Our founder Will recently sat down with Laura, our Head of Design, to talk about how AI search is changing the way websites get designed.
The conversation kept returning to one idea: you now no longer can design for the user only, we need to consider AI systems, and shape decisions around both those two audiences.
A users role on the site is also entirely different to how a search engine or crawl bot may want to use your site.
This article is drawn from that conversation.
Design decisions now have a second reader
Until recently, most design decisions were settled by how the page looked.
Elements were grouped for visual hierarchy, white space was set for rhythm, and the test that mattered was whether a person could scan the page comfortably.
“We used to make design decisions based solely on how things looked on the page. Now we have to consider how the LLMs read a website, and make those decisions at the same time.” – Laura, Head of Design
A team card is a good example of what that means in practice. It holds a name, a job title and an email address. Splitting the three into separate visual containers can look clean, but to a machine reading the page they’re three unrelated items; nothing in the structure says the email belongs to the person. Grouping them in a single container, or joining them with something as simple as a divider line, makes the relationship explicit without changing the design language.
A recent FAQ block build raised the same issue from a different angle.
Each question needed a structural connector to its answer, so that a system reading the page knows which answer belongs to which question rather than treating the section as one continuous run of open text.
“They might seem like micro decisions, but they’re changing how we make choices across the whole design.”
These are the kinds of calls the team now makes weekly, and they compound: a site built from clearly grouped, clearly related elements is one an AI system can quote accurately.
Keeping up when the ground keeps moving
There’s no established playbook for designing this way yet, so the team’s sources are deliberately mixed.
Just as often, the answers come from the SEO side, because most of what’s changing in design right now is arriving through AI search: the same shift that has SEO teams restructuring content for ChatGPT, Claude and Perplexity is rewriting the rules designers work to.
Whatever survives that filter gets folded back into our design system, so the next project starts from the current state of play.
Search is moving inside the website
With more journeys starting in a search box or an assistant, we’ve started designing for that behaviour inside the site itself.
For one client, whose site holds a dense library of resource content that’s genuinely hard to navigate page by page, we’re exploring an AI-powered search within the website: rather than pushing visitors through layers of menus, let them ask for what they need.
Retail is already testing the pattern in public.
ASOS launched ASOS Stylist in May, a conversational shopping experience that turns product discovery into a dialogue rather than a browse.
One of our law firm clients has run search front and centre on their site since before the current wave of LLMs, for the simplest of reasons: their users know exactly what they want, and a search box gets them there faster than any navigation.
Laura’s view is that content-driven sites will keep moving in this direction, and it raises a question she doesn’t think anyone in the industry has answered yet.
Search results are templated; they look uniform because they have to filter in content from everywhere.
If an AI search becomes the main way into a site, the interface that assembles those answers becomes the design problem, and designing simultaneously for crawlers and for people is a challenge the industry is only starting to think about.
The visitors who still click are worth more
Several of our clients are watching traffic shift into AI Overviews, where the answer appears before the click.
The sessions that survive that filter behave differently: someone who reads a ready-made summary and still clicks through usually wants more than the summary, which puts them closer to converting than the average search visitor ever was.
That changes design priorities on the receiving page.
The care that used to be spread evenly across every landing page now goes disproportionately into the journeys those visitors take, and the measurement question gets sharper: which data tells you whether those higher-intent sessions actually convert, and where they stall.
What the next six months change, and what they don’t
We asked Laura whether AI will be doing the design work by the end of the year and the answer is a confident no:
“The gimmick of thinking AI can do the design role is softening. In six months’ time we’ll still be designing the interface manually, presenting it, and iterating on it. What will change is the circumstances we design in.”
The change she does expect sits underneath the interface.
Most sites, ours included, are designed around blocks: a heading, a paragraph and an image grouped into a section.
A component-based design system goes a level deeper, typing each piece (a text group, an image component) so that a system like Claude can understand and work with the individual parts.
Whether the block approach stays relevant, and what a standardised, machine-legible design system would replace it with, is the conversation the team is having now.
Expect AI to become more integrated and more streamlined across the design process rather than transformative of it. Timelines won’t suddenly collapse, and the hands stay on the interface.
Where this leaves you
If you’re planning a redesign and want machine readability handled at the design stage rather than retrofitted afterwards, our UX and UI design team designs for both readers from the first wireframe.
