ARTLOGIC

Knowledge hub

Web design & development

A website is the only asset that has to satisfy three audiences at once: a person deciding whether to trust you, a crawler deciding whether to index you, and increasingly a model deciding whether to name you. Most sites are built for the first and retrofitted for the others. This covers building for all three from the start.

Before you commission anything

The decisions that determine whether the project succeeds are made before any design work begins.

When the site exists and does not perform

Traffic arriving and nothing happening is the most common brief we receive. The diagnosis is rarely aesthetic.

Building a site that can be found

The technical foundations decide whether anything else you do has a chance. They are cheapest to get right at build time and most expensive to retrofit.

Designing a site that can actually rank

Most sites are designed, then handed to someone asked to make them rank. That order is the problem. By then the URL structure is set, the navigation reflects the org chart rather than how people search, and the substance that should be indexable is trapped inside components that render after the crawler has left.

Retrofitting is possible and costs several times what deciding correctly would have. The decisions that matter are made in the week before visual design starts, not in the month after launch.

  • URL structure decided before templates — changing it later means redirects and lost equity.
  • Substance in the initial HTML, not injected after hydration. A crawler that does not execute your JavaScript sees an empty page.
  • Navigation built on how buyers search, which is rarely how the company is organised.
  • One page, one primary intent — the discipline that prevents your own pages competing.
  • Internal linking treated as architecture rather than a footer afterthought.
  • Structured data that matches what is visible. Markup contradicting the page is a negative signal, not a neutral one.
Search Dominance →

Why conversion is a design problem before it is a testing problem

A/B testing gets the attention, and it is the last thing you should reach for. Testing needs traffic volume most business sites do not have, and it optimises within a layout rather than questioning it. If the page never states what the company does in the visitor's language, no button colour recovers that.

The changes that move conversion on most sites are unglamorous: saying the thing plainly above the fold, removing a form field that exists for internal reporting, and making the next step obvious rather than available.

  • Message match — what brought someone here appears in the first screen, in their words.
  • One primary action per page, sized to where the visitor is in the decision.
  • Forms that ask only for what is needed to have the next conversation.
  • Proof placed where the doubt occurs, not collected on a testimonials page nobody visits.
  • Mobile treated as the primary case, because that is where most abandonment happens.
  • Testing only once traffic supports it — otherwise you are reading noise as signal.
Landing Page Optimization →

Performance as a business constraint, not a score

Chasing a perfect audit score is the wrong target. The number is a proxy; what matters is whether the page is usable quickly on the device and connection your buyers actually have. A site scoring in the nineties on a developer's laptop can be unusable on a mid-range phone on cellular.

Performance also erodes rather than breaks. Each deployment adds a script, a font, a tag, and nobody is watching the aggregate. A budget enforced on every deploy is the only control that survives contact with a real release cadence.

  • Measure on the devices and networks your analytics show, not on a build machine.
  • Largest Contentful Paint is usually an image or font problem before it is a code problem.
  • Third-party tags are the most common cause and the least examined — audit what marketing added.
  • Server rendering where it helps; it is not automatically faster for every page type.
  • A performance budget checked on every deploy, because erosion is gradual and invisible.
  • Core Web Vitals matter, and they are a floor rather than a strategy.
Technical SEO Audit →

AI and your website

Two different things called the same name

"AI website" is used for two unrelated projects. One is putting AI into the site — an assistant, personalisation, automated qualification. The other is making the site legible to AI systems so it gets cited when someone asks a question. They share a buzzword and almost nothing else.

Confusing them is expensive. A company that needs to be findable in ChatGPT buys a chatbot instead, and the chatbot does nothing for the problem they actually had.

Putting AI on the site: where it genuinely pays

The honest set is narrower than the pitch. AI on a website earns its place where the volume is high, the answers are already documented, and being wrong is cheap to correct.

  • Answering questions already covered in your documentation — with an escalation path when confidence is low.
  • Qualifying enquiries before a human sees them, using criteria your sales team can state.
  • Internal search that handles synonyms, which converts well above browse on catalogue sites.
  • Drafting responses for a person to review, rather than sending unreviewed.
AI Automation →

Where it does not

A chatbot on a five-page site with low traffic is a cost with no return. Neither is an assistant deployed over documentation that is out of date — it will answer confidently from stale material, and confident wrong answers damage trust faster than no answer.

The test is simple: if a person could not answer reliably from the material you have, a model will not either. It will just be more fluent about being wrong.

Making the site legible to AI systems

This is the other project, and for most businesses it is the one that matters commercially. Generative engines retrieve from the open web. Whether you appear depends on whether an agent can reach you, whether your entity resolves, and whether your substance can be extracted as a passage.

It overlaps heavily with technical SEO, which is why we treat the foundations as one piece of work rather than two engagements.

  • Retrieval agents allowed in robots.txt — the single most common blocker we find, and free to fix.
  • Substance in the initial HTML. Content that exists only after hydration is often invisible to retrieval.
  • An entity that resolves — consistent naming, structured data matching visible content, corroborating profiles.
  • Passages that stand alone, because retrieval works at passage level rather than page level.
  • Claims carrying attribution, which the founding GEO research measured as the strongest single effect.
AI Visibility & GEO →

What nobody can promise

No supplier controls whether a model cites you, and generative retrieval is non-deterministic — the same question produces different answers and different sources across runs. Any guaranteed citation should be treated as a warning sign.

What is measurable is answer share and citation share, sampled repeatedly against a fixed question set and a baseline taken before the work starts. Our full method is published.

GEO methodology →

Where we do this work

Platform choice, honestly

We build on WordPress, Shopify, Drupal, Webflow, Squarespace, Wix, WooCommerce and custom Next.js — and the platform is a consequence of the requirement, not the starting point. Most agencies recommend whichever platform they happen to run. We would rather ask what the site has to do, then tell you which platform makes that easiest to maintain. Sometimes that answer is the one you already have.

By industry

Strategy Call

See Exactly Where You Stand.

Every relationship starts with intelligence, not a proposal. A strategy call gives you a clear picture of your AI visibility, search authority, and competitive gaps — and a realistic view of what is achievable.

Book a website assessmentSee client results