The End of the Homepage as the Front Door

The Visitor Who Already Made Up Their Mind

For thirty years the homepage was the front door. Customers typed your name, clicked a link, and the first few seconds decided whether they stayed. So we optimized for that moment: the hero image, the headline, the emotional hook, the carefully staged proof. The homepage was a handshake, and the whole discipline of web design was the science of a good first impression.

That sequence is breaking. Increasingly the first thing to read your homepage is not a person deciding whether to trust you, but a model deciding how to describe you. By the time a human arrives, an AI system has already read your site, cross referenced it against other sources, and handed the person a summary. The customer does not walk up to your front door anymore. They show up having already been briefed by a machine, and the brief was written without your permission.

This is what we call the Pre-Read, and it quietly inverts the purpose of a homepage. The page is no longer where the decision happens. It is the material the decision is made from. When an AI agent reads your business before any human does, your homepage is no longer a destination: it is evidence.

"It's 2025 and most content is still written for humans instead of LLMs. 99.9% of attention is about to be LLM attention, not human attention."

Andrej Karpathy, AI researcher, founding member of OpenAI and former Senior Director of AI at Tesla. Andrej Karpathy on X, 2025-03-12.

From Impression to Legibility

An impression is something you make on a person. Legibility is something a machine grants you, or refuses to. The two are not the same skill, and a business can be brilliant at one while failing the other. A page can be gorgeous, persuasive, and award winning, and still be an opaque smear of ambiguity to the system tasked with explaining it in one sentence.

Andrej Karpathy, the former director of AI at Tesla and a founding member of OpenAI, put the underlying shift bluntly in March 2025: "It's 2025 and most content is still written for humans instead of LLMs. 99.9% of attention is about to be LLM attention, not human attention." He was talking about software documentation, but the logic generalizes to every business with a website. If the next reader of your page is overwhelmingly a model, then the question is no longer how the page feels. The question is how cleanly it can be parsed, extracted, and repeated.

This is the heart of the shift. In an AI-mediated market, the first impression is not made by your design: it is made by the summary a model writes about you. You can no longer control that summary by controlling the visit, because the visit now happens upstream, in a context you never see. You can only influence it by being legible enough that the machine has nothing to guess at.

The Homepage as Source Material

Treat your homepage as a press release written for one reader who will never thank you and will paraphrase you to everyone who matters. That reader does not care about your gradient or your animation. It is looking for answers to a fixed set of questions: who is this, what do they do, who do they serve, where, and can any of it be corroborated. Everything that does not help answer those questions is, to the model, noise to be discarded.

Source material has different virtues than a destination. A destination wants to delay you, to keep you scrolling, to build a mood. Source material wants to be quoted accurately and fast. The best source material states its claims plainly, attributes them, and never contradicts itself across pages. When your site behaves like good source material, the model's summary of you starts to sound like the summary you would have written. When it behaves like a mood board, the model fills the gaps with inference, and inference is where misrepresentation begins.

This is why the old instinct to be clever in your copy now carries a tax. Wordplay, vague taglines, and aspirational abstractions used to signal sophistication to a human skimming for vibes. To a machine compressing you into a sentence, they read as missing data. The business that writes for the Pre-Read says what it is in language a model can lift without distortion.

"We think of agents as systems that combine the intelligence of advanced AI models with access to tools, so they can take actions on your behalf and under your control."

Sundar Pichai, CEO, Google and Alphabet. Google I/O 2025: Sundar Pichai's opening keynote, 2025-05-20.

What the Machine Actually Reads

When an AI system pre-reads your business, it is not admiring your layout. It is resolving a series of questions about clarity and consistency, and our Trust Visibility framework names them as six pillars across thirty one signals. Semantic Clarity asks whether your purpose is stated plainly enough to extract. Entity Consistency asks whether the name, category, and facts on your homepage match what appears everywhere else the model can see you. These two pillars alone decide whether the machine can build a stable picture of you at all.

The other four pillars decide whether that picture is worth repeating. Authority asks whether anyone credible corroborates your claims. AI Discoverability asks whether your structure and markup let a machine find and read the answers without struggle. Trust asks whether the signals around you reduce the model's risk in recommending you. Local Presence asks whether you are anchored to a real, verifiable place in the world. A business can pass the first two pillars and still be passed over on the last four.

The point of measuring across pillars is that legibility is not one thing you fix once. It is a property that degrades quietly: a renamed product here, an outdated address there, a claim on the homepage that no third party backs up. Each gap is a place where the machine substitutes its own guess for your fact, and you never find out, because the substitution happens in the Pre-Read, before anyone visits to tell you they were confused.

Why Impression-Era Tactics Now Backfire

The reason this matters now, rather than someday, is that the reading is turning into acting. At Google I/O on May 20, 2025, Sundar Pichai, the CEO of Google and Alphabet, defined the direction this way: "We think of agents as systems that combine the intelligence of advanced AI models with access to tools, so they can take actions on your behalf and under your control." An agent that takes actions on a person's behalf does not browse your homepage for inspiration. It reads you to decide whether to include you in a shortlist the human may never question.

Once a machine is filtering on the customer's behalf, the tactics built for human persuasion start to work against you. A splash page that hides your offering behind a video, a clever name that never says what you sell, a wall of testimonials with no verifiable source: these once bought you a few seconds of intrigue. To an agent under time pressure, they buy you exclusion, because the agent cannot afford to recommend what it cannot confidently describe.

So the cost of illegibility is no longer a slightly worse impression. It is absence from the consideration set entirely. Legibility is the new conversion rate: a business that a machine cannot summarize cleanly is a business it will not recommend. You do not get to make a second impression on an agent that already decided you were too ambiguous to pass along.

Trust Becomes the Thing You Optimize

When the reader is a person, trust is built over time, across a visit, through tone and design and social proof absorbed gradually. When the reader is a model assembling a summary, trust has to be present in the source itself, because there is no relationship to build. The machine is not persuaded. It is calculating risk, and it lowers that risk by finding consistency and corroboration, not warmth.

This reframes what optimization even means. You are no longer optimizing the experience of a session, because the session that matters is one you cannot observe: the model reading you last week to brief someone today. You are optimizing whether the account of your business survives compression and corroboration intact. The work moves from making people feel something to making a machine able to vouch for you without hedging.

That is the deepest part of the inversion, and it is why we treat Trust Visibility as a measurable discipline rather than a branding exercise. Impression was always partly subjective, defensible by taste. Legibility is not. Either the model can extract a clean, consistent, corroborated account of you, or it cannot, and that outcome is testable. You no longer optimize for the visitor who lands on your page: you optimize for the model that read it last week.

The New Front Door Is the Summary

If the homepage is no longer the front door, something else is, and it is the summary itself: the few sentences an AI system generates about you before a human encounters you. That summary is now the storefront, and you do not own it. You can only shape it by controlling its inputs, which means making every legible surface of your business agree with every other one. The front door moved into the model, and the only way to influence it is to be impossible to misread.

This is not a counsel of despair for design. Good design still matters for the humans who do arrive, and it always will. But its job has changed. Design used to do the convincing. Now the summary does the convincing, and design's job is to confirm, for the human who clicks through, that the brief they were given was accurate. The homepage that wins the next decade is the one a machine and a person describe the same way.

The businesses that thrive in AI-mediated discovery will be the ones that stop asking how their homepage looks and start asking how it reads to something that will never look at all. That is the whole move: from impression to legibility, from destination to source, from owning the visit to earning the summary. The front door did not disappear. It moved somewhere you cannot decorate, and the only key is being legible enough that the machine has no reason to get you wrong.

Legibility is the new conversion rate: a business that a machine cannot summarize cleanly is a business it will not recommend.