Asked from India, answered from America

01 / 07Summary

Buyers in India, the UK and the US ask answer engines the same questions in the same words. We wanted to know whether the answers change with the country. On 5 October 2026 we asked four engines 60 buying questions that never name a country, from India, the United Kingdom and the United States: 720 answers and 8,553 cited sources.

Asked from India, most answers leaned on American websites, and when they gave a price, it was usually in dollars.

02 / 07Why this matters

A search engine has long known where its user is and shown local shops, local prices and local news. Answer engines now do the reading for the buyer and cite a handful of pages. If those pages come from another country, a local bank, insurer or software maker is competing on someone else’s shortlist, and the buyer gets prices they cannot pay in.

Most claims about “local AI visibility” rest on a single screenshot taken from one place. We asked the same questions from three countries on the same day so that the only thing that changed was where the question came from.¹

Asked from India, answer engines cited American sites three times as often as Indian ones.

Study 008, Finding 2

03 / 07Method

We wrote 60 buying questions, 15 in each of four categories: software for small businesses, online courses, consumer products and personal finance. None names a country, city or currency, and none uses a word that belongs to one market, such as “maths”, “ISA” or “401(k)”.² Each question was asked once on each engine from each country, with the location set by the collection service.³ To see how much answers move when nothing changes but time, we asked 5 questions a second time on all four engines (20 answers).

Answers

72060 questions × 4 engines × 3 countries, plus 20 repeats

Citations

8,5535,266 unique URLs on 1,572 domains

Questions

6015 each: software, learning, consumer, finance

Engines

ChatGPT (search on), Perplexity, Gemini, Google AI Mode

Countries

India, United Kingdom, United States, set at the request level

Fieldwork

5 October 2026, 11:13–12:13 UTC; no failed answers after retries

What we counted

A citation is a URL the engine presented as a source. We counted each URL once per answer, after removing tracking parameters and fragments.⁴ A mention is a brand named in the answer text, with the reference list and links removed. A price is any amount with a currency marker (₹, Rs, rupees, lakh; $; £; €).

Where a source is from

Each cited URL was given a country: first from a country edition in the address (uk.pcmag.com, forbes.com/advisor/in/), then from a country domain (.in, .co.uk), then from a map of 878 domains to the home country of the publisher or company. Global platforms such as YouTube and Reddit were coded as international. Sites whose home we could not establish were coded unknown and kept in the denominator.⁵ The full map is published with the data.

Brands and overlap

We matched brands against a dictionary of 1,672 brands and providers built from the answers and checked by hand.⁶ For each question and engine we compared the brands in the India, UK and US answers. Overlap is the share of brands the two answers had in common out of all brands either named.⁷

Disclosure: some brands that appear in this study are clients of Citation Research. Their results were produced by the same questions, engines and rules as every other brand’s.

04 / 07Findings

1. Most sources were not local, except in the US

Of 2,811 citations in answers asked from India, 15% were Indian sites and 51% were American. Answers asked from the UK did a little better, with 24% UK sites, but still cited US sites more often (46%). Asked from the US, 62% of citations were American. In 210 of 231 India-asked answers that cited anything, fewer than half the sources were Indian; 144 cited no Indian site at all. In the US the equivalent figures were 61 and 10 of 230.

The unknown sites do not change the picture. If every one of the 428 unplaced citations in India-asked answers were Indian, the Indian share would still be 30%. Looking at pages rather than publishers gives a similar answer: 264 of the 2,811 citations (9%) were pages whose title or address named India.

The most-cited site in India-asked answers was coursera.org, with 99 citations in 28 answers. NerdWallet, an American personal-finance site, was cited in 27 India-asked answers. The most-cited Indian site was zoho.com, with 43 citations in 8 answers.

2. The engine mattered more than the country

Perplexity returned exactly the same list of sources in India, the UK and the US for all 60 questions; its answers were worded differently but drew on the same pages. Gemini cited few sources (2.7 per answer; 24 of its 180 answers cited none) and almost none of them were local outside the US. ChatGPT and Google AI Mode changed their sources with the country, and Google AI Mode localised most.

ChatGPT shows two kinds of source: the pages it cites inline and a longer list of search results. Counting only the inline citations gives the same result for India: 34 of 178 (19%) were Indian.⁴ The highest local share for India in any engine and category was Google AI Mode on personal finance, where 50% of 132 citations were Indian sites.

3. Prices came in dollars

Of 240 answers asked from India, 97 named a price. Rupees appeared in 43 of them. The other 54 priced only in a foreign currency, and every one of those was US dollars. Asked from the US, all 110 answers that named a price used dollars. Gemini never priced in rupees (0 of 27 priced answers from India) or pounds (0 of 28 from the UK). ChatGPT and Google AI Mode used rupees in 19 of 30 and 24 of 38 priced India-asked answers. Perplexity rarely named a price at all (2 of 60 answers in India).

4. The local answer often came as an aside

Answers asked from India named India in 56 of 240 cases (23%). In 21 of them, India appeared as a condition: “If you’re in India…”, “If you mean India…”. ChatGPT accounted for 19 of those 21. One ChatGPT answer asked from India opened with “Assuming you mean U.S. retirement accounts”. Answers asked from the UK framed the UK the same way 20 times. No answer asked from the US treated the US as a condition (0 of 240).

No US-asked answer treated the US as a condition. 21 India-asked answers treated India as one.

Study 008, Finding 4

5. Brands moved less than sources

For the same question on the same engine, the India and US answers had 62% of their brands in common on average (238 pairs). That is close to the overlap between UK and US answers (64%, 237 pairs), and below the 74% we saw when the same question was simply asked again in the same country (20 pairs). Google AI Mode changed brands most with the country (47% India–US overlap, 60 pairs) and Perplexity least (73%, 58 pairs).

6. Which categories localised

No category reached a fifth of Indian sources. The differences were in prices and brands. Consumer products localised prices: 28 of 31 priced India-asked answers used rupees. Finance localised brands when it localised at all: 11 of 60 India–US finance pairs shared no brand, against none of 60 in software. Software and courses changed least: 2 of 18 and 1 of 12 priced India-asked answers in those categories used rupees.

7. What the answers looked like

The same question, asked on the same day, from India and from the US. Brands are listed as the engine named them, in alphabetical order.

05 / 07What it means for brands

None of this means local pages are wasted. It means the engines do not always go looking for them.

1. Check from every market you sell in.A check run from one country tells you little about another. ChatGPT and Google AI Mode returned different sources in India and the US for all 60 questions.
2. Be on the international pages, not only the local ones.In India-asked answers, American sites were cited 3.4 times as often as Indian ones. A brand that sells in India is often judged on a US roundup.
3. Say which market a page is for.Only 9% of citations in India-asked answers were pages that named India in their title or address. Prices in rupees and a market named in the title make a page easy to place.
4. Plan for the follow-up question.ChatGPT often answered for the US and then offered to redo it for India. The local answer comes on the second turn, and the pages it cites are the ones that answer that turn.
5. Treat each engine as its own market.Perplexity returned the same sources everywhere in our test. Google AI Mode was the most local. A plan that works on one may not reach the other.

06 / 07Limitations

• One run, one day. Each question was asked once per engine and country on 5 October 2026. Our 20 repeat answers show that brand lists move by about a quarter on a second asking, so single cells are noisy. Totals across 60 questions are more stable than any one example.
• Location, not a local user. The country was set by the collection service, with no account, history or language settings. A signed-in user in India may get more local answers. Perplexity’s identical lists may reflect how the service reaches Perplexity rather than what every user in India sees.
• English only. Many Indian buyers ask in Hindi or a mix of Hindi and English. We did not test that.
• Country of a site is a judgement. We coded publishers and companies by their home country. 15% of India-asked citations went to sites we could not place; the bound in Finding 1 shows how far that could move the result.
• Brand matching is a dictionary. Our hand checks found a small error rate, which we fixed. Five answers named no brand we could match, some because product names sat in image cards that reached us without text.
• Small cells. Gemini cited 165 sources in India-asked answers in total; its percentages rest on fewer pages than ChatGPT’s 1,531.

07 / 07Data and reproducibility

The files below are published under CC BY 4.0. Version 1.0 corresponds to this publication; corrections will be issued as new versions with a changelog. The full answer texts are available on request.

• questions.csv60 rowsCSV
• answers.csv720 rowsCSV
• citations.csv8,553 rowsCSV
• domain-country-map.csv1,572 rowsCSV
• brand-overlap.csv710 rowsCSV
• brand-dictionary.csv1,672 rowsCSV

answers.csv has one row per answer with its currencies, countries named, brands and source counts. citations.csv lists every cited URL with its country code and the rule that set it. domain-country-map.csv gives each cited domain, its country and the basis. brand-overlap.csv has the overlap for each pair of countries, question and engine, and brand-dictionary.csv the brands and aliases we matched.

For questions about the data, or a reanalysis you would like us to check, write to research@citationresearch.com.

@techreport{citationresearch2026study008,
  author      = {{Citation Research}},
  title       = {Asked from India, answered from America},
  institution = {Citation Research},
  number      = {Study 008},
  year        = {2026},
  url         = {https://citationresearch.com/research/ai-answers-india-uk-us-sources}
}