A website today has two layers that work completely independently of each other. The AI-Ready layer is universal. It makes your site readable for ChatGPT, Perplexity, Claude and Bing Copilot, no matter where the bot comes from. The hosting layer is regional. It decides who gets to touch your data, how fast the site loads for your customers, and which legal jurisdiction your business sits in. We build the first layer the same way for every client. The second layer we put where your market lives: Hetzner Germany, AWS us-east, AWS London, Infomaniak Switzerland. Multi-region failover on top when you need it.
We have been talking about AI-ready web design and AI visibility for months now. What got short shrift is the honest answer to a question that comes up more and more often: what does the physical location of a server still have to do with visibility in AI answers, when the bots crawl globally anyway? The answer is uncomfortable for anyone looking for a clean pitch. Hosting region and AI visibility have very little to do with each other. Hosting region and trust, latency, compliance and sales effectiveness have a lot to do with each other. The two should not be thrown into the same bucket.
A website now has two layers
When we build a website, we think in two layers. The first layer is the AI-Ready layer. That covers semantic HTML5 as the skeleton, Schema.org JSON-LD as the meaning layer, agents.json and llms.txt as the table of contents for AI crawlers, robots.txt explicitly opened for GPTBot, ClaudeBot, PerplexityBot and Bing Copilot. Server-side rendering so the bot does not stare at an empty div. Factual text that answers direct questions instead of marketing fluff. This layer builds identically whether your site lives in Frankfurt, Dublin, Virginia or Zurich.
The second layer is the hosting layer. That covers the physical server, the data centre, the storage backend, the database, the CDN. This is where you decide who owns your data, which authority can demand access, how fast your site loads for a user in San Francisco or London, and whether a compliance audit in your industry can pass. This layer is not universal. It depends on who your customers are, where they sit, what they buy, and which regulations are breathing down your neck.
Anyone who fails to separate the two ends up either selling GDPR as a magical AI visibility argument or edge CDN as a magical compliance solution. Both are nonsense. The clean separation looks like this: AI-Ready is the visibility question. Hosting region is the business question.
What AI crawlers actually do
The honest question first. Do ChatGPT, Perplexity, Claude or Gemini favour websites based on their hosting region? We researched this carefully and the answer is: no, there is no solid data for that. Cloudflare publishes a half-yearly AI bot traffic report, the relevant AEO platforms like Stackmatix or Metricus track crawl behaviour, and no serious source shows a regional preference algorithm in the major LLM crawlers. Anyone promising you that your Vercel Edge setup will boost your ChatGPT citation rate is selling you a guess as a fact.
What AI crawlers actually do is more boring and more important at the same time. They fetch your HTML page via HTTP request like any other bot. They look at what is in the DOM without executing JavaScript. They follow your robots.txt and either respect it or ignore it depending on the crawler. They do not cache server-side in the classic CDN sense, they request directly. What counts is whether your site responds quickly, whether the content is there without JavaScript, whether your structured data is complete, whether your agents.json exists. Latency at crawl time matters. If your server takes four seconds for the first byte, the chance is high that the bot drops the request and does not come back.
In practical terms: a Hetzner server in Nuremberg with a clean configuration is just as good for a ChatGPT crawl as a Vercel Edge setup with a hundred global nodes. As long as the first byte arrives quickly and the HTML is complete, the bot sees no difference. Edge hosting becomes interesting when you serve real humans across multiple regions. For pure AI visibility it is a nice-to-have, not a must-have.
Why hosting region still matters
If AI crawlers do not care, why are we even talking about hosting region? Three real reasons, none of them anything to do with AI, all three plain business.
The first reason is latency for human visitors. A page that lives in Frankfurt and gets served to a user in San Francisco needs at least 150 to 200 milliseconds more for the first response than a page that lives in us-east-1 or us-west-2, depending on stack and caching. With a single-page app and several API calls the effect multiplies fast. Anyone serving US customers loses noticeable conversion when hosting from Europe. The same the other way around.
The second reason is sectoral compliance. GDPR in normal commerce does not force you to a German server. It forces you to adequate protection levels and clean contractual chains. A US server with standard contractual clauses and EU-DPF certification is legal, just less convenient. Real localisation requirements only exist in specific sectors. Health, government, critical infrastructure, parts of finance. UK GDPR and the Swiss FADP do not force normal companies into local storage either. What almost always matters though is the trust signal. If your UK client wants a data processing agreement and you say the data sits in London on AWS, the conversation is over in two sentences. If you say the data sits in Frankfurt with standard contractual clauses, you spend half an hour in a compliance call.
The third reason is sales effectiveness. In the United States it is taken for granted that American clients sit on American servers. Not because of compliance, but because of expectation and political climate. In Switzerland, serious clients expect Swiss hosting. That is not hard duty, that is market code. Anyone swimming against it burns conversion for no reason. Anyone swimming with it has a sales advantage the competition does not have.
Four regions, four setups
This is how we run it in practice. Four region patterns, each one real and ready to deploy.
EU as the default means Hetzner Germany for the server, Supabase EU for the database. Hetzner has data centres in Nuremberg, Falkenstein and Helsinki. We pick whichever fits the use case but stay deliberately flexible because Hetzner has no Frankfurt data centre. Supabase EU runs on AWS Frankfurt. Both providers have lived in the GDPR world for years, both have transparent sub-processor lists, both are aggressively priced. This is the default for any client without special requirements. Our own SaaS products (Memory, CRM, GEO, Crew) live there too.
US hosting means AWS us-east-1 in Virginia plus Vercel Edge for the frontend CDN. AWS us-east-1 is the largest and most stable US region, also the one with the densest service catalogue. Vercel Edge brings 100+ global POPs, US users hit the closest. The combination is standard in the US market and answers all the compliance questions American B2B customers ask without discussion. HIPAA if you operate in healthcare, FedRAMP if you sell to the public sector, both reachable through dedicated AWS regions.
UK hosting means AWS eu-west-2 in London plus Vercel Edge with active London POPs. UK GDPR and EU GDPR are still largely aligned, but post-Brexit clients in regulated industries want to see their data physically in the UK. AWS London is the clean answer. Vercel Edge keeps responses fast across all of Britain. If your UK client is more Azure-aligned, we can add Azure UK South or build it as the alternative.
Switzerland hosting means Infomaniak Geneva or Exoscale Zurich. Infomaniak is a Swiss provider with its own data centre in Geneva, runs its own cloud infrastructure, is sectorally cleared for critical infrastructure. Exoscale is a Swisscom subsidiary with zones in Zurich and Geneva, fully FADP compliant, suits clients who want a Swiss provider without giving up hyperscaler comfort. The combination covers everything from a small Swiss SME to a pharma company with FADP requirements that go beyond what GDPR demands.
Multi-region failover when you need it
Single region is the right setup for 90 percent of clients. Cheaper, simpler to operate, fewer failure vectors. Multi region becomes interesting when your business has zero tolerance for downtime or you operate across multiple markets with hard latency requirements.
Multi-region failover is part of our pattern. We use it ourselves for our Memory product, which runs with a standby server and automatic switch within seconds of a primary failure. For custom clients we build it on demand depending on requirements. Three patterns we typically recommend:
Active-standby in the same region with a floating IP. Quick to build, switch in under ten seconds, protects against single-server failure, does not protect against region failure. Enough for most business models that need high availability but not full disaster recovery.
Active-standby across two regions with DNS failover. Cloudflare health checks poll every 60 seconds, on primary failure the DNS record switches to the standby in another region. Switch takes one to two minutes depending on TTL. Protects against region failure, costs a bit more because the standby hardware runs alongside. This is the pattern we are currently moving Memory to.
Active-active with a load balancer and replicated state. Both regions take traffic, the load balancer distributes by geo or load. More complex to operate because state replication has to be built cleanly. Makes sense from a certain volume on, or when you actively want both regions for latency optimisation.
Which pattern fits depends on what your real requirements look like and how big your budget is. Most clients do not need active-active. Many need more than just a single server. Active-standby in the same region is usually the sweet spot.
What stays the same everywhere
The AI-Ready layer does not change with hosting region. What we install on every setup is the same bundle. Semantic HTML5 with clear section, article, nav and main elements. Schema.org JSON-LD in the head, at minimum for Organization, WebSite, BreadcrumbList and Service or Product. agents.json under /.well-known with the tools your site offers AI agents. llms.txt at the root with the AI-readable content overview. robots.txt with explicit allow for GPTBot, ClaudeBot, PerplexityBot, Bing Copilot, Google-Extended. And server-side rendering or static site generation so the bot actually finds text to read instead of an empty React div.
Whether Frankfurt, London, Virginia or Zurich, this stack runs identically. That is also why we do not couple AI visibility to hosting. We build the AI-Ready layer the same for every client and then place it on the region that fits your market. You get the visibility without forcing your US clients to accept their data living in Germany, or your Swiss client to accept hosting in Ireland.
How we decide in practice
When you start with us we clear three questions before we lock in the server stack. First question: where do your buyers sit? Not where your company sits, where the people are who buy. A DACH SME serving mostly US software clients needs US hosting, even when based in Hamburg. A Mallorca hotel chain with German guests needs EU hosting, even if the owner is British.
Second question: which industry, which compliance? Health, finance, government, critical infrastructure need sectoral answers and sometimes hard localisation. Normal SMEs and e-commerce do fine on EU or US, depending on the market. UK-specific compliance like FCA requirements or Swiss FINMA regulation often demand local presence. We can call out the pain points before the build starts.
Third question: do you need failover or is single region enough? 90 percent of SME websites do not need multi-region. Anyone running a high-volume online shop, a SaaS app with real SLA pressure, or a booking flow for a hotel with 200 rooms should think about active-standby. Everyone else lands well with a properly monitored single server setup.
From these three answers the region choice usually drops out by itself. We turn that into a concrete setup with provider names, compliance documentation, backup plan and monitoring. You get an answer in one sentence, not a 30-page architecture document.
What to take from this
AI visibility and hosting region are two topics, not one. Anyone selling you the opposite is doing marketing, not architecture. The AI-Ready layer you build cleanly once and then it stands. The hosting layer you place where your customers, your compliance and your business sit. Both decisions are independent and both should be made consciously, not from the gut.
We build exactly this separation into every web design project. Anyone working with us gets an AI-Ready stack that has a real chance in any LLM answer, plus a hosting setup that fits their market. Serving a US market gets US hosting, compliance-relevant Swiss clients get Swiss hosting, anyone wanting DACH default gets EU hosting. Multi-region failover on top when the business needs it. Otherwise spared and kept simple.
If you are wondering whether your site is built for AI answers and whether your hosting setup fits your market, take a look at our web design page or book a call directly. We sort the question in 15 minutes.
