The usual argument goes like this: AI now answers most questions directly, so fewer people click through to websites, and therefore websites matter less. 

That’s not entirely true. Instead, there’s a split happening. A website used to carry a buyer through an entire decision, from awareness through conversion. Now the early research, the "what's out there," "is this worth it" stage, happens inside a chat interface before a buyer ever reaches a domain. What's left for the website to do is create confirmation for the buyer who's already narrowed the field and wants to check the product’s fit before booking a call or filling out a form.

This is part of what Alessia Sannazzaro, co-founder of Code & Wander, and I discussed on a recent episode of Building the Next Web. A site that used to do everything now only needs to do one job well, and a growing number of brands are responding by questioning whether they need a site at all. That's backward. As Alessia put it: without a website, you lose a chunk of control over your brand’s message in citations, and when a buyer eventually does seek you out and “click” there’s nothing there for them to grab onto.

That's the framing I’d like to build from, and it's why the real question is what the website's job becomes now that AI has already taken over the top of the funnel.

Why AI systems can’t cite what doesn’t exist

Large Language Models increasingly target sources deliberately instead of skimming the open web. Reporting from Search Engine Journal on a recent shift in ChatGPT Search found that the model's background searches started scoping heavily to individual domains, going to specific sites directly rather than pulling broadly from wherever content happened to surface.

That change penalizes brands whose story is scattered across old campaign pages, outdated service descriptions, and messaging that shifted 18 months ago without anyone updating the site to match. If a model can't find a consistent, extractable answer on your own domain, it either guesses, pulls from a third party who may not represent you accurately, or leaves you out of the answer entirely.

A brand is AEO-ready when its site says the same specific, defensible thing about itself everywhere a person or a model goes looking, clear enough that an AI system can summarize it accurately in a single sentence.

What happens when your visibility depends on someone else’s platform

In August 2026, Reddit's share of citations in ChatGPT Search fell from an average of 3.83% to 0.52% in about a week, an 86% relative drop, according to tracking data from Promptwatch. The timing lines up with the same fanout change mentioned above: ChatGPT started scoping far more of its background searches to specific domains on August 8, and Reddit's citation share cratered days later.

What’s notable is that nobody, including the firm that measured it, is fully certain why it happened. 

If a large chunk of your AI visibility strategy depends on earning mentions on Reddit, Quora, or any platform you don't own, you're exposed to changes you'll never see coming and can't influence when they land. A site you control doesn't carry that same risk. It can still lose visibility, but for reasons you can act on.

The CRO shortcut that skips the problem

There's a version of this problem that shows up before a brand ever thinks about AI citations. A marketing team notices conversion rates are soft, so they strip fields from a form, add autofill, and tighten the copy on a call-to-action button. The number moves and everyone feels good for a quarter, but none of that addresses whether the page underneath the form is saying something clear, true, and consistent with the rest of the site. 

A shorter form doesn't fix a brand that describes itself three different ways across three different pages. It just speeds up the wrong-fit leads and does nothing to help an AI system summarize the company correctly. What the company does, who it's for, and what makes it different have to be settled before any funnel optimization means anything. Skip that step and you're polishing a page a model still can't accurately describe.

Confirmation is a narrower job than persuasion, and most sites are still built for the wrong one

Once the early research shifts to AI chat interfaces, the site's remaining job is confirmation, not discovery. That changes what “good” looks like. A homepage doesn't need to convince a cold visitor from zero. It needs to hold up under a closer look from someone who already has a shortlist, with claims that are specific enough to be checked and consistent enough to survive a visitor reading three different pages back to back.

This is the same problem seen from two directions at once: a site has to be clear enough for a model to summarize accurately and credible enough for a skeptical buyer to trust. A brand clarity assessment is usually the right starting point, since it surfaces exactly where a brand's story diverges from page to page before that inconsistency shows up as a bad AI citation or a lead who drops off mid-form. 

From there, a Webflow AEO audit can show where the site's structure is helping or hurting extraction, and Edgar Allan’s Visibility Engineering and Optimization methodology ties brand, SEO, AEO, and CRO together so none of them get optimized in isolation. Clarity first, structure second, a system that keeps both aligned after that.

So the next time someone suggests cutting the website budget because AI does the research now, the useful response is to ask a better question: if a model pulled up your homepage right now and had to summarize what you do in one sentence, would it get it right?

If the answer is no, or if it depends on which page the model happens to land on, that's the problem you need to solve before any CRO tactic gets touched. Start with whether the story holds up. Everything downstream depends on it.