AI Mode in this study. Reddit is second in AI Mode (18 citations) but cited only 3 times in AI Overviews. Social and video platforms take 14% of AI Overviews citations and 11% of AI Mode citations, against 0% in ChatGPT; company sites are the largest category in both.

**Do different AI engines cite the same sources?**

Rarely. Of 1,395 cited sites, 1,091 (78%) were cited by only one of the seven engines, and only one, forbes.com, by six. ChatGPT shares at most 1 of its 25 most-cited sites with any other engine.

**Do AI engines cite local websites outside the US?**

It depends on the country. In Japan 84%, Sweden 79%, Italy 70% and Germany 65% of citations go to sites serving that country; in France 32% and in Spain 25%. In India only 8% do: answers to questions asked from India mostly cite US and international sites.

**Which AI engine cites company websites most?**

Gemini. 58% of Gemini's citations go to a company's own site, the highest of any engine, against 16% for ChatGPT.

**Are international sites cited less where they have fewer users?**

Not mainly. For 57 of the most-cited international sites, TIN compared each site's search audience by country with its citations by country. In the US, UK and Australia a bigger local audience goes with more citations (rank correlation 0.42 to 0.53), and in India 0.35. In the six non-English markets these sites drew just 44 citations in all, even where they have a real audience: YouTube gets 2.17% of its search audience from Germany and 3.65% from Japan.

**How were these AI citation sources measured?**

TIN wrote 60 real searched questions across 15 everyday topics, ran them in 10 countries through Ahrefs Brand Radar on 7 engines, and counted the sites each answer cited. Two coders classified every site into one of 10 categories without seeing citation counts; they agreed on 88.5% and a third coder settled the rest. The method was published before any data was read.

## Related

- [Generative engine optimization: what the controlled evidence actually shows in 2026](https://theinternetninja.com/blog/generative-engine-optimization-what-the-controlled-evidence-shows-2026/)
- [ChatGPT ranks passages, not pages: what its retrieval stream shows, and the numbers nobody can source](https://theinternetninja.com/blog/ai-search-optimization-what-the-retrieval-record-shows-2026/)

## Editions and errata

This edition: 2026-10-02. A refresh republishes on the same URL, and every prior edition's dataset stays downloadable.

No erratum.