Eight problems derail most web design projects: building before you know the audience, picking a tech stack that costs you performance, navigation nobody can hold in their head, accessibility left until the end, SEO retrofitted after launch, security treated as someone else’s job, responsive work that only holds at three widths, and a site that looks good but does not convert. Every one of them is cheaper to fix in a wireframe than in production.
Key points
- Most web design failures are decisions deferred, not mistakes made.
- 95.9% of home pages had detectable WCAG failures in February 2026 — a figure that got worse, not better.
- Core Web Vitals targets are LCP under 2.5 s, INP under 200 ms and CLS under 0.1.
- Settle navigation and content structure before visual design begins.
- Audit early: accessibility and SEO cost most when retrofitted after launch.
Table of Contents
1. Designing before you know who it is for
Brand recognition is rarely the actual goal. The site has to attract the right people and get them to do something. The challenge is that “the audience” is usually asserted in a kickoff deck and never checked again.
Fix it by making two questions answerable in one sentence each before any visual work starts: who is this for, and what do they need to accomplish here? If nobody on the project can answer without hedging, you are about to design a site that pleases the stakeholder who talks most. Analytics from the existing site, five customer support transcripts and three user interviews will beat a persona workshop.
2. A tech stack that quietly costs you performance
Stack choices are made early, by whoever is available, and paid for by every visitor afterwards. The measurable consequence is Core Web Vitals, and Google publishes the thresholds you need to hit at the 75th percentile of page loads.
Three decisions do most of the damage: a JavaScript framework used for a site that is mostly documents, a font loading strategy that shifts text after paint, and images shipped at whatever size the CMS produced. All three are fixable, but only the first is expensive once you have committed. Set a performance budget before the stack is chosen and measure against real devices rather than your laptop.
3. Navigation people cannot hold in their heads
Smooth navigation lets people find what they came for without learning your org chart. Use descriptive labels drawn from what users call things, not internal product names. Keep the top level short enough to scan, give every page a sensible path back, and include a search field once the site passes a few dozen pages.
The cheapest test is a tree test: give ten people a list of tasks and your menu structure with no visual design at all, and watch where they click. It costs an afternoon and routinely overturns decisions that took a month.
4. Accessibility left until the end
This is the biggest gap in the industry, and in 2026 it got wider. WebAIM’s annual analysis of a million home pages, published in February 2026, found detectable WCAG 2 failures on 95.9% of them — up from 94.8% the year before — with an average of 56.1 errors per page, a 10.1% increase. WebAIM links the reversal to heavier use of third‑party frameworks and libraries, and to automated and AI‑assisted coding.
Look at what is actually failing. Low contrast text on 83.9% of pages, missing alternative text on 53.1%, missing form labels on 51%. These are not obscure technical requirements — they are things a designer decides in Figma and a content editor decides in the CMS. Which is good news: most of the list is fixable without touching the framework.
Contrast in particular is worth designing for rather than checking afterwards, because a palette that fails will fail everywhere at once. Getting your brand colour into a form you can reason about is the first step; our colour code converter gives you HEX, RGB, HSL and OKLCH for any colour, and OKLCH is the one that makes lightness comparable across hues.
5. SEO bolted on after launch
Search visibility is mostly decided by things designers control: the heading hierarchy, whether page titles describe the page, whether images have alt text, whether the URL structure survives the redesign, and how fast the thing loads. None of that can be sprinkled on at the end by a specialist.
The single most expensive mistake is changing every URL in a redesign without a redirect map. The second is shipping a page whose <h1> is a logo. Agree both before the build, and give whoever owns search a look at the wireframes rather than the staging site.
6. Security treated as someone else’s problem
For most design‑led sites the risk is not a sophisticated attack; it is an abandoned plugin, an unpatched CMS and a form that emails unvalidated input. Keep the plugin count low and check each one is still maintained before you add it. Serve everything over HTTPS, keep the platform patched on a schedule rather than when something breaks, and make sure whoever inherits the site knows they own those updates.
7. Responsive design that only works at three widths
Designing three fixed comps — phone, tablet, desktop — guarantees the layout breaks between them. Design for the ranges, not the devices: decide what happens as a container gets narrower, and let the content dictate the breakpoints rather than a list of popular screen sizes.
Test at 400% browser zoom as well as on a phone. It is a WCAG requirement, it catches horizontal scrolling nobody meant to ship, and it is the fastest way to find layouts that were secretly fixed‑width. Long labels are where this usually shows up first — and if you are tempted to solve a cramped axis by turning the text on its side, read our note on clockwise versus counter‑clockwise rotated text before you do.
8. A site that looks good and does not convert
A beautiful design that does not move anyone to act has failed at its job. This usually traces back to the design team never being told what the page is supposed to achieve, so the page ends up describing the company rather than helping the visitor decide.
Give every significant page one primary action. Put the evidence a sceptical person needs — pricing, proof, an answer to the obvious objection — above the point where they have to commit. Write the call to action as the verb the user would use. And measure it, because conversion arguments settled by opinion tend to get re‑litigated every quarter.
Catching these before launch
Five of the eight — performance, accessibility, SEO basics, responsive behaviour and broken structure — have a mechanical component that a tool can find faster and more reliably than a person reviewing pages by hand. That will not tell you whether your navigation matches how users think, but it clears the noise so your review time goes on the questions that actually need judgement.
We have compared the current options in our roundup of the best AI website audit tools, which covers what each one genuinely detects and where it overstates its results.
Want a starting point right now? Run a free audit of your site — it checks the contrast, alt text, form labels, heading structure and performance signals described above, and gives you a prioritised list you can take into your next sprint. No sign‑up needed.
Checklist
- Can you state the audience and their goal in one sentence each?
- Is there a performance budget, agreed before the stack was chosen?
- Has the navigation been tested with real people, without visual design?
- Does the palette pass contrast before it reaches a component?
- Is there a redirect map for every URL that changes?
- Does the layout survive 400% zoom without horizontal scrolling?
- Does every key page have exactly one primary action?
None of these challenges is new, and none of them is solved by working harder at the visual stage. They are solved by deciding earlier. For the practical side of that, our design tools cover the colour, layout and content jobs that come up on almost every project.
