Picture a SaaS dashboard where the "no data yet" screen for brand-new accounts is a single gray box that says, literally, "No data." Nothing else. New users hit that screen on day one, during their first five minutes in the product, and nobody ever designed it. It just happens, by default, because someone forgot to.
That's a common pattern. Teams spend weeks polishing the happy path, then ship whatever the framework generates for zero-data states, timeouts, and 404s. Those "edge cases" aren't edge cases at all, they're screens real users hit constantly: a fresh account, a search with no matches, a broken link from an old bookmark.
TL;DR: Empty states and error pages are frequently a new user's first real screen, not a rare exception. Nielsen Norman Group frames them as "teachable moments", well-designed ones explain what happened and offer a next step instead of leaving a blank wall.
Why Do Empty States Get Ignored During Design?
Because they don't show up in most Figma mockups. Designers build the "populated" version of every screen, five rows of data, a full dashboard, a rich profile. Then engineering ships the zero-state as whatever the component library defaults to, usually just a placeholder div with no thought behind it.
Here's the thing teams learn the hard way: empty states happen more often than the "happy path" for specific user segments. A brand-new user sees zero data on every single screen until they've done real work in your product. A power user who filters aggressively hits "no results" constantly. If you only design for the full state, you're designing for a minority of actual screen views.
Empty states happen more often than the happy path for entire segments of your users, and most teams never design them.
What Makes a 404 Page Actually Work?
Every 404 needs four things: clear language that this page doesn't exist, working navigation back into the site, a search bar if your site has meaningful depth, and ideally a shortlist of popular pages (UXPin, 2026). Pull that shortlist from real analytics data, not intuition: whatever your top five most-visited URLs are this quarter.
A good 404 page is also a safety net during a messy migration. When old URLs break for a while before redirects catch up, a custom 404 with a working search bar and a "browse by category" module gives visitors a path forward instead of a dead end, and a dead end is what turns a broken link into a lost visit.
Rather than argue the four rules in the abstract, here's each one as a live page that gets it right. These were captured in August 2026 and checked again on October 16, 2026, so they're current designs, not the versions everyone keeps reposting from 2019. Where a page changed in between, the text says so.
Cloudflare handles rule one better than almost anyone: "It may have been moved, renamed, or no longer exists." Three plausible causes, named. Most pages stop at "we can't find it", which leaves the user guessing whether they mistyped or the site is broken.
Airbnb is my go-to reference for rule four. Under "Oops!" sits a labelled list: Home, Search, Help, Traveling on Airbnb, Hosting on Airbnb, Trust & Safety, Sitemap. Those are the seven things a lost visitor was probably trying to do. It also prints "Error code: 404" as text, which support agents need and which most brands hide to protect the vibe.
Vimeo is the cautionary case for the search rule, and it moved the wrong way between the two captures. The August capture had search twice: a field in the header and a second, wider one in the error block, right where your eyes already are. The October 16 capture has neither. The page still says "Make sure you've typed the URL correctly, or try searching Vimeo," and then offers a cyan "Why am I seeing this?" button and no field to type into. With millions of items, search is the only realistic recovery path, so telling people to use it without handing them the box is the one instruction a 404 shouldn't give.
Dropbox is the cheapest version and it still works: six task-oriented links under one sentence, a small drawing, no clever copy. If your team has one afternoon, ship this one.
Which 404 Pages Get the Voice Right?
The four rules are the floor. What separates a 404 you remember from one you tolerate is tone, and tone is where teams either freeze up or overshoot.
Webflow takes the blame in four words: "And that's on us." Nothing pinned on the user, one button back home, a single cyan headline on black. It reads as confident rather than apologetic, which is a hard trick in an error state.
Starbucks is a reminder that voice is a decision someone can reverse. The August capture was built around a spilled Vanilla Latte, with a line saying it was "just a little spilled Vanilla Latte" and an illustration to match. By October 16 both were gone. What's left is a green "404", "We can't find the page you're looking for," the same three links (place an order, check a gift card, return to the home page) and an Error ID in small type at the bottom. The joke went and the part that does the work stayed, which is a fair argument for writing the links first and the voice second.
Basecamp goes the other way and I still like it. One card: "Hmm, that isn't right", then in red, "You may have typed the URL incorrectly", then advice specific enough to act on, check the spelling and capitalization. It's the only page here that treats the commonest cause of a 404 as something the user can fix in three seconds.
The Guardian is the only page in this set that offers a way to tell them about it: "If you believe this to be an error please report it." For a news site with decades of archive URLs loose on social media, that link is a free broken-link crawler staffed by readers. Why more publishers skip it, I genuinely don't know.
Kualo, a UK host, turned theirs into playable Space Invaders where the aliens spell the company name and scoring over 1,000 points wins a coupon for 15% off a new shared hosting plan, according to the page's own script. It's the most-copied joke 404 on the web. But look at what it keeps: the error code top-right, the controls, and a "click to go back to where you were" link under the headline. The game is additive, never a replacement for the exit.
Steve Madden is the commerce pattern, and it's smarter than it looks. The error block is almost nothing: "404", "Page not found", "Continue shopping". But the page keeps the whole store around it, nav across ten sections, working search, cart count, six-column footer. On a shop, the furniture is the recovery path and the error block only has to stay out of the way.
How Should Zero-Data States Be Written?
Answer why, then answer what next. Those are the two jobs of an empty state's copy, and skipping either one is the most common mistake I see in reviews.
"No invoices yet" is fine as a headline, but it's incomplete. Pair it with "Create your first invoice to start tracking payments" and a button right there. That single sentence turns a dead-looking screen into an active prompt, and it costs you nothing beyond ten minutes of copywriting.
A few concrete patterns worth stealing:
- New account, no data: explain what this section is for, then offer the first action (a template, a quick-start wizard, a sample dataset toggle).
- Filtered search, zero matches: name the filters that are active, offer a one-click "clear filters" action right next to the message.
- Completed task list: celebrate it. "All caught up" with a small positive visual reads better than a blank list that looks broken.
- Permission-gated empty state: tell the user why they can't see data (role, plan tier) and who to ask, don't just show nothing.
Is a generic "No results found" ever acceptable? Only as a fallback, never as your primary design. It tells the user nothing about why or what to do about it.
| Empty state type | What to explain | Next action to offer |
|---|---|---|
| New account, no data | What this section is for | First action: template, quick-start wizard, sample dataset |
| Filtered search, zero matches | Which filters are active | One-click "clear filters" |
| Completed task list | Celebrate it | Small positive visual, not a blank list |
| Permission-gated | Why the user can't see data | Who to ask |
What Should Server Error Pages Actually Say?
Never show a raw stack trace or a generic "Something went wrong" that swallows the whole page. Scope the failure. If one API call fails on a dashboard with six widgets, only that widget should show an error state, the other five should keep working. Otherwise a single flaky endpoint can take down the entire page behind a modal. Component-level error boundaries limit the damage to one card.
Error copy should tell the user three things: what happened in plain language, whether it's their fault or the system's, and what to do next (retry, contact support, wait). "Error 500" tells a user nothing useful. "We couldn't save your changes, try again in a moment" does.
Slack's page is the cautionary one, and I say that as someone who admires the illustration. Ask for a URL that doesn't exist and you get "There's been a glitch..." followed by "We're not quite sure what went wrong." That's server-error copy on a navigation failure. The system knows exactly what happened, the page isn't there, and calling it a mystery invites the user to sit and refresh. Beautiful floating islands, vague sentence, and the only real exit buried mid-paragraph. The wording was identical in the October 16, 2026 recheck, but the navigation bar and footer that framed the page in August have been removed, so the card and the Help Center link are now all that's left of the way out.
How Do You Test These Screens Before Launch?
Force them. Most teams never see their own empty states because they're testing with seeded demo data that always has content. Before you ship, manually trigger every zero-data path: a brand-new account, an aggressive filter combination, a deliberately broken URL, a timed-out API call.
Keep a checklist for this on every project. It's one of the cheapest QA passes there is, and it catches things nobody would ever notice in a demo. Is your team actually testing these paths, or just assuming the framework default is fine? If you're not sure, that's your answer.
The image at the top of this article is our own 404 page, because writing the rules and then failing them would be embarrassing. Plain message, search field, navigation, four article cards, the error code as real text rather than a decorative graphic, and a report link that arrives prefilled with the URL that broke. Auditing our own page against this list turned up three things to fix, which tells you something about how easy the drift is.
The shortlist was the worst of them. It's now built from our Search Console export, ranked by impressions over the last 28 days, with every URL checked against the published collection first and a newest-first fallback so the grid never renders empty. Until this week it sorted by publish date and called itself "most read", which is exactly the intuition-over-analytics shortcut I'm warning about here.
The pages that get the least design attention are often the ones a meaningful slice of your users see first. Treat them with the same care you'd give the dashboard, and you'll fix a real drop-off point that most competitors never bother to touch.
Empty states are one small piece of a bigger discipline: our visual hierarchy in UI design principles guide covers the broader system these micro-states have to fit into, and our dashboard design patterns piece shows what the fully-populated version of these same screens should look like.