The new standard for interaction and the changing role of a website.
People increasingly meet your brand inside an AI before they ever reach your website. That changes what a website is for.
IN SHORT
Think about what you did the last time you had to find information online or wanted to research a product or services.
If you had a specific question, there is a good chance you typed it into an AI assistant and read the answer… rather than a Google search, opening three tabs and scanning each page for the piece of information that was relevant to you.
This is people doing what people have always done, which is take the shortest path to what they need. And when that shortest path is a natural language answer, the traditional path through website pages starts to feel like friction and quite frankly, in the too-hard-basket.
ChatGPT is probably close enough. Right?
Unfortunately the part where someone forms their first impression of you is quietly moving into answer engines like ChatGPT and Google’s AI responses, and let’s be honest here… there’s only so much you can do with GEO (generative engine optimisation) and that’s a whole another subject to talk about.
But as consumers gain better understanding of the flaws of generic AI chatbots, I believe the new gold standard for brands will be… to deliver the exact same natural language experience, but in your brand’s environment with your own voice, guaranteed facts and strict guardrails.
They arrive already briefed: AI is now part of the buyer journey
The consequence is that by the time a person reaches your site, a conversation about your brand has already happened somewhere you could not see it.
This is clearest in business-to-business buying. In 6sense’s 2025 Buyer Experience Report, a global study of nearly 4,000 buyers, 94% said they used large language models during their buying journey. More striking, 95% of the time the vendor that eventually wins is already on the shortlist the buyer drew up on day one, up from 85% only a year earlier. And 6sense’s earlier research found the vendor a buyer reaches out to first goes on to win close to 80% of the time.
In other words, the persuasion now happens before the first actual conversation. 6sense’s own analysis adds a detail that should make every website owner sit up: LLMs are reducing curiosity-driven website visits. The casual browsing and comparing that used to happen on your pages is increasingly happening inside an answer engine instead, summarised by something that has never spoken to you.
The same shift is underway with consumers, and it is already visible in New Zealand. In an Adobe survey of just over a thousand New Zealanders, fielded in early 2025, 62% said they had used AI assistants more than once, in line with the global average. Among the Gen Z and Millennial early adopters, more than half were already using those assistants to research brands and products, and a third were using them in place of traditional search during the discovery stage.
This is not a distant trend to prepare for. It is already how a meaningful share of people form their first view of a company.
So the person landing on your homepage is not a stranger reading your menu for the first time. They arrive half-briefed, carrying a version of your brand that something else wrote.
And, that something else… is confidently wrong a good deal of the time.
In August 2025, NewsGuard found that the ten leading chatbots repeated false claims on news topics about 35% of the time, up from 18% a year earlier, in part because the models had stopped declining to answer questions they could not answer well.
“94% said they used large language models during their buying journey”
“browsing and comparing that used to happen on your pages is increasingly happening inside an answer engine”
RAG done wrong: why off-the-shelf AI answers still get it wrong
Nor is this solved by basic off-the-shelf retrieval systems. RAG, or retrieval-augmented generation, is the technique of grounding an AI’s answers in a defined set of documents instead of letting it answer from memory, and it is the architecture behind most serious business AI. Done casually, it is not enough. A peer-reviewed Stanford study of purpose-built legal research tools found they still produced incorrect answers between roughly 17% and 33% of the time, against about 43% for a general model.
And it’s not just a distant possibility when it comes to ramifications. In Moffatt v. Air Canada, a tribunal held the airline responsible for a refund policy its chatbot had invented, and flatly rejected the argument that the chatbot was a separate entity responsible for its own words.
If a machine speaks for you, you own what it says.
But like with most things new, the industry is adjusting, the technology is improving and more of us have access to powerful tools to fix this.
“If a machine speaks for you, you own what it says.”
RAG done well: the branded AI assistant you don't need to be scared of
The old job of a homepage was to orient. It greeted a stranger, explained who you were, and pointed them toward the page that held the generic answer.
The new job is different. It is to continue a conversation the person has already started elsewhere, and to turn whatever third-party version of your brand they arrived with into something accurate, in your voice, and matched to their actual situation. And now that person knows they are getting brand approved answers.
None of this means the website disappears. The honest version of the shift is not that websites stop mattering, it is that they change role. As discovery moves into AI, the site becomes the trusted source of record: the place a person goes to verify what the generic machine told them, to find the depth a summary could never carry, and to complete the high-value, high-trust actions that a generic chat window is not built for.
Similarweb’s own chief executive described the behavioural half of this plainly, noting that people now start inside AI assistants, shaping their preferences and deciding who to trust before they reach a website. The site that wins is the one built to serve two audiences at once: the human who wants a confident, branded answer, and the machine that is reading you on that human’s behalf.
“the trusted source of record: the place a person goes to verify what the generic machine told them”
Getting it right, in order: strategy, knowledge architecture, then technology
The end state most brands are moving toward is a branded, on-site assistant that knows the company, speaks in its voice, and operates inside proper guardrails with continuous auditing.
Vanta, a compliance company, is a clean example. It runs a branded assistant as a single front door for customer questions, and in Intercom’s case study of the deployment, Vanta’s support lead reports fielding 45% fewer inbound inquiries over email even as its user base doubled.
The interesting part is not the deflection figure. It is the posture: one branded, governed place that answers, instead of a static site the customer has to excavate.
The cautionary tale sits right beside it. Klarna rolled out an AI assistant in 2024 that quickly took on the work of hundreds of support agents, and then, in 2025, publicly acknowledged it had leaned too far into automation at the expense of the harder, higher-value conversations, and began rebuilding human support for them.
The lesson is not that the assistant was a mistake. It is that an assistant without judgement about what to hand back to a person is a liability, and that the strategy has to come before the tool.
This is the part that is easy to skip and expensive to ignore. A brand assistant is only ever as trustworthy as the knowledge behind it and the guardrails around it. Point a capable model at a pile of undifferentiated content and you get a fluent version of the same confident wrong answers the public models already produce, except now it is wearing your logo. Get the knowledge architecture right first, decide what the assistant is allowed to say and what it must escalate, audit it continuously, and only then does it turn AI-mediated discovery from a risk into an advantage.
“Vanta’s support lead reports fielding 45% fewer inbound inquiries over email even as its user base doubled.”
The conversation about your brand has already started
Someone, somewhere, is asking an AI about your industry right now, and there is a reasonable chance your brand comes up.
The only open question is… what happens when that person finally arrives? Will they meet a directory of web pages asking them to start hunting for information, or a branded, accurate, well-governed continuation of the conversation they were already having?
The risks are real, and you’ve heard the cautionary tales. But when you do the knowledge and the architecture first, with the guardrails and the auditing throughout, you’ll provide your customers with the new gold standard of a brand experience.
Quick answers
Do websites still matter now that people research with AI?
Yes, but the role changes. As discovery moves into AI assistants, the website becomes the source of record: where people verify what the machine told them, find real depth, and complete high-trust actions. The sites that win will serve the human reader and the AI reading on their behalf at the same time.
Why do AI chatbots get facts about brands wrong?
Because they are built to produce plausible answers, not verified ones. NewsGuard found leading chatbots repeated false claims about 35% of the time in 2025, and even purpose-built retrieval tools in a Stanford study erred 17 to 33% of the time. Accuracy comes from the knowledge architecture behind the AI, not the AI alone.
What is a branded AI assistant?
An AI assistant that lives on your own website, answers in your voice from your verified information, and operates inside strict guardrails with continuous auditing. It gives visitors the natural language experience they now expect, with answers your brand actually stands behind.
What is GEO (generative engine optimisation)?
The practice of structuring your content so AI assistants describe your brand accurately and cite you as a source. It is worth doing, but it only influences what other people’s machines say about you. A branded assistant is the part of the AI experience you fully control.
Linkki builds BrandRAG systems: branded, retrieval-first AI assistants built on structured, verified source material, with strict guardrails and continuous auditing.
This article is part of an ongoing series on how AI is changing the way customers find and trust brands.
