Most companies still brief their digital work as if the customer journey is stable: publish a website, get indexed by Google, win the click, convert the visitor on the site.
That journey still exists. It is just no longer safe to treat it as the only journey.
For a Luxembourg SME or mid-market company, the practical decision this quarter is simple: keep the website as the authoritative source, but start making your product, service, price, availability and trust data usable by AI agents.
The old web was built around pages. The next layer is being built around answers, actions and small pieces of interface that appear inside ChatGPT, Claude, Google AI Mode, Perplexity or another assistant. A customer may not browse your product gallery. They may ask an agent to shortlist options, compare prices, check delivery, book a meeting, draft the email and complete the purchase.
That is not science fiction anymore. It is early, uneven and full of open questions. But the direction is now visible enough that companies should stop thinking only in terms of website traffic.
The strongest evidence is not that search is dying
The useful signal is more specific: the click is losing its monopoly.
Google said in March 2025 that it sees more than 5 trillion searches annually. That is still a massive distribution system. Alphabet also told investors in July 2025 that AI Overviews had more than 2 billion monthly users across more than 200 countries and territories, and that AI Mode already had more than 100 million monthly active users in the U.S. and India. Google also said AI Overviews were driving more than 10% more queries globally for the types of queries where they appear.
That does not look like the end of search. It looks like Google turning search into an answer and conversation layer before someone else does.
OpenAI is building the same shift from the other direction. An OpenAI economic research paper says that by July 2025, ChatGPT had more than 700 million weekly users sending 18 billion messages per week. OpenAI later said ChatGPT served more than 800 million users every week. ChatGPT search became broadly available in February 2025, and OpenAI has since moved into shopping, checkout, apps and ads inside ChatGPT.
Perplexity is smaller, but the direction is also useful. In June 2025, TechCrunch reported that Perplexity's CEO said the service received 780 million queries in May 2025 and was growing more than 20% month over month. Even if that growth moderates, it shows that a meaningful number of users are willing to use an answer engine instead of a classic search results page.
The numbers are not directly comparable. A ChatGPT message is not the same thing as a Google search. Perplexity's number was a CEO statement reported by the press. Anthropic does not publish a simple Claude query-volume number. But the broad pattern is clear enough: more discovery, research and decision-making is happening inside AI interfaces.
The click data is the warning sign
The most important question for businesses is not whether people still search. It is whether those searches still become website visits.
Pew Research Center analysed U.S. Google browsing behaviour in March 2025. When an AI summary appeared, users clicked a traditional result in 8% of visits. When no AI summary appeared, they clicked a traditional result in 15% of visits. Pew also found that users clicked links inside the AI summary in only 1% of visits to pages with such a summary.
That is the business problem. A company can be used in the answer without receiving the visit.
SparkToro and Datos found a similar long-running pattern before AI summaries became the main story: in 2024, 59.7% of European Union Google searches and 58.5% of U.S. Google searches resulted in zero clicks. Cloudflare made the publisher-side version of the same argument in July 2025: the old trade was that search engines copied content and sent traffic back; AI systems can consume more content while sending less traffic.
This does not mean nobody will visit websites. It means the website visit becomes a later, more selective step. The first screen may belong to the agent.
Interfaces are moving into the assistant
MCP Apps became the first official extension to the Model Context Protocol in January 2026. The core idea is simple: a tool can return an interactive interface, not just text. That interface can render directly inside the conversation as a chart, form, dashboard, map, document viewer or checkout flow.
OpenAI's Apps SDK and the MCP Apps extension both move in this direction. Companies are no longer only publishing pages to browsers. They can publish small, contextual interfaces to assistants.
That matters because most websites are not designed for agents. A human can interpret navigation, tabs, forms, vague copy and marketing language. An agent needs clean facts, current data, clear permissions and safe actions. It needs to know what can be quoted, compared, priced, booked or bought.
The strategic question becomes less "how do we drive everyone to our homepage?" and more "how does an agent reliably understand, recommend and act with our business?"
My opinion
My opinion: the web will not disappear in the next three to five years. The stronger prediction is that the website will stop being the default front door for many informational, comparison and transactional journeys.
That is still a large change.
For many businesses, the homepage will become less important than the structured truth behind it. Product feeds, service pages, pricing logic, availability, reviews, policies, APIs, booking systems and agent-accessible workflows will matter more. The companies that win will not necessarily have the most beautiful website. They will have the clearest public facts and the safest action paths.
There is a real pushback. Gartner's January 2026 consumer research found that only about one-third of consumers believe generative AI chatbots are as effective as search engines for learning new information. The same research found that AI summaries can lengthen research: 31% of surveyed consumers said AI summaries made them spend more time searching, compared with 16% who spent less.
That is why "websites are dead" is too blunt. Trust-heavy purchases, local services, regulated decisions, B2B deals and anything involving risk will still need proof, human contact and owned digital spaces. But those owned spaces may increasingly feed an agent before they receive a visitor.
What Luxembourg companies should do now
The useful move is not to stop building websites. It is to build the website as an agent-readable source of truth.
- Create an AI visibility register. Pick 20 buying questions your ideal customer might ask, in English and French if relevant. Test them monthly in Google AI Mode or AI Overviews, ChatGPT, Perplexity and Claude.
- Rewrite key pages as decision pages. State who the service is for, who it is not for, where you operate, what it costs or how pricing works, what is included, what constraints apply, and what proof supports the claim.
- Clean the machine-readable layer. Add or improve structured data, product feeds, merchant feeds, business profiles, review profiles and current opening or availability data.
- Separate the interface from the action. If quote creation, appointment booking, stock checking, payment, onboarding or support triage matters commercially, decide how it could be exposed safely through an API, connector, MCP server or app experience later.
- Measure beyond normal analytics. Track branded search changes, direct traffic, assisted conversions, sales-call source questions, ChatGPT and Perplexity referrals, server logs and mention or citation visibility.
The strategy
The next digital strategy is not "website versus agents." It is website plus agent distribution.
The website remains the legal, commercial and trust anchor. It is where your owned content lives, where humans can verify you, and where higher-risk conversions still happen. But it should no longer be designed only as a destination. It should also be designed as a data source, evidence base and action layer for AI systems.
The companies that prepare early will not need to guess which assistant wins. They will have a cleaner source of truth, stronger structured data, better comparison pages, safer workflows and a direct audience they own. That work helps SEO today. It also makes the business easier for agents to understand tomorrow.
The risk is not that websites vanish overnight. The risk is that customers stop needing to visit yours before they decide.