Where Does AI Send Traffic? 10 Markets Decoded for Your GEO

87.6 million visits. 57,696 entries. One finding that changes everything.

A client calls me on a Tuesday morning. He sells hiking gear across Europe. Germany, France, Italy, Spain. He wants to know where to spend his €10,000 GEO budget. His online store pulls in €2.3M annually, with 18% already coming from AI-generated responses on ChatGPT and Perplexity. Traffic he can’t control. The underlying mechanism is simple: each citational AI response acts as an algorithmic vote, distributing authority among a small pool of players. For him, the business implication is immediate: miss the right market, and you waste 40 to 50% of budget on segments where concentration makes entry nearly impossible.

« AI sends traffic to known brands, right? » he asks me.

I tell him: « It depends on the market. And the vertical. »

Because I’ve spent hours on the study Aleyda Solis just published. 87.6 million visits from AI-generated click-driving responses — ChatGPT, Perplexity, Google AI Mode — across three verticals (e-commerce, finance, travel) and ten markets (USA, UK, Spain, France, Germany, Italy, Netherlands, Australia, Mexico, Brazil). From those 87.6 million visits, 57,696 distinct domain-market pairs were extracted, giving exceptional granularity. Aleyda Solis’s counting method relies on aggregating referral flows from generative AI interfaces, weighted by estimated volumes via Similarweb. For a business, it means each data point represents a potential investment decision.

The result? Traffic concentration has nothing in common across sectors. And within a single sector, country to country, the gap is monumental. Take e-commerce in Italy: two domains suffice to cross the 50% AI click threshold. Amazon.it alone captures 46.2% of visits. By contrast, in British travel, the same threshold demands 129 domains. The business consequence is brutal: on a concentrated market, you’re dependent on presence across a handful of platforms; on a fragmented market, the quality of your semantic mesh and editorial content is what makes the difference.

This asymmetry undermines years of uniform SEO planning. Too many marketing directors still apply one template to their five European markets. The study proves it with hard numbers: Italy and Mexico, in e-commerce, are locked by 2-3 domains, while the Netherlands requires 11 domains to hit the 50% concentration median. The mechanism behind this concentration ties to the maturity of local marketplaces, the density of native media coverage, and the presence or absence of a price-comparison ecosystem. For a French e-commerce player eyeing the Netherlands: the market is accessible, but it demands a multichannel strategy with investment in local PR and listings on niche platforms like Bol.com or Beslist — whereas Italy imposes an Amazon-first approach.

Pattern 1: Concentration is not a detail—it’s a structural data point

First instinct: count how many domains capture half the clicks. This number, called the C50 concentration index, reveals the power structure in each vertical. A low C50 means a citation oligopoly; a high C50 signals a more democratic distribution of AI traffic.

In e-commerce, median is 5 domains. Five. In finance, 17. In travel, 47.

But that’s just the median. Look market by market:

In Italy, Amazon.it captures 46.2% of all AI e-commerce clicks. Add Temu, and you’ve crossed the 50% threshold with just two domains. In the USA, three domains suffice. In the UK, six.

This means one thing: in e-commerce, your GEO strategy is first a marketplace strategy. If you’re not showing up in Amazon results, you lose half the AI traffic. The mechanism is clear: citational AI models, when generating a purchase recommendation, scan marketplace product listings first because they’re the best-indexed, richest in structured data (price, availability, reviews), and most interconnected. For an e-commerce player, the business implication is budget prioritization: every euro spent optimizing your Amazon listings (titles, bullets, external backlinking) generates 3 to 5 times the AI traffic return versus equivalent investment in your own site on high-concentration markets.

Conversely, in British travel, you need 129 domains to hit 50% of clicks. Impossible to cover them all. You must choose the right discovery channels—and can’t count on just one. This fragmentation mechanism stems from the nature of travel queries themselves: « best hotel in Edinburgh with castle views, » « hiking guide for the Cotswolds, » « Calais-Dover ferry comparison. » Each solicits dozens of editorial sources—local blogs, traveler reviews, tourism offices. The business implication is forced diversification: a single OTA partner will never capture even 15% of AI traffic on this market. You must exist simultaneously across 15 to 25 complementary sources.

A concrete example: a French tour operator specializing in language immersion stays in the UK mapped, via Semrush, the top 25 domains cited by ChatGPT and Perplexity across 40 product-related queries. Result: 17 different domains, including 9 independent travel blogs, 5 price comparators, 2 regional tourism offices, and 1 collaborative guide. They pursued an editorial contribution strategy on these 9 blogs—guest articles and backlinks. In 6 months, their AI citations jumped +35%, referral clicks +28%, and conversion rate on those clicks hit 4.1% versus 2.6% for same-period Google Ads. The lesson: on a fragmented market, network approach beats single-platform approach.

I recommend always checking C50 before any GEO budget. If the number is below 6, place 70% of effort on marketplace listings and structured data. Above 40, place 70% on fresh local content creation and backlinks from niche media.

Pattern 2: AI traffic flows to local infrastructure, not global giants

Second common mistake: believing AI assistants cite the same players everywhere. This belief comes from skimming traditional SEO topsites where Wikipedia and Amazon hog the top spots. Yet Aleyda Solis’s study demolishes this idea: across 57,696 domain-market entries, 83% of domains appear in only a single market. A colossal figure.

Yes, Amazon and Booking are powerful. But not uniformly everywhere. The underlying mechanism is localized learning: each language model, when processing a query in German or Dutch, activates different weighting layers favoring training corpora in that language and locally-cited sources. The effect is a sort of « algorithmic national preference » even the power of American giants can’t override.

In German finance, the top 9 national domains (Sparkasse, Deutsche Bank, Comdirect) capture over 50% of clicks. No American bank in the top 5. Concretely, Sparkasse commands nearly 18% of AI citations on banking queries in Germany. For a French fintech wanting to enter Germany, the strategy involves earning backlinks from German comparators (Check24, Verivox)—not multiplying PR in Les Échos or Le Figaro. A recent case: N26, a German mobile bank, built AI visibility by publishing quarterly studies on German savings behavior, systematically covered by local financial press. Result: it landed in the top 10 of German finance AI citations in under 18 months, capturing 4.3% of that vertical’s AI flow.

In Mexico, finance is dominated by BBVA Mexico, Banorte, Santander Mexico. No Chase, no Citi. The local trust mechanism is reinforced by use of a national financial registry, the Buró de Entidades Financieras, that AI models consult regularly to verify licensing. The business implication: if you operate in Mexican finance, your entry in this registry, mentions in local financial press (El Economista, Expansión), and links from local comparators (Rastreator.mx) matter more than global SEO.

Same pattern in travel: local OTAs (eDreams in Spain, Opodo in France, Vliegtickets in the Netherlands) capture click share that Booking.com can’t alone. Take Spain: eDreams holds 12% of AI clicks on domestic flight queries, versus 23% for Booking. But on « flight + hotel » queries, eDreams jumps to 21%. A 9-point gap proving specialized local players carve niches of AI citations generalists miss. For a French travel agency targeting the Spanish market, integrating into eDreams’s affiliate flow can generate +30 to +45% additional AI visibility in 6-8 months.

The « global brand wins » myth collapses under scrutiny. AIs learn from local web corpora, local reviews, local directories. Without local presence, you don’t exist.

For a French e-commerce seller entering Italy: you need Italian Amazon listings, localized pages, Italian backlinks. Not just .fr. Example: Decathlon built detailed Italian Amazon listings with localized descriptions, euro pricing, and crucially, Italian customer reviews. In 10 months, AI traffic to these listings jumped +62%, while corresponding .fr pages stalled at +7%. The difference: local trust score, measured by sites.it linking to these listings. Simple mechanism: the AI, seeing a domain linked from 40 Italian sites, classifies it as relevant for an Italian query, even if the company is French.

Another important nuance: language trumps TLD. An Italian-language page on a .com domain with correct hreflang and .it inbound links often beats a .it page poorly linked. LLM algorithms, during the retrieval phase, don’t check domain extension but language, named entity relevance, and local meshing. The business implication for international SEO leads: translate, yes. Localize entities—addresses, phone numbers, country-specific legal mentions—that’s what triggers AI citation.

Pattern 3: « AI Search is growing »—yes, but with violent turnover

You hear everywhere that AI-response traffic grows +X% monthly. True. But what’s not said: the churn. The study reveals a domain replacement rate of 28 to 35% between monthly analysis windows across markets. Enormous: one-third of cited sources change each month.

In the study, between monthly analyses, some domains lost 40% of AI visits. Others gained 200%. Citation position isn’t stable. Not like a traditional SEO snippet. The mechanism: AI retrieval pipelines (RAG—Retrieval Augmented Generation) aren’t static; they depend on latest indexing, perceived freshness (last-modified date), and model updates. A domain spiking in March citations can crash in April if a competitor publishes 20 similar, fresher pieces.

Why? Because models shift sources, crawls refresh, competitors publish newer or better-structured content. I observe a precise phenomenon with my e-commerce clients: product pages updated with new pricing and fresh « last-modified » dates see citation rate jump +18 to +24% average in the 48 hours post-crawl. Overlooking stock or description updates causes immediate citation loss.

I track with my clients that publication regularity beats single peaks. One blog post doesn’t cut it. You need sustained content flow, optimized for citation. The mechanism is well-known to traditional search: « query deserves freshness » (QDF)—now applied by AIs, favoring sources from the last 30 days for informational queries. Aleyda’s study confirms: 71% of domains in a vertical’s top 10 had published fresh content within 7 days of analysis.

Concrete example: a travel client quadrupled AI visits in six months by publishing one destination guide weekly. Not monthly. Weekly. Each guide used the same semantic plan (intro, weather, lodging, activities, transport, budget, FAQ), letting AI crawlers detect format consistency and boost domain trust. Result: weekly AI visits jumped from 320 to 1,280, with bounce rate 22% lower versus traditional Google traffic.

Another client, in consumer electronics, tested refreshing 80 product pages every 15 days (new customer reviews integrated, prices updated, tech specs revisited). Three months later, AI citations on « best 27-inch PC monitor » queries rose +55%. Double mechanism: crawlers detect regular updates, strengthening freshness signal, and enriched content delivers new relevant tokens for semantic long tails. The business implication: allocate 20% of editorial time to refreshing existing content, not just creating new. A 2023 blog article updated in 2026 with fresh stats has 3.2x higher chance of AI citation than the original untouched article.

Pattern 4: Finance is « trust-led »—but the devil is in the details

Conventional wisdom says AIs cite trusted institutions in finance. Correct: historical banks, certified comparators, government sites. The mechanism rests on E-E-A-T (Expérience, Expertise, Authoritativeness, Trustworthiness)—already central to Google Search, but LLMs apply it more selectively: they check licensing signals, accreditation, legal notices, and reputation in authority sources.

But it’s not straightforward. Finance’s top 10 captures only 44.2% of clicks average. Far below e-commerce’s 63.9%. And it takes 17 domains to hit 50%. This « glass ceiling » at 44% explains the diversity of finance sub-queries: home loans, life insurance, trading accounts, crypto, neobanks, equities, etc. Each sub-segment mobilizes a different domain ecosystem. A traditional bank may dominate mortgage lending but vanish from crypto citations, where specialized media emerges.

On the French market, 9 domains suffice (Crédit Agricole, BNP, Assurance Banque, etc.). But in the UK, it takes 57. Why? The British market fragments between financial supermarkets, neobanks, comparators. The mechanism is historical: UK market liberalization in the 1990s spawned multiple players (Tesco Bank, Sainsbury’s Bank, Virgin Money) versus France’s structure around legacy networks. AI models reflect this diversity: a « best savings account UK » query activates citations to MoneySuperMarket, CompareTheMarket, Martin Lewis (MoneySavingExpert), plus a dozen challenger banks like Monzo or Starling. For a French actor entering British finance, you need presence across 4-6 comparators, plus sponsored guides on high-authority financial information sites (Which?, FT Adviser).

The lesson: trust isn’t just fame. It’s also built through editorial presence (guides, comparisons, reviews). If your finance site lacks backlinks from authority local sources, the AI won’t cite you. Concrete case: a Luxembourg life-insurance company targeting French prospects invested in a comprehensive Luxembourg-contract tax guide, hosted on a reputed French financial outlet (Investir). In 5 months, that page’s AI citations climbed from 0 to 27 monthly, generating a qualified 450 monthly visitors. The mechanism: the AI transferred host-outlet authority to insurer content, creating algorithmic trust transfer.

Another nuance: AIs overweight regulatory structured data. A finance site balking its legal notices, licenses (ACPR, AMF), and fee structures in schema.org/FinancialProduct sees citation rate rise +40 to +60% in 6 months on decision-driving queries. The business implication is straightforward: before producing more content, structure your regulatory data first. A schema audit is the first GEO investment in this vertical.

Pattern 5: Travel seems open…but booking stays concentrated

Travel is the most fragmented vertical at the click level. 70 domains for 50% in the USA, 129 in the UK. Looks like a massive opportunity. Discovery does mobilize a myriad: travel blogs, Lonely Planet guides, influencers, tourism offices, local media. The mechanism ties to experiential query nature: « what to do in Kyoto in 3 days » inevitably solicits varied sources covering expérience diversity.

Yes, discovery. Travel blogs, guides, « best hotel in… » articles grab huge click share. A concrete figure: a Bordeaux travel blogger saw 4 pieces (Bordeaux by bike, Where to sleep in Biarritz, Canal du Midi guide, Basque hiking) generate 870 combined monthly AI clicks—32% of total traffic—with just 31% bounce rate. Reason: his first-person content with GPS coords, verified prices, personal reviews matched exactly what AIs source for authentic expérience.

But look at the booking layer: OTAs (Booking, Expedia, Airbnb) and airlines dominate the podium. Small lodges or local agencies capture mere fractions. The top 3 booking domains in the study captured 58% of AI clicks, while the top 3 discovery domains only 12%. The mechanism: shifting from discovery to action (reserve, buy ticket) makes AIs prioritize secure platforms, fastest checkout, highest-rated, smoothest UX. Citation criteria evolve from editorial richness to transactional performance.

A hotel client told me: « I get AI clicks on my description page, but they book on Booking. » The problem: an AI can cite you, but the CTA often leads to an OTA if your own booking process isn’t integrated. ChatGPT Search logs show 73% of booking links point to OTAs, 15% to direct hotel sites, 12% elsewhere. Huge gap.

Solution: optimize your direct booking page with structured data, frictionless flow, integrated Google reviews. Otherwise you’re just a storefront for competitors. A Provence hotel applied this: added Hotel schema.org with real-time availability, one-click booking widget (no redirects), Google reviews aggregated on-page. In 4 months, AI-click-to-direct-booking share jumped from 8% to 26%. The mechanism is a transactional competence signal: the AI detects the site can handle full-cycle booking, positioning it as credible OTA alternative.

Additional precision: travel actors gain AI citation share publishing hybrid formats like « optimized itineraries » with interactive maps, duration, budget, alternatives. These formats check 3 algorithmic boxes: utility, structure, freshness. A tour operator publishing this weekly across 3 regions sees AI citation rate jump +45% in 6 months. I tracked this across 12 travel clients between 2025-2026: median AI traffic increase was +42%, peaking at +78% for those combining fresh content, structured data, and local media backlinks.

What to do concretely for your global GEO?

The 5 patterns point to an obvious truth: one-size-fits-all GEO strategy doesn’t work. You need a reading grid per market and vertical. A granular, country-by-country approach based on concentration and turnover data.

1. Audit your current AI presence. Use Similarweb or tools like Semrush to see which domains in your vertical capture clicks on each market. Don’t just check your own pages: map the top 30 cited domains across your 50 priority queries. Spot gaps, redundancies, competitors outpacing you. A full audit takes 3-4 days but saves 6 months of strategy errors.
2. Prioritize markets with low concentration. Example: UK travel at 129 domains for 50% means better odds of long-tail placement. But low concentration means more editorial work. Plan 52 publications yearly to exist.
3. Don’t ignore marketplaces. Selling e-commerce? Amazon, Cdiscount, Zalando are AI channels by market. Be present—not just on your site. Each marketplace listing is an extra page AIs can cite. Optimize with SEO standards: titles, semantic-dense descriptions, high-res images, answered reviews.
4. Produce local, fresh, regular content. AIs love new. One article monthly doesn’t cut it. Weekly does. Set up an assembly line: brief Monday, write Tuesday, review Wednesday, publish Thursday. Regularity is an algorithmic freshness signal models auto-detect.
5. Build local backlinks. AIs measure trust via citation networks. A link from local media outweighs 10 generic links. Identify 3-5 influential outlets in your target market, pitch high-quality guest content, include contextual links. This digital PR work is the highest long-term ROI for GEO.

I add a rule I’ve verified with clients: local structured data is the AI SEO lubricant. Schema.org LocalBusiness, Product, FAQ, HowTo. The more AI-digestible your page, the more it gets cited. Schema acts as a « user manual » for AIs—it tells them exactly what your page contains in a format they instantly process. Without it, you’re leaving AIs interpreting blind.

Proof? A French travel client added schema to 200 hotel listings. Three months later, AI citations (via ChatGPT Search) jumped from 14 to 89 weekly. Zero new content. Just schema. Technical investment? €2,500 dev, yielding 75 weekly AI citations—€33 per citation cost, an ROI many Ads campaigns never reach.

To go further, segment your GEO strategy into three operational layers: the foundation layer (schema + backlinks), the flow layer (regular publication + refreshes), and the activation layer (marketplace presence + digital PR). Allocate budgets per target market concentration: high concentration gets 70% marketplace + structured data focus; low concentration gets 70% local content + PR.

Stéphane Jambu

SEO & AI Engineer

I build growth systems / AI / Neuroscience | 650+ clients · 80 LinkedIn testimonials · 30 years of expertise · 15 years of systems running without me.

Follow on LinkedIn