In this post
A new mediation layer is created between the user query and the website
Digital assistants, AI-powered search, and early systems for operating graphical interfaces changed access to company information in 2024. Users could get answers, compare sources, or prepare individual steps without having to search through every website completely themselves.
This did not mean that companies would replace their websites. Their role became more demanding. Content had to be so clear, structured, and substantiated that it could be read directly as well as correctly classified via mediating systems.
Starting point
Classic search had long mediated access to websites. Users formulated a search query, received a list of results, and decided which page to visit. The website remained the primary place for the actual explanation and interaction.
This division of labor was postponed until October 2024. In 2023, Microsoft had introduced an AI-powered Bing with conversational answers. Google introduced AI Overviews in search in May 2024 and expanded them internationally in October. At the end of October, OpenAI presented ChatGPT Search with direct source links. In parallel, on October 22, 2024, Anthropic demonstrated an experimental model that could operate graphical computer interfaces via screen, mouse, and keyboard.
These developments were at different stages of maturity. However, they showed a common direction: systems no longer just read pages for an index. They summarized content, combined sources, answered follow-up questions, and in initial scenarios, began to take over operational actions.
What lies behind the problem
The website is no longer always fully visible
A user can receive an answer without immediately opening the source. They can compare multiple providers via a summarized view or adopt a single detailed question from a longer expert article.
This does not make the website unimportant. But its effect can begin even before a visit. Heading, page structure, clear statements, sources, and recognizable authorship influence whether content is selected at all and conveyed meaningfully.
Machines reconstruct contexts
People can partially compensate for unclear transitions through experience. They recognize that a service page belongs to a specific target group or that a technical article qualifies a statement.
An intermediary system requires clearer signals. It processes visible text, HTML structure, links, metadata, and other technical cues. If terms are contradictory, page roles are unclear, or important information is hidden only in images, the risk of incomplete classification increases.
Summarization changes context
A website can carefully limit a statement. An external summary can shorten the sentence, connect it with other sources, or detach it from a larger argument. Companies do not fully control this representation.
The obvious reaction would be to formulate statements as generally and risk-averse as possible. However, that would also be problematic. Unclear texts are harder to summarize correctly. Precise statements with visible conditions, sources, and limits are more robust.
First action systems increase requirements
The computer use functions presented in October 2024 were still experimental. Nevertheless, they made a qualitative difference visible. An assistant could not only read but also view an interface and suggest or perform operating actions.
This brought forms, buttons, states, and error messages into a new context. What was unclearly labeled or technically fragile for humans could also become unreliable for automated operation.
Strategic Classification
Direct and indirect use must match
In 2024, companies did not need a separate website for AI systems. The reliable foundation remained a good website for people:
- Unique page roles
- semantic headings
- understandable texts
- descriptive links
- accessible forms
- visible authorship
- Sources and timeliness
- technically accessible content
These properties also supported search systems and digital assistants. The most important step was therefore not a new file format or a special AI page, but the elimination of existing structural ambiguity.
Content becomes citable units
A website should explain contexts. At the same time, individual statements also had to remain understandable outside the immediate paragraph.
This did not mean breaking texts into short answer blocks. It meant answering central questions clearly, using terms consistently, and attaching conditions directly to the statement. A citable statement is precise enough for a summary and complete enough not to seem misleading.
Evidence is gaining importance
When systems combine information from multiple sources, provenance becomes more important. Companies should make it clear who is responsible for a statement, on what experience or source it is based, and when a technical review is necessary.
A source block at the end of a post alone is not always sufficient. The contextual relationship between assertion and evidence must be recognizable.
Interactions require robust states
For forms and other functions, the proven web fundamentals continued to apply: clear labeling, programmatically recognizable fields, understandable error messages, confirmation, and secure transfer.
This structure supported people with different abilities and devices. It also facilitated technical interpretation. The assistant trend thus reinforced requirements that had long been known from accessibility and good web development.
Perspective from practice
In many corporate websites in 2024, the biggest problems were not due to a lack of AI technology. They were due to contradictory service names, unclear page hierarchies, outdated PDFs, poorly linked technical articles, and forms without a traceable process.
Such weaknesses were already disadvantageous for humans. Mediated use made them more visible. An assistant could only pick up what was accessible and interpretable. It could not reliably decide which of three different service descriptions was binding.
The practical priority was therefore:
- Clarify binding statements,
- Assign page roles,
- Structure content semantically,
- Make evidence and authorship visible,
- Make interactions robust,
- observe external mediation without chasing every product change.
The mediated website access
The graphic separates three processes. The user can visit the website directly. They can receive mediated feedback. Or an assistant can prepare a limited interaction. All three paths access the same technical and professional foundation.
Framework of action
1. Check central statements
Clearly define the scope of services, target audience, procedure, and limitations. Different pages may elaborate, but must not contradict.
2. Organize page roles and hierarchy
Use headings, navigation, and internal links to indicate which page provides an overview, which goes into detail, and which enables a specific action.
3. Make sources and authorship visible
Connect factual statements with comprehensible sources, experience, and accountability. Outdated or anonymous content weakens direct and mediated use.
4. Design forms robustly
Use understandable field labels, clear mandatory fields, comprehensible confirmations, and accessible error messages. Check the entire transfer, not just the submission.
5. Observe mediated representation
Periodically check how central questions about your company are presented in search and answer systems. Do not react with hectic optimization, but with corrections to your own knowledge and page structure.
What companies should not do
Companies should not prematurely label every website as 'agent-capable' in 2024 or treat experimental functions as a new mandatory standard. The systems were in rapid development and were inconsistent in reach, reliability, and access options.
Nor should content be written exclusively for machine extraction. A website remains a human decision-making space. Machine readability must support comprehensibility, not replace it.
Consequences for companies
Digital assistants shifted access to the website, but not the company's responsibility. The more information is mediated, the more important clear statements, semantic structure, visible sources, and robust interactions become.
The strategic task in 2024 was to treat one's own website as a reliable source and not just a landing page. Later agentic applications could build on this foundation. They did not have to be anticipated.
Subject-matter connection
Check website for direct and indirect use
A website strategy assesses whether content, page roles, evidence, and interactions are understandable for people and intermediary systems. SDC-Discovery creates a prioritized basis for this, without placing experimental technology before professional order.
Classify website structure and mediated usage
Sources and technical foundations (7)
- Microsoft, Reinventing search with a new AI-powered Microsoft Bing and Edge, your copilot for the web, February 7, 2023. Open source
- Google, Generative AI in Search: Let Google do the searching for you, May 14, 2024. Open source
- Google, AI Overviews in Search are coming to more places around the world, October 28, 2024. Open source
- Anthropic, Developing a computer use model, October 22, 2024. Open source
- OpenAI, Introducing ChatGPT search, October 31, 2024. Open source
- World Wide Web Consortium, Headings, Web Accessibility Initiative. Open source
- World Wide Web Consortium, Forms Tutorial, Web Accessibility Initiative. Open source
