Reimagining an AI-native website, on both the front-end and the back-end

For a brief moment, when everyone was crushing hard on Claude and ChatGPT, declarations that “the web is dead” ricocheted across LinkedIn and X. (We do seem to love that X-is-dead cliché.) Why would anyone need to visit a website ever again? What would be its raison d’être in an AI world?

Well, the initial answer was pretty obvious. Where are those AI answer engines going to source their answers from? Reddit and LinkedIn only provide some of the angles on companies, products, and services. For many questions, you want authoritative information straight from the horse’s mouth. So AEO — answer engine optimization — breathed new life into the oh-so-briefly deceased.

But it turns out that’s only part of the story. Especially with the rise of personal agents — Instinct, Meta Muse, OpenAI Dots, Claude Duet, Town, et al. — it’s clear that companies will need a digital surface to serve those agents too. That surface has to live at a domain, with addressable URLs, associated with human-readable content when someone wants to be in-the-loop to verify or approve their agent’s recommendation.

It doesn’t take a lot of squinting to see that as a website. Granted, a new kind of website, serving a new kind of audience. Even with new kinds of protocols and functionality. But still a digital presence “home” for a business or a brand.

However, perhaps the most surprising revival is the realization that people do still want to connect directly with businesses and brands. Experience matters. And the experience within a third-party AI assistant is the experience of that AI — not the experience of the brands referenced within it.

AI assistants are a lossy channel, compressing brands into a more generalized experience. That’s fine — even desirable — for certain decisions or stages of a buyer’s journey. But when you want direct answers from a business, or you want direct interaction with one of their services, or you want a direct feel for their product, service, culture, community, vibe, or je ne sais quoi, their website offers an experiential source of truth that has no substitute.

Ironically, when an audience has been “primed” through their AI assistant, and they’re ready to engage with a brand directly, the experience they have in that next moment has become more strategically important.

But — and this is a big but — the expectations that people have for what a great digital experience should be have been forever altered by AI. The website as a digital presence “home” isn’t dead. But the days of sclerotic sites that force visitors to adapt to them, rather than the other way around, are numbered. No one wants to spelunk through a sprawling information architecture — for many websites, I use that term loosely — to piece together a story from fragmented content that is often out-of-sync or out-of-date.

So what might a reimagined, AI-native website that addresses all of the above look like? How might it be architected differently?

That’s the latest research project I took on with Knotch, a content intelligence company that has evolved into an AI experience infrastructure platform. The result is a new paper, The Future of Websites: A Brand Context Model to Power Human & Agent Experiences, which you can download (free and ungated).

The high-level summary of this future as I see it: dual-surface websites — an intent-responsive surface for humans and a semantic/service-oriented surface for agents — both powered by a common “brand context model” underneath.

Let’s start with the concept of intent-responsive web experiences. It’s a nod to responsive web design, which was really device-responsive. AI assistants such as Claude and ChatGPT illustrate being intent-responsive. The user states what they want, and the AI serves it up to them. The whole experience bends to the user’s will.

That was such a dramatically better experience than the previous back-and-forth, hunt-and-click slog of search engines and websites that AI assistants were adopted faster than any other technology in history. So much better that they overthrew the behemoth that was Google Search.

As noted, AI assistants aren’t the brands represented within them. But can we bring that kind of intent-responsiveness into the owned experiences of brand websites?

“Like a chatbot on our website? We already got one.” No. Not something stuck on top, awkwardly competing with the underlying page. A new generation of web experiences that are natively designed to be intent-responsive.

One incarnation of that can be something that resembles an open-ended chat, à la a ChatGPT session — albeit with the benefit of being able to authoritatively answer questions specific to that business and that customer (e.g., “What’s the status of my order?”). Rather than being squeezed in a small, disembodied chat box, such a conversational experience can be fully embedded in the page itself. Here’s an example of what this looks like with Think with Google.

But in between two ends of the spectrum — from static pages to open chats — there’s a whole continuum of possibilities. The downside to purely open chat, which I think of as an “improv” type experience, is that it puts the burden on the user to know what they want and express it in words. That’s an efficient mode of interaction if they’re ready for that. It’s a dead end if they’re not.

An alternative is to design “scaffolded” intent-responsive experiences. They dynamically adapt to each individual within varying degrees of structure and guidance. This might take the form of choose-your-own-adventure branches, a “composed” comparison chart or planning board, or a tailored interactive explainer. The user may be presented with choices, options, or wayfinders to help them discover, develop, and express their intent.

The entire content of the page then immediately adapts to their intent. A good example of this is the Zillow Rentals solutions page, which offers both explicit personalization choices and an open-ended prompt box (which also includes a few one-click suggestions). It’s a conversational experience in its flow, but it’s not rendered as a “chat.”

It’s worth pointing out that “intent” over the past decade has largely been about inferred intent. Based on a user’s browsing and history, we infer (i.e., guess) what they want or how to best serve them. It’s been better than nothing — and it can still feed into intent-responsive experiences — but the explicitly expressed intent that visitors communicate in these conversational designs has much higher fidelity.

As a bonus, this expressed intent, when aggregated, offers far greater insights than traditional web analytics. It’s much closer to market research, but better incentivized for soliciting actual interests, preferences, and concerns — and doing so at scale.

A great experience for a human and a great experience for an agent differ. Yes, AI can simulate human interaction when necessary. But it’s not efficient. And now that we’re moving past the flawed tokenmaxxing frenzy, efficiency will be a larger factor.

Agents are more interested in the canonical than the adaptive. They work best with access to authoritative content, structured data with semantically clear facts, and — increasingly — callable services that enable agents to take action on behalf of the humans they’re working for. AI-era websites should give this to them as a first-class experience.

I’m certainly not the first person to recognize this. Many of the leading AEO and web platforms have embraced this approach over the past year. So I won’t dive deeper into it here. I cover more in the The Future of Websites paper.

The key point is that we want both: adaptive, intent-responsive experiences for humans and semantic, service-oriented experiences for agents. And we want them to be perfectly in sync, yet not require two different streams to maintain.

How can we enable — and maintain — these more adaptive experiences as the number of combinatorial possibilities that can be composed and generated grows exponentially? How can we keep parallel agent-optimized surfaces in sync? Orphaned and outdated content has plagued websites for three decades. Will this ideal next-generation website we’re envisioning only make it worse?

I believe the solution is to rethink the back-end that powers these experiences with a more AI-native architecture — a brand context model.

“Context” is certainly the word of the year in martech. To be fair, marketing has always invested in understanding the context of the customer, even if we haven’t used that term du jour. What’s new is thinking more explicitly about the context of the company — the many facets of the brand, the processes and priorities of the business, its decisioning rules, the sum of its knowledge, etc. — and making that context programmatically accessible to agents.

A brand context model is an incarnation of such a company context layer to power these AI-native, dual-surface websites. It’s a canonical, continually updated representation of the brand that both human and agent web surfaces can draw upon.

Rather than a rigid hierarchy of database records, it’s more of a graph of all the content, assets, components, brand guidelines, policies, and capabilities of the business — and the knowledge of how they’re related and interconnected — that can be combined on-the-fly in a more adaptive fashion.

This isn’t a CMS or even necessarily a replacement for a CMS. In fact, existing CMS, DAM, PIM, knowledge bases, etc., are all great authoritative sources that can feed a brand context model. It’s also not an LLM, that disassembles and then regenerates content probabilistically. The integrity of source material is deterministically maintained, along with all relevant rules and relationships. However, this model can provide context to an LLM, which might be part of a conversational experience you’re serving.

The team at Knotch has built a brand context model like this, called ACE, which is definitely worth taking a look at. But the architecture and experiences that I describe in The Future of Websites represent an approach that anyone can take. If you decide to read the paper, I’d love to hear what you think.

For 30+ years, websites and the tech stack that powers them have evolved in an incremental fashion. AI gives us the opportunity to rethink what’s possible in a more fundamental way.

Time to turn the page,

Scott