Zero click is real. Declaring every website dead is nonsense.
Zero click has been studied: Pew observed fewer clicks on search results when Google displayed an AI summary; the cited sources were rarely opened. [1] My forecast: as more information needs are met directly in the answer, more of those clicks will disappear.
But getting information, buying a product and trying a tool are different things. An answer is enough for a definition. I learn how a tool feels by using it. Whether I hire a company also depends on how it presents itself, what it has delivered and how well it understands my problem.
Our website needs to do two jobs: supply AI with clear, current and technically accessible information. And invite people to experience the product, service or company. The answer helps people find us. The visit needs to give them a reason to choose us.
A citation does not pay the bills.
I expect fierce competition over which businesses appear in AI answers. SEO has new work to do; GEO, optimisation for visibility in generative answers, supplies the new acronym.
The uncomfortable question remains: what exactly are we winning?
Is my website cited as a source, or is my business recommended as a provider? If my explanation of heat pumps supports an answer, AI has not necessarily sold my heat pump. A mention can build trust and lead to an enquiry later. It can also remain a useful footnote.
We need to look separately at the questions where we appear and the commercial results that follow. Qualified enquiries, purchases, repeat purchases. Conversations with customers about how they found us. Attribution will not always be clean. But a growing citation count alone feels far too convenient as proof of success.
If a person still clicks, we cannot afford to screw it up.
I mean this quite explicitly for ordinary AI search: someone reads an answer, takes an interest in our offer and visits our site. At that moment, we need to deliver.
Whether that visitor is more ready to buy than before needs measuring. Losing them because the next step is badly built is a problem we can spare ourselves.
If we fight for every relevant mention and then bury the visit we earned in confusing product choices, we have built ourselves a fairly expensive dead end.
For me, conversion optimisation means making a decision possible. What do I get? Does it fit my needs? What does it cost? Why should I trust you? What happens next? Those questions need useful answers. In the right place. Without detective work.
In a shop, that includes understandable variants, transparent total costs and a checkout that does not make buying unnecessarily difficult. For a service, a relevant case study, a clear scope and a simple enquiry can make the difference. That is what our web design needs to be judged against.
Closing the deal does not mean pushing something on every visitor. If the offer does not fit, they should understand that quickly as well. I want to win the people we can actually help. They should not leave because we built the final step badly.
With less traffic, the maths becomes more demanding. A model example: 1,000 visitors at two per cent conversion produce 20 completions. With 500 visitors at three per cent, the result is 15. The rate improves; the business shrinks. All else being equal, four per cent would be needed to get back to 20 completions.
That is not an argument against conversion optimisation. It shows how much we need it to achieve. We should consider rates, absolute completions and customer quality together, and test the critical journeys. A UX audit can help identify the biggest obstacles before we rebuild the whole site on a hunch.
And no, a twenty-minute visit does not prove delight. Someone may have spent twenty minutes looking for the price. A good website is allowed to do its job in two minutes.
A brand needs to do more than sit beside an answer.
The more interchangeable the answer boxes become, the more a brand needs something of its own to say. Our website gives us room to do that: show a product, make its impact tangible and give the company a face.
AI does not have to preserve the tone of voice we developed so carefully. It can mention us and still sound like itself. We cannot count on a cheeky brand voice automatically producing a cheeky AI summary.
A brand needs more substance: a distinct position, recognisable product qualities, demonstrable experience and promises that hold up in daily use. On a website, we can turn that into an experience. Through language, imagery, interaction and design willing to take a few risks.
I have no appetite for a future in which every business looks like black text on a pale grey background. I also have no appetite for navigating an art installation just to find the shipping costs.
Good design needs to do both: make you feel something and handle an ordinary task brilliantly. That is why we see branding as the basis of what a business promises and how it feels. The website has to deliver on that promise. Otherwise, the beautiful brand is simply disappointment in a very good outfit.
Loyalty grows when people have a reason to return. A good product, a helpful service or a relationship that holds up. A useful newsletter or customer area may do more for that than another fight over generic search terms. We have to look after that relationship, too.
Your CMS has a job to do.
“Replace your old CMS or AI will never find you” would be a comfortable sales pitch for an agency. It does not hold up.
Google lists the familiar foundations for supporting links in AI Overviews or AI Mode: a page must be indexed and eligible to appear with a search snippet. It does not require a particular generation of CMS. [3]
The technical foundation needs to make information accessible and get people where they need to go quickly. Clear structure, current content, short loading times and an interface that works on a phone as well as at a desk.
The question is what the system actually enables. Is important content accessible? Does data stay current? Can the team manage content themselves? Can new requirements be implemented without every change becoming a minor emergency?
If an old system does that reliably, its age is no reason to replace it. If it keeps preventing necessary work, we should act. A new system can solve those problems. The word “headless” on a proposal certainly does not.
Machine readability and human accessibility share some foundations. The W3C explains, for example, how meaningful HTML structure and headings help people using screen readers or keyboards navigate. [4] Human accessibility also requires adequate contrast, understandable forms and usable controls, among other things. A crawler reads text. It does not tell us whether someone can complete a purchase.
Search access and AI training are separate decisions as well. OpenAI provides independent settings for OAI-SearchBot and GPTBot. A site can allow search access while declining training use. [5]
At Silberpuls, that is why we choose the technical foundation around the task. Framer or Webflow may suit a website, a headless CMS may suit complex content distribution, and Shopify may suit commerce. Quality comes from a suitable choice and good implementation. That is part of our web development. No tool guarantees good UX, accessibility or AI visibility.
Please do not put an FAQ sign on another SEO wasteland.
Clear information matters. A useful answer should not be buried between twelve variations of the same heading.
Squeezing everything into the shortest possible FAQ would still be the next shortcut. A price needs a clear figure. A complex decision may need a proper explanation. Someone choosing a technology partner deserves more than three sentences and “Contact us now”.
For its generative search features, Google explicitly recommends useful, original content with first-hand perspectives and experience. [6] An FAQ is one possible format. It does not replace that substance.
Put the answer first, then offer the depth someone actually needs: examples, limitations, results. If a piece of content can easily be replaced by a generic AI answer, we should look more closely at the contribution it makes.
Otherwise, we are building another SEO wasteland. This time with accordions.
The agentic web needs the business behind the website to work.
Alongside this, a second job is emerging. An agent does more than answer questions. It can also act on a person's behalf: select products, prepare a cart or work through a booking process.
Shopify documents product discovery, carts, checkout and order status through the Universal Commerce Protocol, or UCP. Available actions depend on factors including access and trust tier. [2] That is concrete infrastructure. It is not yet proof that everybody will delegate their shopping.
That changes the requirements. A business could generate revenue without the buyer first navigating its shop themselves. Variants, prices, availability and delivery conditions need to be accurate. Interfaces need to work reliably and clearly limit what an agent is allowed to do.
Two jobs run alongside each other here: people visiting themselves need a convincing experience and a good route to completion. People sending an agent need reliable data and usable processes. Unfortunately, a hero video will not tell the agent which shoe size is still in stock.
Fewer clicks are no reason to give up.
Fewer information clicks demand better websites. We need to examine which visits disappear and which new ones emerge for our own offer.
That is why we build websites that do more. With a brand people consciously choose. A product worth the visit. Technology that makes content and processes accessible. And conversion work that covers the journey from first interest to completion.
We all need to step up. Including us as an agency. Test designs with real users, keep working after launch and rebuild a beautiful solution when it does not do its job.
We can earn a place in an AI answer and still lose the customer on our own website. Then we have won the visibility and blown the business.
AI needs usable information. People need a reason to become customers. Our website has to serve both. When someone clicks, we need to deliver.





