Original study, October 2026

Does your authority cross the border?

The same question, asked in French, German and English from France, Germany and Switzerland. This is what Google AI Overviews, ChatGPT and Perplexity cite.

Study by Alizée Baudez, independent international SEO and GEO consultant, Strasbourg. About me

1,510 answers analysed, 12,414 links cited. Collected 25 September to 2 October 2026, logged out, with a VPN set to Paris, Frankfurt and Zurich.

Get full access Read the seven findings

Who are you? See what to watch first.

Three questions about your business. The findings and recommendations reorder themselves to fit you. The sorting happens in your browser: nothing is sent.

Where do you sell?

FranceGermanySwitzerland

In which languages does your site exist?

FrenchGermanEnglish

What do you sell?

To businesses (B2B)To consumers (B2C)

Tick at least one answer: your three points to watch appear here.

To watch first, for you

Seven findings

Four findings hold in both sectors, three are specific to one. Each gives its question, its result with its base, how to read it and its caveats.

Finding The result holds in both sectors and rests on stable citations.
Vertical-specific The result holds for one sector, cookware or B2B. It is not generalised to the other.

Row: same language, three countries. Column: same country, three languages.

FindingFor you

Language decides the sources, more than country

When the country or the language changes, do the cited sources change?

Overlap is 1 when two answers cite the same sites and 0 when they share none. On Google, two answers in the same language from two countries share 0.32 of their sites. From the same country in two languages, 0.08. On Perplexity, 0.75 against 0.05. ChatGPT is the exception: 0.39 across countries, 0.38 across languages. There, country counts as much as language.

Try it: what does your set-up share with the other eight?

Each cell is a set-up: a question language (row) asked from a country (column). The number is the share of cited sites that set-up has in common with the ringed cell. 1 means the same sites, 0 means no site in common.

Same language, other country (location effect)Same country, other language (language effect)Country and language both changeStarting cell: the same test, repeated (noise floor)

Try this: stay on Google, then compare the cell to the right of the ringed one (same language, other country) with the one below it (same country, other language). Click any cell to make it the starting point.

On Google and Perplexity, language changes the sources, country does not

Source: Alizée Baudez, Cross-Border AI Visibility Study 2026 (overlap of sources). Base: pairs of answers to the same question (n on each bar); whiskers: 95% interval, resampling the 24 questions; Perplexity: 141 tests run twice. Collected 25 September to 2 October 2026.

See the chart data

Sector	Tool	Same test, repeated	Same language, other country	Same country, other language	n (same language, other country)	n (same country, other language)
Both sectors	Google AI Overviews	0.45	0.32	0.08	818	777
Both sectors	ChatGPT	0.51	0.39	0.38	1,287	1,210
Both sectors	Perplexity	0.79	0.75	0.05	214	214
Cookware (B2C)	Google AI Overviews	0.54	0.39	0.11	299	304
Cookware (B2C)	ChatGPT	0.39	0.25	0.19	550	505
Cookware (B2C)	Perplexity	0.80	0.76	0.01	106	106
B2B (components)	Google AI Overviews	0.40	0.28	0.06	519	473
B2B (components)	ChatGPT	0.60	0.50	0.51	737	705
B2B (components)	Perplexity	0.78	0.74	0.08	108	108

Reading. On Google and Perplexity, the German version of your site is not a translation of the French one. It is a presence on another market, facing other sources. At domain level, 56% of the 603 domains Perplexity cites appear in all three countries, against 15% of the 327 domains Google cites stably.

Caveat. Bases: 818 and 777 pairs of Google answers (location, then language), 214 and 214 for Perplexity. The gap "location minus language" is positive on Google (0.199 to 0.276) and on Perplexity (0.637 to 0.763), and includes zero on ChatGPT (−0.041 to 0.079). 24 questions, two sectors, one week.

All cells, English row included: the country of the test counts even without a local language.

FindingFor you

ChatGPT assumes the country you are writing from

Do the tools localise their sources by the country of the test?

ChatGPT cites 33.2% sites from the tested country, Google 18.1%, Perplexity 18.8%. Even in English, ChatGPT stays anchored: 26.7% tested-country sites, against 6.4% and 2.1%. On three questions that do not say where you are, ChatGPT takes the country of the test for the person’s own in 52 of 81 answers, Google in 19 of 67, Perplexity in 5 of 27.

ChatGPT cites nearly twice as many sites from the country as the other two tools

Source: Alizée Baudez, Cross-Border AI Visibility Study 2026 (hypothesis 1, the most localised tool; country assumed by the answer). Base: stable citations for Google (708) and ChatGPT (328), all citations for Perplexity (1,774); whiskers: 95% interval. Assumed country: answers to the three questions where the person’s country matters (CW12, EA03, EA12). Collected 25 September to 2 October 2026.

See the chart data

Tool	Both sectors	Cookware (B2C)	B2B	English answers	Test country taken for the person’s own
Google AI Overviews	18.1%	20.5%	15.8%	6.4%	19 / 67
ChatGPT	33.2%	28.8%	36.5%	26.7%	52 / 81
Perplexity	18.8%	19.9%	17.7%	2.1%	5 / 27

Reading. A company that sells in both France and Germany cannot read a test run from a single country, even in English.

Caveat. Each country was collected on a different day: the effect of the country and the effect of the day cannot be separated. The mechanism (IP address or otherwise) is not observed, only the effect is. Bases: 328 stable ChatGPT citations, 708 Google, 1,774 Perplexity citations.

Swiss column: answers obtained from Zurich, in French and in German.

FindingFor you

From Zurich, answers cite the neighbouring countries first

From Zurich, do answers cite Swiss sites or those of the neighbouring countries?

From Zurich, in German, Perplexity cites 75.3% German domains against 5.6% Swiss ones (198 citations). In French, 59.0% French domains against 9.5%. Google points the same way: 40.5% against 15.5% in German, 32.2% against 21.1% in French. ChatGPT is closer to balance, but its base is narrow (34 stable citations).

From Zurich, Google and Perplexity cite neighbours more than Swiss sites

Source: Alizée Baudez, Cross-Border AI Visibility Study 2026 (hypothesis 3, Switzerland and its neighbours). Base: citations in answers obtained from Zurich, stable for Google and ChatGPT, all for Perplexity; n on each line. An asterisk marks a base of fewer than 30 citations. Collected 25 September to 2 October 2026.

See the chart data

Sector	Tool	Answers in French: french domains	Answers in French: swiss domains	Answers in German: german domains	Answers in German: swiss domains	n
Both sectors	Google AI Overviews	32.2%	21.1%	40.5%	15.5%	90 / 84
Both sectors	ChatGPT	29.4%	26.5%	35.3%	23.5%	34 / 34
Both sectors	Perplexity	59.0%	9.5%	75.3%	5.6%	210 / 198
Cookware (B2C)	Google AI Overviews	36.4%	29.5%	43.9%	24.4%	44 / 41
Cookware (B2C)	ChatGPT	31.2%	25.0%	20.0%	20.0%	16 / 15
Cookware (B2C)	Perplexity	61.9%	10.5%	84.3%	4.9%	105 / 102
B2B (components)	Google AI Overviews	28.3%	13.0%	37.2%	7.0%	46 / 43
B2B (components)	ChatGPT	27.8%	27.8%	47.4%	26.3%	18 / 19
B2B (components)	Perplexity	56.2%	8.6%	65.6%	6.2%	105 / 96

Reading. A Swiss reader gets answers built on the ecosystem of the neighbouring country that speaks their language. On Google and Perplexity, authority earned in that country counts for more than authority earned in Switzerland. That does not make Swiss sites poorly adapted.

Caveat. Zurich stands for all of Switzerland; Geneva or Lausanne might give something else. Bases: 34 to 210 citations depending on the tool. The direction is the same in both sectors for Google and Perplexity; ChatGPT ties in two of the four sector splits.

All cells: the result covers all answers.

Vertical-specificFor you

Authority depends on the sector before it depends on the tool

Which types of site does each tool cite, by sector?

In B2B, the manufacturer’s site is the first source: 78.8% of ChatGPT’s stable citations, 46.2% of Google’s, 51.2% of Perplexity’s citations. In cookware, the same share falls to 20.9%, 19.0% and 9.5%. When a site exists in several country versions, the version for the tested country is the exception: 24% of links on Google, 26% on ChatGPT, 22% on Perplexity.

In B2B, the manufacturer’s site dominates; in cookware, media and users

Source: Alizée Baudez, Cross-Border AI Visibility Study 2026 (site types by tool). Base: stable citations for Google and ChatGPT, all citations for Perplexity; n on each line. Share of citations, in %; shares under 4% are not labelled. Collected 25 September to 2 October 2026.

See the chart data

Sector	Tool	Brand site	Third parties: media, testing, advice, affiliates	Trade: retailers, distributors	Users: forums, video, social	Other	n
Both sectors	Google AI Overviews	33.1%	11.2%	21.6%	30.5%	3.7%	708
Both sectors	ChatGPT	54.3%	24.0%	7.0%	5.2%	9.4%	328
Both sectors	Perplexity	29.9%	39.4%	18.9%	8.1%	3.7%	1,774
Cookware (B2C)	Google AI Overviews	19.0%	20.2%	18.1%	41.9%	0.9%	342
Cookware (B2C)	ChatGPT	20.9%	56.1%	5.8%	8.6%	8.6%	139
Cookware (B2C)	Perplexity	9.5%	66.5%	10.5%	11.9%	1.4%	905
B2B (components)	Google AI Overviews	46.2%	2.7%	24.9%	20.0%	6.2%	366
B2B (components)	ChatGPT	78.8%	0.5%	7.9%	2.6%	10.1%	189
B2B (components)	Perplexity	51.2%	11.0%	27.6%	4.0%	6.0%	869

Reading. In B2B, presence in the answers starts with the company’s own site. In cookware, it runs through third parties: media that test, sites that compare, forums, video. A cookware brand that looks after only its own site acts on one citation in five on Google and ChatGPT, one in ten on Perplexity.

Caveat. Site types were coded by AI agents, with no second human coder: 6.8% of links are on low-confidence domains. Read the shares as orders of magnitude.

All cells: the result covers all Google searches.

Vertical-specificFor you

In cookware, the AI Overview recedes as the buyer nears the purchase

Does the AI Overview appear at every stage of the buying journey?

Overall, 62.7% of Google searches show an AI Overview (406 of 648). In cookware, an AI Overview appears for 90.7% of problem searches ("my pans keep sticking") and for 3.7% of those where the buyer looks for where to buy. In B2B: 79.6% at the problem stage, 84.4% at shortlist, 63.0% at purchase.

The nearer the buyer gets to purchase, the less the AI Overview appears and the more the brand is linked

Source: Alizée Baudez, Cross-Border AI Visibility Study 2026 (AI Overview trigger rate, answers by stage of the buying journey). Base: 648 Google searches, from 54 to 270 depending on the stage; for the share of linked brands, answers from all three tools to the questions that name no brand. Learning and purchase: two questions each, one per sector. Collected 25 September to 2 October 2026.

See the chart data

Stage	Cookware (B2C) (AI Overview)	B2B (components) (AI Overview)	Both sectors (AI Overview)	Named manufacturers also linked	n (stage)
Problem	91%	80%	85%	11%	108
Learning	82%	96%	89%	50%	54
Shortlist	35%	84%	60%	21%	270
Evaluation	32%	75%	54%	32%	162
Purchase	4%	63%	33%	81%	54

Reading. In cookware, the place where the buyer chooses is where the AI Overview steps back: Google sends people to its classic results. AI Overview visibility there is pre-purchase visibility. In B2B it stays present through to purchase.

Caveat. The stages rest on few questions: one per sector for learning and for purchase. The drop at purchase is therefore first of all the gap between two questions.

All cells: brand-site links cited by the three tools.

FindingFor you

ChatGPT cites language folders on a global domain

Which URL structure do the tools cite for brand sites?

ChatGPT cites country or language folders (for example se.com/fr/) in 67.1% of its brand-site links. Google’s top 10 has only 38.1%. ChatGPT cites fewer national domains (.fr, .de, .ch): 7.8% against 29.5%. Google and Perplexity stay close to their top 10 (39.2% and 37.2% of folders, against 37.1% and 38.1%). ChatGPT’s pattern holds in both sectors: 77.6% in cookware, 64.9% in B2B.

ChatGPT mostly cites language folders, few national domains

Source: Alizée Baudez, Cross-Border AI Visibility Study 2026 (URL structure, brand sites). Base: brand-site links cited: Google 1,771, ChatGPT 1,929, Perplexity 927; Google organic top 10: 1,792 links. Collected 25 September to 2 October 2026.

See the chart data

Tool	Language or country folder (se.com/fr/): cited by the tool	Language or country folder (se.com/fr/): Google top 10	National domain (.fr, .de, .ch): cited by the tool	National domain (.fr, .de, .ch): Google top 10	n
Google AI Overviews	39.2%	37.1%	22.5%	28.4%	694+
ChatGPT	67.1%	38.1%	7.8%	29.5%	1,294+
Perplexity	37.2%	38.1%	29.3%	29.5%	345+

Reading. ChatGPT turns to a global domain with language folders rather than to national domains. This is an observed association, not a demonstrated effect: the data do not say that URL structure makes a site get cited.

Caveat. ChatGPT’s 1,294 folder links rest on 40 domains, while ChatGPT cites 81 brand domains in all. Six of them carry 65% of these links (endress.com, se.com, siemens.com, hager.com, zwilling.com, sick.com): the finding mostly describes these large sites.

French and German rows: questions asked in the brands’ home language.

Vertical-specificFor you

Brands are named for their language, clearly in B2B

Are a country’s brands named more often when the question is asked in their language, or from their country?

Across the 16 questions that name no brand, the answers contain 2,012 manufacturer mentions. The share of French brands rises from 21.9% to 34.1% when the question is asked in French (12.2 points more). For German brands, from 35.1% to 50.4% in German. Location weighs less: from 2.7 to 5.3 points depending on the brands’ origin. The language effect is mostly a B2B effect: 16.5 points for German brands in German, 4.8 points for French brands in French. In cookware the gaps point the same way (19.1 and 12.7 points), but their intervals include zero.

A country’s brands gain points when the question is asked in their language

Source: Alizée Baudez, Cross-Border AI Visibility Study 2026 (brands of origin, gaps). Base: 2,012 manufacturer mentions in answers to the 16 questions that name no brand, three tools together. Gap in points between the share of home brands when the condition is met and when it is not. Whiskers: 95% interval; hollow dot: the interval includes zero. Collected 25 September to 2 October 2026.

See the chart data

Brands’ origin	Gap, in their language (points), 95% interval	Gap, from their country (points), 95% interval
French brands	+12.2 (3.2 to 26.9)	+4.0 (−0.5 to 9.9)
German brands	+15.3 (5.2 to 27.9)	+5.3 (1.3 to 10.7)
Swiss brands	–	+2.7 (1.1 to 5.2)

Reading. Manufacturers are named more often in their own language, clearly in B2B. Swiss brands gain little when the question is asked from Switzerland. One reading remains to be validated: having no language of their own, Swiss champions would build their authority in their neighbours’ languages.

Caveat. Mentions come from a brand dictionary: a brand that is missing is not counted, and a mention is not a recommendation. 95% interval of the language gap for French brands: 3.2 to 26.9 points.

Five hypotheses, two failures

The five hypotheses were written before collection. Two of them fail, and the study keeps them.

AI Overviews are the most localised tool

Share of local domains among sources: ChatGPT 33.2%, Perplexity 18.8%, Google 18.1%.

Not confirmed

English answers bring up global sources everywhere, even from Zurich

Share of global domains in English answers, by country: Google 80.8 to 84.4%, Perplexity 76.3 to 79.1%, ChatGPT 52.3 to 58.3%.

Confirmedin part for ChatGPT

Switzerland borrows its neighbours’ sources

From Zurich, in German, Perplexity cites 75.3% German domains against 5.6% Swiss ones.

Confirmedfor Google and Perplexity; mixed for ChatGPT

B2B relies on distributors and directories, cookware on resellers and review media

In B2B, manufacturers’ sites make 46.2% of Google’s stable citations, distributors and directories 23.8%.

In part

Champions are named more than linked, giants are both

Champion: a brand with revenue under €5 billion; giant: €5 billion or more. Manufacturers named that are also linked: giants 27.4%, champions 25.8%.

Not confirmed

The full verdict on each hypothesis, with bases and intervals, is in part 7 of the full report.

What becomes of the starting thesis

The starting intuition was that a brand’s authority does not cross a border on its own. The data refine it rather than confirm it. On Google and Perplexity, authority travels from one country to another as long as the language stays the same, and does not travel from one language to another. On ChatGPT, the country matters too.

For a company that sells on both sides of the Rhine, the consequence is concrete. For these tools, the German version of its site is not a translation of the French one. It is a presence on another market, facing other sources and often other competitors.

What this changes for a champion

A champion is a brand with revenue under €5 billion. Each recommendation rests on a result of this study. None promises traffic: the study does not measure any.

Treat each language as a marketFor you

What the results show. Two answers to the same question in two languages, from the same country, share only 0.08 of their sources on Google and 0.05 on Perplexity. In B2B, German brands gain 16.5 points of share of mentions when the question is asked in German, French brands 4.8 points in French.

What they do not show: That translating the site makes the brand appear in the target language. The study measures cited sources, not the effect of a piece of content.
To check on your side: In which language is your brand cited when the question is asked in each of your selling languages?