Project

1 / 14

Nuxsol Website

Research

Site structure

Web design

Design system

Motion rules

2026

The task

Nuxsol is an AI-native consulting and custom-software company in the Netherlands, working across the EU and EEA. It sells brand identity, digital design, product design, marketing and growth, and development, including websites that AI systems can read.

A company that sells readable websites needs one of its own. The task was a site that presents the five services and lets a visitor check every claim about readable websites on the site itself.

One site, two readers

Buyers now ask an AI assistant before they visit a site. So the site has two readers: a person in a browser and a machine fetching the text. We planned every page twice, as a designed page and as a Markdown copy at its own address, both made from the same content. Map It did the research, the structure and the design. Nuxsol developed the site.

How we studied it

Desk research. We studied 45 agency and consultancy sites, ranked 14 AI-readiness features by the evidence behind each, reviewed the owner's reference site and studied three template sites.

What we had to solve

  • No site in the sample combined brand, software, search visibility and agent-readable websites, so the site is built on one idea that holds them together: two readers, people and machines.

  • AI-readiness is often sold as one file, so we ranked 14 features by evidence and put crawler rules and copy in the HTML ahead of llms.txt.

  • Service pages are hard to quote, so all five follow one pattern that opens with a stance line and a one-sentence definition.

  • A case shown as a cover says little, so every case page shows the problem, the work per service, four decisions and the system.

  • The machine layer is normally invisible, so a terminal in the hero and a Human and Machine switch in the header put it in view.

  • Two versions of a page drift apart, so the Markdown copy is generated from the same data as the designed page.

How the Nuxsol site was made

Fourteen steps, from a study of 45 agency sites to a site where every page exists twice: designed for people, Markdown for machines.

1Understand
Board: The first screen.
The first screen
01

The brief

We set the task: one site that presents the services and proves its own claims.

  • Present the services
  • Read as AI-native
  • The site as its own proof
The problem
Nuxsol sells brands, websites, products and software, and websites that AI systems can read. A company that sells readable websites cannot have one that reaches a crawler as an empty page.
Our thinking
Buyers now ask an AI assistant before they visit a site. So the site has two readers: a person in a browser and a machine fetching the text.
What we did
We wrote down what the site had to do: present the services, read as AI-native rather than as a classic agency, and let a visitor check every claim about readable websites on the site itself.
How we did it
Desk research: a study of 45 agency and consultancy sites, a ranking of AI-readiness features, a review of the owner's reference site and a study of three template sites.
What changed
The home page opens on the line "Built for people. Readable by machines." A terminal under it shows the site's own machine files.
Board: The reference field.
The reference field
02

The reference field

We studied 45 agency and consultancy sites in three groups.

  • 45 sites, three groups
  • Machine files checked
  • One gap in the field
The problem
No independent ranking of AI-consultancy websites exists. The lists that do exist are driven by reviews or published by the agencies that top them.
Our thinking
Look at what the sites show and publish, not at what the lists say about them.
What we did
We grouped 45 sites three ways: sites with a similar service mix, award-level agency sites, and Nordic and brand-strategy benchmarks. We checked each home page for llms.txt, structured data, language tags and crawler rules.
How we did it
For each site we used the award bodies' own records, a check of the home page on 6 November 2026 and the proof shown on the page.
What changed
No site in the sample offered brand strategy, custom software, search visibility and agent-readable websites in one menu. That gap gave the site its idea: two readers, people and machines.
Board: AI-readiness, ranked by evidence.
AI-readiness, ranked by evidence
03

The evidence

We ranked 14 AI-readiness features by the evidence behind each.

  • 14 features ranked
  • Crawler rules come first
  • llms.txt as a convenience
The problem
AI-readiness is often sold as one file. The research we read showed no link between having an llms.txt file and being cited in AI answers.
Our thinking
Design first for what the evidence supports. Add the rest as a convenience for agents, and say so on the site.
What we did
We ranked 14 features. Crawler rules, copy in the first HTML response, quotable passages and semantic HTML lead the list. llms.txt is rank 10 and Markdown copies for agents are rank 12.
How we did it
Each feature is listed with what to implement, the evidence for it and the effort it takes.
What changed
The site names nine crawlers in its robots file and carries all copy in its HTML. One of its three notes is titled "llms.txt: what it does, and what it does not".
2Structure
Board: The site map.
The site map
04

The site map

We set five menu items and gave each service, case and note its own page.

  • Five menu items
  • Five service pages
  • 27 addresses in the sitemap
The problem
The site has to carry services, work, a way of working, the studio and notes. A visitor and a crawler both need each of them at one clear address.
Our thinking
One thing, one address. If a service has its own page, an assistant can point to it and a buyer can be sent to it.
What we did
We set the menu to Services, Work, Approach, Studio and Insights, with a contact button beside it. Each of the five services, each case and each note has its own page.
How we did it
The structure follows the feature list from the template study: a multi-page site with collections behind work and notes, filters on the indexes, a site search, a contact form and legal pages.
What changed
The sitemap lists 27 addresses: home, the services index and five service pages, the work index and ten case pages, approach, studio, the insights index and three notes, contact and two legal pages.
Board: The home page, from claim to proof.
The home page, from claim to proof
05

The home page order

We ordered the home page from the claim to the proof.

  • Ten scenes after the hero
  • Proof before the offer
  • Questions at the end
The problem
A home page that only lists services asks the visitor to take every claim on trust.
Our thinking
Show what a machine finds on this site first, then the work, then what is for sale.
What we did
We set ten scenes after the hero: what an agent finds, the belief, work, services, approach, a walk-through of an agent reading the page, three ways to start, notes, questions and contact.
How we did it
Each scene has one job. Nine of them can flip to their own part of the page's Markdown copy.
What changed
The first scene after the hero counts what an agent finds on the site: pages, Markdown copies, schema types, named crawlers and languages.
Board: One pattern for every service page.
One pattern for every service page
06

Service pages

We gave all five service pages the same seven parts.

  • One stance line per service
  • Seven parts, same order
  • How we measure it
The problem
Service pages often describe a service in adjectives. Neither a buyer nor an assistant can quote a page that never says what the service is.
Our thinking
Start each page with a definition that can be quoted on its own. Then say how it works, what you get and how it is measured.
What we did
We set one pattern: a stance line, then What it is, What it covers, How it works, What you get, How we measure it, A good fit when and In the work.
How we did it
The pattern follows the ranked list: each service page carries a one-sentence definition that a machine can lift as a passage.
What changed
Every service page has the same headings in the same order. The Development page opens on the stance line "The design, built exactly."
Board: Every case shows its decisions.
Every case shows its decisions
07

Case pages

We gave every case page one order, from the problem to the system.

  • Problem first
  • What we did, per service
  • Four decisions with reasons
The problem
A case shown as a cover image and one paragraph says little about how the work was done.
Our thinking
A buyer reads a case to judge how a studio thinks. So each case has to show its decisions and the reasons for them.
What we did
We set the order: the case, the problem, what was built, who it is for, what was done per service with its steps, four decisions, the system, how it works and what it is built to change.
How we did it
The system table lists each project's four tones with their roles. Contrast ratios are computed from the project's own colours when the site is built.
What changed
Ten case pages follow this order. The work index filters them by the five services and by text.
3Design
Board: The system.
The system
08

Colour and type

We set the system as tokens: a near-black page, one lime signal, two typefaces.

  • Lime as a signal only
  • Geist and Geist Mono
  • Nothing below 12 px
The problem
Nuxsol already had a logo and a palette: lime, near-black and off-white. Lime is strong. Over large areas it tires the eye quickly.
Our thinking
Lime marks what matters and nothing else. Type does the rest of the work.
What we did
We set a dark framed direction: a near-black page, white text, and lime for the mark, one word in the headline, arrow boxes, status dots and the ticker. It follows the language of a bought Framer template, rebuilt in the site's own code.
How we did it
Colours, type sizes, radii and easing are tokens in one file. Geist sets everything people read. Geist Mono sets labels, addresses, file names and Markdown.
What changed
Headings scale with the screen: the main heading runs from 2.5 to 5.625 rem. No text is set smaller than 12 px.
Board: The machine files, in the hero.
The machine files, in the hero
09

The terminal

We put the site's machine files in the hero, in a terminal with three tabs.

  • Three tabs
  • llms.txt, schema, robots
  • The machine layer in view
The problem
Structured data and crawler rules are normally invisible. A visitor has to take it on trust that they exist.
Our thinking
If readable websites are what the company sells, the machine layer should be visible in the first screen.
What we did
We designed a terminal for the hero with three tabs: llms.txt, schema.json and robots.txt. It sits under the headline, next to the opening text and the first button.
How we did it
The hero is a framed panel: a headline with one boxed lime word, a white button with a mono link, the terminal, and a lime ticker along its lower edge.
What changed
The first screen of the home page shows what a machine reads on the site next to what a person reads.
Board: Human or Machine, one switch.
Human or Machine, one switch
10

Human and Machine

We put a switch in the header that shows any page as a machine reads it.

  • On every page
  • Same words both ways
  • Markdown over the page
The problem
A claim that a page reads the same for people and machines is hard to check without opening a second address.
Our thinking
Let the visitor compare the two versions in one place, with one control.
What we did
We designed a Human and Machine switch for the header of every page. Machine shows the page's Markdown copy over the designed page.
How we did it
The switch sits in the header beside the search control. On the home page, single scenes can also flip to their own part of the Markdown.
What changed
Open any page, press Machine and compare. The Markdown comes from the same content as the designed page.
Board: Five services as rows.
Five services as rows
11

Services as rows

We drew the five services as numbered rows that stick and stack on scroll.

  • Five numbered rows
  • Transform and opacity only
  • A switch to reduce motion
The problem
Five services in a grid of cards read as five equal boxes. Motion added for show slows a page and tires a reader.
Our thinking
Motion should confirm what the reader just did. It should never be needed to read the page.
What we did
We set the services as five numbered rows that stick and stack as you scroll. We limited motion across the site to transform and opacity.
How we did it
A switch in the footer and the system setting both turn motion down. Scroll and animation run on one clock.
What changed
The home page shows the services as five numbered rows. A Reduce motion switch sits in the footer of every page.
4Build
Board: What an agent finds.
What an agent finds
12

Every page, twice

Nuxsol built the site in Astro: 28 pages, each with a Markdown copy.

  • Built in Astro by Nuxsol
  • All copy in the HTML
  • Counts taken at each build
The problem
Many sites reach a crawler as a script and a nearly empty page. A text copy written by hand drifts away from the designed page over time.
Our thinking
Generate both versions of a page from one source, so they cannot differ.
What we did
Nuxsol developed the site in Astro as static HTML: 28 pages, with all copy in the HTML. Each has a Markdown copy at a predictable address, made from the same data as the page.
How we did it
One function per page builds the Markdown from the data the designed page uses. Motion runs on GSAP and Lenis.
What changed
The home page shows five counts taken at each build: pages, Markdown copies, schema types, named crawlers and languages.
Board: Files for machines.
Files for machines
13

Machine files

Nuxsol added llms.txt, a robots file, a sitemap and structured data.

  • Nine crawlers named
  • Eight schema types
  • A preferred citation line
The problem
An assistant answering a buyer needs a short description it can quote and a clean copy of each page. A crawler needs to know it is welcome.
Our thinking
State the rule for each crawler by name. Hand agents an index instead of making them guess.
What we did
Nuxsol added llms.txt with a summary, a line per service and a preferred citation, a second file with the full text of every page, a robots file that names nine crawlers, and a sitemap.
How we did it
Structured data sits on every page as one graph. Eight schema types are used, among them Organization, Service, Article, CreativeWork and FAQPage.
What changed
llms.txt links to the Markdown copy of every service, case, note and page. The sitemap lists 27 addresses, each with a date set at build.
Board: Search, and the phone.
Search, and the phone
14

Search and phone

Nuxsol built a search palette over the whole site and a layout that scales to a phone.

  • Opens with Cmd K or /
  • Index from the Markdown
  • Margins from 16 to 30 px
The problem
A site with 28 pages needs a quick way to reach any of them, on a laptop and on a phone.
Our thinking
The Markdown copies already hold every page's text. Search can use them, with no second index to keep by hand.
What we did
Nuxsol built a search palette that opens with Cmd K, Ctrl K or the slash key. It searches an index made from the Markdown copies, and a search page lists every result.
How we did it
On small screens the menu becomes an overlay. Page margins and type sizes are fluid tokens, so one layout runs from phone to desktop.
What changed
Type a word in the palette on any page and it lists the matching pages, services, cases and notes.

Where it is now

  • 28 pages built in Astro, with all copy in the HTML a crawler receives.
  • A Markdown copy of each page, made from the same content as the designed page.
  • llms.txt, a robots file that names nine crawlers, a sitemap and structured data on every page.
  • A Human and Machine switch in the header of every page, and a search palette over the whole site.

The brief and the field

Nuxsol sells websites that AI systems can read, so its own site had to prove the claim. We studied 45 agency and consultancy sites and ranked 14 AI-readiness features by the evidence behind each.

A yellow Bolo Lemon box with a sliced lemon drizzle cake on blue
An orange Bolo Carrot box with a carrot cake and a cut slice on mint

Structure

We set five menu items and gave each service, case and note its own page. The home page runs from the claim to the proof. Service pages share seven parts, and case pages share one order.

A pink Bolo Birthday box with a sprinkle cake and candles on yellow
A white Bolo bag sealed with a red order sticker on orange
A sheet of red Bolo order seals on a paper bag on blue

Design

A near-black page, lime as a signal only, Geist and Geist Mono. The machine layer is part of the design: a terminal in the hero shows the machine files, and a switch in the header shows any page as Markdown.

A pink Bolo Birthday candle box with twisted candles and ribbon
A box of four Bolo cupcakes on tomato red
A Bolo Carrot box tied with red ribbon on a sunny café table

Built for two readers

Nuxsol built the site in Astro: 28 pages, each with a Markdown copy made from the same content. It added llms.txt, a robots file that names nine crawlers, a sitemap, structured data and a search palette.

A slice of chocolate cake beside a Bolo Chocolate box on yellow
A slice of lemon drizzle cake on a pale blue plate on yellow

What we delivered

  • Research: a study of 45 agency and consultancy sites in three groups, a ranking of 14 AI-readiness features, a review of the owner's reference site and a study of three template sites.

  • Site structure: five menu items, a home page of ten scenes after the hero, one pattern for the five service pages and one order for the case pages.

  • Two versions of every page: the plan for a designed page and a Markdown copy at its own address, from the same content.

  • Design system: tokens for colour, type, radii and easing. A near-black page, lime as a signal only, Geist and Geist Mono.

  • The hero: a framed panel with a headline, one boxed lime word, a terminal with three tabs for the machine files, and a lime ticker.

  • Human and Machine: a switch in the header of every page that shows the page's Markdown copy over the designed page.

  • Page designs: home, services, work and case pages, approach, studio, insights, contact, search and legal pages.

  • Motion rules: transform and opacity only, with a switch in the footer that turns motion down.

What Nuxsol built

Nuxsol developed the site in Astro as static HTML: 28 pages, each with a Markdown copy. It added llms.txt and a full-text file, a robots file that names nine crawlers, a sitemap of 27 addresses, structured data with eight schema types and a search palette.

Credits

Map It: research, site structure and design, including the design system, the page designs and the motion rules.

Nuxsol: development of the site, the Markdown copies, the machine files, structured data and search.

Third-party tools: the site is built with Astro. GSAP and Lenis run the motion. Geist and Geist Mono are third-party typefaces. The visual direction follows the language of a bought Framer template and was rebuilt in the site's own code.

Client

Nuxsol, an AI-native consulting and custom-software company in the Netherlands

Category

Website

Year

2026

Services

Research

Site structure

Web design

Design system

Motion rules

Contact

© 2026 Map It, KVK 98282859, VAT NL005321010B77

Cookies and storage

This site sets no cookies of its own. With your permission it loads the booking calendar straight away and remembers your planner and camera choices in your browser. Privacy and cookies