“Do you have flowy, colourful trousers in my size that work for Lindy Hop on a warm dance floor?“
When users ask with this level of nuance, they expect an answer, not a menu to browse. Large language models like ChatGPT and Perplexity already interpret such intent and generate useful responses. Online users increasingly expect the same from your website: not merely to display information, but to clarify, guide, and convert.
As WeAreBrain’s Head of UX and Design, I’ve been exploring how AI-driven websites represent a fundamental shift in digital experience. This article shares what we’ve learned about why human-centred design principles remain essential and how to navigate this transition thoughtfully.
Why the ‘brochure and catalogue’ model no longer suffices
We’ve been building websites primarily as brochures or catalogues. That model continues to serve a function, but I’m seeing its limitations become increasingly apparent:
Cognitive overload
Visually dense, navigation-heavy interfaces force users to work too hard to find what matters. Complex menus, unfocused content, and competing calls to action obstruct straightforward answers.
Search inadequacy
When a user’s task is specific, their instinct is to search. Yet standard site search returns only links to pages, not answers.
Strategic vulnerability
If your website remains a passive data source for LLMs or external aggregators, you risk invisibility. Others will harvest your content and redirect your users.
Without a conversational, answer-driven layer embedded in your site, you relinquish control over the interface mediating between your users and your offerings.
The question shifts from “Do we add a chatbot?” to: How do we transform your website into an intelligent, interactive surface that serves as your organisation’s best representative?
What endures in an AI-driven paradigm
Whilst AI evolves rapidly, human cognition remains bounded by long-standing limitations. In our design work, we ensure conversational interfaces respect these truths:
Limited working memory
Users cannot hold multiple threads simultaneously. Break tasks into digestible steps and maintain visible context throughout interactions.
Need for predictability
AI feels powerful, yet users require safety. Provide clear cues, control mechanisms (reset, go back), and transparency about how their input is used.
Pattern-seeking behaviour
Users expect organised information. Conversational UX must mirror mental models rather than force unfamiliar taxonomies.
Spatial navigation
Without stable visual anchors, users become disoriented. Retain structure, breadcrumbs, and persistent context panels to prevent confusion.
Latency sensitivity
Users will not wait. Even complex tasks must feel responsive through background processing, optimistic UI, or progressive reveal strategies.
Anthropomorphism
Resist the temptation to create an overly human persona. Aim for an experience that feels powerful because it leverages machine intelligence, not because it pretends to be human.
Voice interfaces add accessibility and emotional connection, yet consistency matters: the same intent understanding and brand personality must persist whether users read, type, or speak.
Challenges requiring immediate attention
In our projects, we’ve learned that anticipating obstacles proves as crucial as designing the vision.
SEO and discoverability
Your site requires crawlable pages, metadata, and schema markup. Conversational output should complement this structure, never replace it. Search engines must continue indexing your core content.
Prompt injection and abuse
Conversational interfaces face misuse. We implement filtering, sandboxing, rate limits, and human review for critical areas.
Hallucinations and trust
LLMs can generate plausible yet incorrect answers. We display confidence indicators, cite sources, and enable users to escalate to stable pages or human support when needed.
Domain drift
Models lose relevance over time. We keep knowledge modular, retrain regularly, and version content to enable updates without system-wide disruption.
Maintaining narrative flow
The traditional brand narrative funnel still matters. Dialogue should reinforce your story, not override it.
Graceful degradation
When AI cannot respond adequately, we redirect to pages, forms, or people. Transparency builds trust.
Real-world applications from our projects
We’ve explored this territory through several projects that illustrate both promise and pitfalls.
Product discovery optimisation
For Praxis/Maxeda, we built a Product Locator combining store layout data with conversational layers. Customers ask “Where is product X?” and receive precise aisle information. Staff manage layout metadata to reflect real-world configurations. The conversational interface integrates deeply into the product discovery path.
Process automation
We created “Harry,” a workbot guiding vendors through complex procurement workflows. This illustrates how conversation, UI integration, and process orchestration reduce friction and errors.
Security lessons
We deployed a chatbot on our own site but removed it due to spam and prompt-injection attempts. This experience informed our approach to sandboxing and abuse prevention, and it’s why we now emphasise security from inception rather than as an afterthought.
A pragmatic roadmap we follow with clients
We help organisations transform their websites incrementally rather than attempting wholesale replacement:
1. Audit and modularise content
Map existing pages into domain modules and knowledge graphs. Identify high-value conversational candidates (product discovery, FAQs, support).
2. Pilot limited domains
Select a narrow vertical (one product category or support flow) and add a conversational assistant that clarifies, recommends, and links to pages.
3. Embed guardrails from inception
Implement input validation, fallback pathways, human handoff, logging, and moderation.
4. Observe and iterate
Monitor task success, user satisfaction, and fallback rates. Use collected data to refine prompts and train domain modules.
5. Evolve towards blended interfaces
Gradually introduce conversational surfaces that coexist with content, visuals, and transaction flows.
6. Govern and refresh
Maintain domain knowledge as your single source of truth. Version changes, review answers, and remove outdated information.
Over time, your website transforms from static brochure into trusted, intelligent guide where structured content, conversational immersion, and transactional capabilities work in harmony.
Comparing approaches: Traditional vs AI-driven websites
| Aspect | Traditional Website | AI-Driven Website |
|---|---|---|
| User interaction | Click through menus and pages | Ask questions in natural language |
| Search results | List of page links | Direct answers with context |
| Information architecture | Fixed navigation hierarchy | Dynamic, intent-based pathways |
| Personalisation | Segment-based (limited) | Individual, contextual understanding |
| Updates | Manual page edits | Knowledge module updates |
| User effort | High (browse, filter, compare) | Low (ask, receive, decide) |
| Scalability | Add more pages/sections | Train on more domains/intents |
| Maintenance | Update individual pages | Maintain knowledge graphs |
Vision: Your digital representative
Consider a website greeting visitors with a simple prompt: “What brings you here today?“
You type: “Do you have flowy, colourful trousers in my size that work for Lindy Hop on a warm dance floor?”
Your digital representative interprets the intent (movement, comfort, style, context) and presents realistic visualisation of how such trousers perform. Behind the scenes, generative AI models collaborate to translate natural language into meaningful, visual answers. Each interaction enriches the system’s understanding of customer intent and domain-specific nuance.
The traditional, crawlable site persists for transparency and discoverability, now enhanced by an adaptive, multimodal layer that authentically represents your brand and serves your users.
Why this matters now
For decades, interaction evolved from desktop to mobile to voice, always requiring humans to adapt. LLMs reverse that pattern: they adapt to us. These systems interpret intent even when language is imperfect (typos, fragments, missing context) and respond with precision.
This ability to understand before we fully articulate represents a genuine paradigm shift. Websites built around menus and click paths feel outdated because users now expect to ask and receive answers.
If your site remains static, you risk becoming invisible. Move too slowly, and others will control how your content is represented. Yet if you experiment now with design sensitivity, cognitive respect, and technical rigour, you can lead this transition.
Our approach at WeAreBrain
At WeAreBrain, we’ve explored this shift through conversational systems and hybrid interfaces, applying lessons from client projects and our own experiments. From a UX design perspective, I’ve learned that successful AI-driven experiences require:
Design thinking first
Technology enables experiences, but human needs drive design decisions. We start with user research, journey mapping, and clear problem definition.
Iterative prototyping
We test conversational flows with real users early, identifying friction points before development.
Cross-functional collaboration
UX designers, engineers, and domain experts must work together from day one to balance ambition with technical reality.
Ethical considerations
We design with transparency, user control, and privacy as core principles, not afterthoughts.
Measurable outcomes
We define success metrics upfront: task completion rates, user satisfaction, fallback frequency, and business impact.
We understand both the promise and the pitfalls, and we know where to begin safely.
If your organisation is ready to evolve beyond the brochure model and create a digital representative that authentically understands and serves your audience, let us design and build what comes next together.
