A chief executive recently told me that nobody looks at websites anymore: “I don’t read them,” he said. “I use ChatGPT.”
It is an odd but interesting view. Why navigate a manufacturer’s corporate website, open six tabs and download three PDFs when an AI assistant can compare suppliers, summarise their capabilities and produce a shortlist in seconds?
The observation is directionally right. The conclusion is dangerously wrong.
People may visit fewer websites in future. But that does not mean websites become less important. It means their role changes. The corporate website is evolving from a destination people must navigate into a body of evidence machines can retrieve, interpret and use.
The question is no longer simply: will a buyer find us?
It is: will an AI system understand us well enough to recommend us?
Where does an AI answer come from?
Ask an LLM which European manufacturers can design, industrialise and produce a complex mechatronic module, and the answer may feel remarkably informed. It may identify companies, distinguish their capabilities and even offer a view on which one appears best suited to the brief.
But the model did not acquire this knowledge through industrial experience. Nor did it independently form an opinion in the human sense.
Its response may draw on several layers of information: knowledge acquired during model training; live web search and retrieval; search-engine indexes; structured databases; third-party publications; and, where permitted, private files or connected company systems. The model then organises that material, identifies patterns and generates a conclusion.
What sounds like an opinion is usually a synthesis or inference. And the quality of that inference depends heavily on the quality of the available evidence.
This is why withdrawing from the web because buyers increasingly use AI would be like removing your company from industry directories because procurement teams use databases. You would be eliminating the source material at precisely the point machines are becoming more influential in the buying process.
OpenAI makes the connection explicit. Its OAI-SearchBot exists to surface websites in ChatGPT search answers; sites that block it will not appear as sources in those answers, except potentially as navigational links. Importantly, OpenAI separates search inclusion from model training, allowing a company to permit search crawling without granting the same permission for training. OpenAI’s crawler documentation makes this distinction clear.
No website, or a thin, vague and technically inaccessible one, means less first-party evidence for an AI system to retrieve.
Your website is becoming evidence infrastructure
For the past 20 years, many B2B websites have been designed around a familiar journey: attract a visitor, explain the proposition, capture a lead.
That journey will not disappear. Long-cycle B2B decisions still involve people who need reassurance, technical scrutiny and organisational confidence. An engineer, procurement leader or potential employee may arrive after an AI recommendation and want to examine the company behind it. Essentially, website visits are a few steps later in the user journey than they were in the past.
But the website now has another audience: retrieval systems.
This changes what good content looks like. Broad claims such as “engineering tomorrow” or “your trusted innovation partner” provide little material from which either a person or a machine can make a meaningful judgement. They communicate aspiration without supplying evidence.
An effective content footprint answers much more specific questions:
What can the company actually make? At what tolerances, volumes and levels of cleanliness? Which responsibilities can it assume? What standards does it meet? Which markets does it understand? How has it moved a product from prototype to repeatable production? What did it solve for a customer, and what was the measurable result?
For complex B2B businesses, the strongest website may become less like a digital brochure and more like a well-organised technical library: capability pages, applications, case studies, engineering perspectives, certifications, product data, manufacturing locations and named subject-matter experts. It becomes your brand identity.
This does not mean publishing every piece of intellectual property. It means making the company’s competence visible.
AEO is not a bag of new tricks
Answer engine optimisation, generative engine optimisation and AI search optimisation are all competing labels for the attempt to improve visibility inside generated answers. Inevitably, a small industry has appeared promising special files, magic schema and techniques for “ranking in ChatGPT”.
Much of this overstates what is known.
Google says the fundamentals of SEO continue to apply to AI Overviews and AI Mode, with no separate technical requirements or special AI markup needed. Pages must still be crawlable, indexed and eligible to appear in conventional search. Google recommends internal linking, accessible text, a good page experience and structured data that accurately reflects the visible content. Google Search Central is refreshingly unromantic about it.
The emerging evidence does, however, suggest that the way information is written matters. Research into generative engine optimisation found that clear sourcing, relevant quotations and statistics could improve visibility in generated responses, while keyword stuffing performed poorly. The research reported improvements of up to 40% in its experimental settings, although results varied by subject and system. The GEO research paper should be treated as evidence of direction rather than an eternal ranking formula.
Microsoft’s Bing Webmaster Tools now measures citations in AI-generated answers and the queries used to retrieve them. Its own guidance encourages depth, clear headings, tables, FAQs, supporting evidence and current information. Microsoft’s AI Performance guidance sounds suspiciously like good publishing—because that is largely what AEO is.
From ranking pages to supplying answers
Traditional SEO encouraged companies to think in keywords and ranked pages. AI discovery encourages them to think in questions, evidence and entities.
A specialist manufacturer should not merely optimise a page for “precision engineering”. It should publish the material required to answer the real questions buyers ask: Can this company maintain micron-level tolerances at production volume? Does it have cleanroom assembly? Can it support design for manufacture? Has it managed a technology transfer between sites? Which engineers have experience in the relevant discipline?
The advantage belongs to the company that offers the clearest body of connected, verifiable knowledge, not necessarily the one that publishes the most content.
That knowledge should not live exclusively inside glossy PDFs, sales presentations, employees’ heads or disconnected LinkedIn posts. It needs a durable, crawlable home. Third-party coverage, customer references and authoritative external citations then corroborate the company’s own claims.
In other words, the content footprint extends beyond the website, but the website remains its centre of gravity.
The website after the website
Ten years from now, many buyers may never browse a corporate website from the homepage down. An AI agent may conduct the first stage of research, compare suppliers and retrieve only the fragments relevant to the task.
Yet when the shortlist becomes serious, people will still want proof. They will examine the factory, the people, the record, the detail and the risks. In a high-value sale, confidence cannot be generated by a paragraph of synthetic text alone.
The winning B2B website will therefore perform two jobs simultaneously. It will give machines information they can accurately retrieve and give people reasons to believe it.
So, does a company still need a strong website to come out on top in AI-driven discovery?
Yes, but not simply because people will keep visiting it.
It needs one because the website is becoming the company’s public knowledge base: the place where its capabilities are defined, its claims substantiated and its expertise made legible to both humans and machines.
The CEO may be right that he no longer reads websites.
ChatGPT, however, still needs something worth reading.
