Over the past few months, a growing share of a website's visitors have had neither a screen nor a mouse. They are AI bots. They read pages, summarise them, sometimes recommend them, and leave without a trace in Google Analytics. According to Cloudflare, in July 2026, OpenAI crawled more than 200 pages for every visitor it sent back to a site. Anthropic, more than 2,000. Google, around 5.
This traffic sends fewer people, but more decided ones: a visitor coming from ChatGPT or Perplexity arrives already informed, already having compared, and converts noticeably better than a visitor from traditional search. Naturally, this changes the way we build a website. Here are four things we put in place on our projects.
The first concerns the copy.
SEO isn't going away, it's getting broader. Alongside optimising for Google, we now write to be picked up and cited by generative engines, what we call GEO. In practice: we answer first, then elaborate. Subheadings written as real questions, short paragraphs, content that is signed, dated and sourced. What helps an AI extract information also helps a reader in a hurry.
The second happens in the code.
An AI doesn't see your layout. It reads code. If the price, the address or the opening hours are only readable visually, they don't exist for it. So we add a layer of structured data (schema.org) that explicitly describes what the page contains: organisation, product, service, FAQ, article. And we make sure the main content is present in the HTML delivered by the server, because many bots don't run JavaScript.
The third is what the user sees first. A visitor sent by an AI doesn't come through the homepage. They land directly on a product or service page, often from their phone. The first screen therefore has to answer four questions within seconds: am I in the right place, what exactly do you offer, can I trust you, what should I do now. This hierarchy is decided before colours and animations.
The fourth is speed, with AI on the developers' side this time. Here it is no longer a visitor but a tool. Our developers use it to audit code, spot what slows things down, generate the right image formats, prioritise loading and test Core Web Vitals before going live. It also helps manage bot traffic itself: distinguishing in the robots.txt between training crawlers, which can be blocked, and answer bots, which can cite the site and send visitors back. A lighter site for humans, an easier one to read for machines.
AI traffic doesn't replace human traffic. It filters it. A website ready for this reality is one that is clear in its copy, explicit in its code, reassuring from the first screen, and fast. None of this is reserved for large companies: these are working methods, just as valid for a showcase site as for an online shop.
joy in creating joy in growing joy in thinking joy in creating joy in growing joy in thinking