Marketing knowledge you can trust
Tell Google you read us – and we’ll show up more when it matters.
Your website is one of the few channels that you actually own. Everything else is rented: reach on Google, space in an AI response, attention in a feed, clicks in an ad account. Someone else sets the prices and the rules there. On your own site, you set them yourself—and at the same time, you provide the raw material from which all other channels are now built.
That’s why working on your own website in 2026 pays off differently than it did three years ago. It’s no longer just the destination that all channels point to. It’s the source they draw from.
This article summarizes what a website must do today to work effectively for people, search engines, AI systems, autonomous agents, and paid campaigns all at once.
The good news up front: There aren’t five tasks. There’s just one.
In this Article
Your Website as the Hub
We’ll build you a website that really works in the age of AI.
Three visitors are reading the same page
Imagine three visitors—who couldn’t be more different—browsing your products.
People scan. They read headlines, look at images, skip around, and decide within seconds whether to stay. They need guidance, evidence, and a clear next step. They stumble upon clichés, walls of text, hidden prices, and forms with dozens of fields.
The crawler reads the code. It scans the page, breaks it down into text sections, and evaluates whether they can be quoted or, in the case of Google and Bing, displayed in search results. It needs high-quality content in the rendered HTML, a clean hierarchy, and paragraphs that stand on their own. It stumbles over content generated by JavaScript, a sea of containers, and a messy hierarchy.
The agent doesn’t want to read at all. It wants to take action: compare, book, submit, add to cart. He needs labeled fields, stable elements, and a path that leads to the goal without endless clicking. He stumbles over CAPTCHAs, consent layers, and unlabeled input fields.
If you focus on just one of these three, you might lose the other two. But here’s the beauty of it: Almost everything that helps one also helps the other two. More on that in a moment.
Rankings and clicks have become two different things
About 20 percent of German search queries now display an AI summary. When it appears, the click-through rate for the top result drops from about 27 percent to 11 percent—as measured by SISTRIX across more than 100 million German keywords. For transactional searches, the rate remains largely unchanged; people who want to buy or hire a service continue to click.
In practical terms, this means that visibility without a click has become the norm. How-to content is losing traffic, even when it ranks well. The value of a page is shifting from “brings visitors” to “is citable and contributes to the decision-making process.” The money is where it’s always been, just more visible than before: on service, comparison, price, and review pages. How-to guides remain important—as proof of expertise, not just as pure traffic generators. Our SEO expert Corinna has written about this in more detail in her post on zero-click searches.
Being cited works differently than ranking
An AI response is generated in three steps. A single question is broken down into many search queries (grounding queries). Instead of web pages, the system collects text excerpts. From these excerpts, the system formulates a response and cites sources.
This leads to a rule that transforms half of the editorial work: Every paragraph should be able to stand on its own. A section that requires a “see above” reference will not be cited. Feel free to spell out the subject instead of using “it” or “the.” The brand name can be repeated (just not in every sentence).
On top of that, there’s a hard technical fact that surprises many: GPTBot, ClaudeBot, and PerplexityBot do not execute JavaScript. They only see what the server returns. Unfortunately, this is precisely the issue we’ve seen more and more in new client projects over the past few months.
Google renders, Bing renders, and Gemini renders on direct access—many others do not. So anything that only appears after a click or reload simply doesn’t exist in the AI world. And: These crawlers end up on error pages much more frequently than Google does. Broken links cost more here than they used to.
The third visitor doesn't want to read your website, they want to take action
An agent is an AI that performs actions on websites on behalf of a human. That might sound like something from the distant future, but it’s already well underway: payment gateways from OpenAI, Stripe, Google, and Shopify are live, and some purchases are already being completed right in the chat. In e-commerce, the product data feed is the key—without structured product data, an agent cannot purchase a product.
What reliably causes agents to fail today is surprisingly trivial: forms that can only be submitted via JavaScript and have no submission target in the HTML. Unlabeled fields. Radio buttons instead of actual dropdown menus. CAPTCHAs and hidden honeypot fields in the main path. Appointment booking behind the consent layer. And the phone as the only way to complete a transaction.
If an agent can’t submit your contact request, a person using a screen reader usually can’t either.
This sentence epitomizes the core of this article. It connects tomorrow’s technology with today’s legal obligation—and with a simple conversion question. More on that in a moment.
Create and Develop Websites with WEVENTURE
We build websites that work for people, search engines, and AI systems alike—from structure and design to development and analytics. And we’ll tell you upfront what makes the biggest difference.
One initiative, five channels
Now let’s get to the heart of the matter. Almost none of the measures listed below have an impact in just one area. That’s exactly why the effort pays off.
| Measure | SEO | AI / GEO | Agents | SEA | Social |
| Content in the delivered HTML | ✓ | ✓ | ✓ | ✓ | – |
| Custom title, one H1, meta description | ✓ | ✓ | – | ✓ | – |
| Clean heading hierarchy | ✓ | ✓ | ✓ | ✓ | – |
| Semantic, lean HTML | ✓ | ✓ | ✓ | ~ | – |
| Structured data and feeds | ✓ | ~ | ✓ | ✓ | ✓ |
| Labeled, submittable forms | ✓ | – | ✓ | ✓ | ✓ |
| Price and provider transparency | ✓ | ✓ | ✓ | ✓ | ✓ |
| Do not block crawlers and AdsBot | ✓ | ✓ | ✓ | ✓ | ✓ |
| Clean event and consent architecture | ✓ | ✓ | – | ✓ | ✓ |
This is best illustrated using an input field. First, a field with a correctly linked label is a requirement for digital accessibility and will therefore be a legal obligation for many companies starting in June 2025. Second, it’s what a screen reader reads aloud. Third, it’s what an agent needs to fill out the field correctly in the first place. Fourth, it helps anyone using autofill finish faster—which lowers the abandonment rate. And because more submitted forms mean more accurate conversion signals, that same label can ultimately improve the performance of your Google Ads campaigns.
One budget, five results. This isn’t a marketing calculation—it’s simply the result of all systems accessing the same file.
A Website That Works
From web development and design to SEO and content optimization to paid ads: We’re your full-service agency, ready to turn your website into a sales machine by 2026.
Optimizing a Website for People, Google, AI, and Agents: Where to Start?
The order is more important than completeness. Without the first two points, nothing else matters.
1. People, crawlers, and AI agents must be able to access your website
Before anyone can understand it, they have to be able to access it. Your hosting must work, and your site must be accessible—to both people and crawlers. In your robots.txt file, block only what really needs to be blocked, and explicitly allow AdsBot-Google. Then check the server level: firewalls, bot protection from your hosting provider, and CDNs often block crawlers even though robots.txt allows them. These blocks take effect before the CMS, which is why you won’t see them in any SEO tool.
The fastest test in the world: Request your homepage using the identifiers for GPTBot, ClaudeBot, PerplexityBot, Googlebot, and AdsBot, and check the status code. 200 or 403? This reveals more in five minutes than half a tool audit.
2. The most important content on your website must be in the HTML
If only one rule from this article is to survive—especially when it comes to optimizing for ChatGPT, Claude, and the like—then let it be this: Everything important must be delivered ready-to-view from the server—no clicks, no consent, no JavaScript. JavaScript can enhance the experience, but it should never be the only way to access key information. Accordions and tabs are fine as long as the text is embedded in the HTML. Images need a real `src` attribute, not a JavaScript placeholder attribute; otherwise, they don’t exist for image search or AI.
And then there’s the point that’s new to most projects: the structure of the code itself. Website builders generate three containers for every heading. You can’t see it, but it takes up valuable space. In client projects, we regularly measure homepages with 450 KB of code for 1,000 words of text—a text ratio of less than 2 percent. To an extractor, a page like this looks like a skeleton without a building.
Here are a few target metrics to include in the briefing: fewer than 800 DOM elements in the body, a maximum of about 30 nesting levels, at least 3 percent text content relative to the code, exactly one `main` element per page, and `article` elements used only for truly standalone content units. In addition, use the correct elements instead of containers for everything: bulleted lists as lists, comparisons as real tables with labels, and counterarguments as expandable elements. And include navigation only once—repeated menu structures are nothing but noise for extractors.
3. Language That Stands on Its Own
This is what determines whether you’ll be cited. Definitions should be formulated as definitions: term, colon, one-sentence explanation. Anything with “vs.” or “difference” belongs in a table. Numbers need a source—verified numbers measurably increase the likelihood of being cited, while superlatives do not. What you can do without, however: word count targets, keyword density, and an authoritative tone without substance. And hidden text intended for AI systems is a violation of guidelines and a security risk.
Two building blocks that are often overlooked because they’re invisible: alt text and ARIA labels are the only places where you can provide context that isn’t included in the visible text—without overloading the page. An icon link without text is equally silent to humans, machines, and agents; an `aria-label` solves this in a single line. The only important thing is the hierarchy: Visible text always takes precedence; ARIA supplements where text is missing.
And, very importantly, your brand must be recognizable as a single entity. Use a consistent spelling everywhere—every variation creates a separate entity. A fixed organizational node in the code that everything references. Links to all of your own profiles. Authors as real people with their own profiles, not as “Editorial Team.”
One more honest take on structured data: It’s essential for clarity and SERP display, but the best available test found no positive effect on AI mentions. Mark it up properly, yes; sell it as a visibility silver bullet, no. The same goes for llms.txt: it costs almost nothing, but so far has had virtually no demonstrable effect.
4. Website Forms: CRO for People and AI
The website form requires a slightly modified approach to conversion rate optimization.
First and foremost: Every form field must serve a purpose, because every additional field reduces the chance of a good conversion rate. So: Anything that will be asked during the initial consultation anyway doesn’t need to be included in the form. Each field gets a real, visible label—placeholder text disappears when you start typing, and autocomplete on standard fields makes filling out the form faster. Errors appear directly next to the field without losing any input.
For this same approach to work for agents as well, three additional attributes are needed: a real `action` target in the HTML, real dropdown lists instead of buttons that write to a hidden field, and a main path without a CAPTCHA. Security measures belong on the server side, not in the user’s path. Incidentally, these are exactly the same points required for accessibility.
This includes accessible completion paths: a phone number as a genuine `tel:` link with an international number, a visible email address, and an appointment link that’s accessible even without consent to the cookie banner. External calendar integrations may be blocked—but the path to completion itself must not be. Otherwise, you’ll lose everyone who declines.
And then there’s tracking, which many tackle too late: Define three to five KPIs, assign them a value, and only then make them technically measurable. Our tracking expert Lara explains how to set this up properly in her article on website tracking and analytics.
5. Pace, Rules, and Hygiene
PageSpeed is still particularly important for Google rankings, since these are heavily influenced by user experience. However, it has less of an impact on mentions in ChatGPT, Claude, Perplexity, and similar platforms (unless your PageSpeed is so poor that the crawlers can’t properly read the page).
Nevertheless, metrics such as LCP, INP, and CLS should be included in the briefing. LCP under 2.5 seconds, INP under 200 milliseconds, CLS under 0.1. A maximum of two font families. Images in AVIF or WebP, with dimensions specified in the HTML. No more than ten third-party script domains, all documented. And testing is performed automatically with every deployment, not just once a year manually.
Legally, the Accessibility Enhancement Act has been in effect since June 28, 2025. It applies to e-commerce services provided to consumers, with an exception for micro-enterprises with fewer than ten employees. Penalties include fines, injunctions, and cease-and-desist letters under competition law. The measures with the greatest impact are unspectacular: accessible labels for all input fields, sufficient contrast, keyboard accessibility with visible focus, clickable areas of at least 24 by 24 pixels, and descriptive link text. By the way, overlay widgets don’t solve this—they just hide it.
That leaves the “behind-the-scenes” elements that no one sees but that scale with volume: exactly one H1 per page, a unique title and description for each URL, no internal links to redirects or error pages, a complete sitemap, and complete social preview images. For a website with a thousand URLs, that quickly adds up to a hundred pages with the same error. And since Google generates ad copy from titles and descriptions, this is no longer just a cosmetic issue. Setting aside a fixed monthly time slot for this is more effective than one big cleanup effort per year—much like internal linking, which only works if it’s maintained.
The Underestimated Lever for Your Website: Say What It Costs
A lack of cost transparency is one of the most common reasons for abandoning a project. What’s new is that it now comes at a triple cost. For people, it leads to them walking away. In AI responses, the question “How much does X cost?” is answered by the providers who give a specific figure—those who don’t are excluded. And an agent who’s comparing options needs a specific figure; “upon request” is simply a deal-breaker for them.
This can be resolved without risk. Use price ranges instead of fixed prices: “Projects between X and Y,” “Starting at Z.” Provide a calculation method that explains what drives the price—scope, languages, competition, intensity. And specifying a lower limit even protects against inappropriate inquiries: “Below X, it doesn’t make sense because …”. If your business model really doesn’t yield a specific number, that’s fine—but it should be a conscious decision, not a gap. The middle ground is to explain the process and the price drivers without naming a specific number.
Why the Perfect Website Often Fails in Everyday Use Anyway
The list above isn’t a secret formula. Still, we rarely see it implemented in its entirety—and the reason almost never lies in skill, but rather in the process.
In most setups, each page is created twice: once as a draft in the design tool, and once as a replica in the builder. The same design decision is made twice; changes made after approval must be applied in two places; and every newly built page introduces new container levels and new dependencies. The overhead mentioned in the previous chapter is no coincidence—it’s the direct result of this workflow. Quality then depends on the person who is currently building—not on the system.
The solution is unspectacular and takes effect immediately: a pre-approved building block consisting of thirty to fifty sections, in which the rules from this article are already hard-coded. In the design phase, you simply assemble what’s available in the building block. Quality is determined once and for all, rather than being renegotiated for each page. The price for this is creative freedom on a per-page basis—and a team must be willing to pay that price consciously. Those who think ahead cast the blueprint into their own components and turn the checklist into a default setting. For projects with a lot of content or multiple distribution channels, it might also be worth looking into headless CMS.
And AI? It’s genuinely helpful in this context, but not to the same extent in every area. It’s reliable for structuring and organizing content based on a briefing, for drafting and creating variations of text, for producing alt text in bulk, for generating and verifying structured data, for migration and initial translation, and for visual comparisons and link checks. With well-organized drafts, a time savings of around 40 to 50 percent in the initial development phase is realistic—the remaining thirty percent still requires manual work, and the gaps are always the same: semantics, accessibility, adherence to the design system, and state logic. No tool today delivers code that’s ready for publication. Anyone who promises that is just selling a demo.
Where would you start?
If you want to know how your website measures up in these areas, we’ll take a look at it—crawler access, structure, forms, analytics, and accessibility. You’ll receive a prioritized list of actions rather than a list of tools.
Conclusion: Three Takeaways About the 2026 Website That Will Stand the Test of Time
It’s one task, not five. Being clear and trustworthy benefits users, SEO, AI responses, agents, and your paid channels all at once. You don’t have to build anything five times.
The content must be in the HTML. Everything important comes ready-made from the server—no clicks, no consent, no JavaScript. If only one rule survives, let it be this one.
Quality belongs in the system, not on a checklist. As long as every page is created by hand, you’ll lose that standard with every new project.
And the statement from the beginning holds twice as true in the end: Your website is the only asset that truly belongs to you. It’s the only place in marketing where care isn’t rented but built—and where it continues to work for years to come.
FAQ: Optimizing a Website for People, Google, and AI Agents
Will I even need my own website in 2026?
Yes, more than ever. Every other channel is rented: someone else determines the reach, the rules, and the prices there. At the same time, your own website is the source from which search engines, AI systems, and even ads draw their content. It doesn’t replace any channels—it provides them with input.
What is the difference between SEO and GEO?
SEO aims to rank in search results. GEO (Generative Engine Optimization) aims to be cited as a source in AI responses from ChatGPT, Perplexity, Gemini, Copilot, or Google’s AI mode. The technical fundamentals are nearly identical. The difference lies in the text: For GEO, each section must be understandable on its own, because AI systems collect sections, not pages.
Why isn't it enough for the page to look good in the browser?
Because most AI crawlers do not execute JavaScript. They only see the HTML that the server returns. Content that appears only after a click, consent, or reload does not exist for them. This is less of an issue for Google, since Google renders the content—but it is crucial for ChatGPT, Claude, and Perplexity.
What are agents, and do I need to start thinking about them now?
Agents are AI systems that perform actions on websites on behalf of people: comparing, filling out forms, booking appointments, and making purchases. The good news is that you don’t have to build anything extra to make this work. Everything agents need—labeled fields, submitable forms, real links, and structured product data—is already required for accessibility, conversion, and search.
Does accessibility really help my SEO?
Indirect, but clear. Accessible names, clean heading hierarchies, descriptive link text, alt text, and semantic elements all form the foundation that enables search engines and AI systems to correctly categorize a page. The tree structure that a screen reader uses to read a page is the same one that an agent uses to process it. One budget, multiple results.
How fast does my website need to be?
The target metrics are a loading time of less than 2.5 seconds for 75 percent of real users, a response time of less than 200 milliseconds, and layout shifts of less than 0.1. Speed is a prerequisite, not a growth driver: a fast page without a clear message won’t sell anything. More important than the final point in the score is that these metrics are automatically checked with every deployment.
Does Structured Data Enhance the Clarity of AI Responses?
As things stand today, this cannot be measured. The best available test found no positive effect on AI mentions. Structured data remains mandatory nonetheless: for clarity, for display in search results, for shopping feeds, and for agent-based purchases. The only important thing is that it appears in the server-side HTML and matches the visible text.
Do I have to list prices on my website?
You don’t need to give a specific number, but you do need to provide some guidance. A range, a clear calculation method, or a stated minimum are sufficient. Without any indication of cost, potential customers will lose interest; AI responses will name the competitors who provide a specific number, and agents comparing options will exclude you. If you deliberately choose not to give a number, at least explain what determines the price.
Is a relaunch worth it, or is a tweak enough?
This almost never comes down to design, but rather to structure. If the content is in HTML, the hierarchy is correct, and the page types are clearly defined, you can get a long way by making adjustments. If, on the other hand, each page was built individually and the quality depends on the person who built it, a new content management system is the more cost-effective option—even if it seems more complex at first.
