Our site looked great at launch and looks tired eighteen months later — what happened?

We go back into sites we designed all the time — for an extension, an audit, a second phase. The pattern is the same nearly every visit. The pages built during the project still look like the project. The pages built after it mostly don’t.

On a site that launched with thirty pages and now has a hundred and ten, the eighty new ones are the exposed surface. Nobody edited the design. Somebody just needed a page for a webinar on Thursday and built one with the tools in front of them.

That is normal, and it is not the client’s fault. When it happens badly we now treat it as our failure. We shipped a set of pages and a design language, and we did not ship a way for someone else to keep making pages in it. A launched site is a snapshot of the pages that existed on launch day. Everything after that is authoring.

Each of these came from someone trying to finish a task with the tools they had. None of them is a mistake anyone should feel bad about.

• Spacing drift. Gaps between sections that are 40px, 44px, 60px and 72px on four consecutive pages, where the design used two values. It happens because the page builder lets someone type a number, so they type a number — one that looked right on their laptop at four in the afternoon.
• The colour that isn't in the palette. A hero band in a blue that is nearly, but not, the brand blue. Usually pulled off an old deck or sampled from a stock photo. Someone needed a variant for a campaign, there wasn't one, so they made one.
• Heading inflation. Every new page opens with the largest heading style, because the largest heading style looks the most important. Hierarchy flattens. Everything shouts and nothing leads. Nobody told the author that the top heading level is structural, not decorative.
• Component fatigue. The original site had six section types. There are now nineteen, and eleven of them are one-off variations built for a single page that is still live. Each request was reasonable on its own. Nobody was counting.
• Image entropy. Art-directed photography at the top of the site, then compressed conference photos from three years ago and screenshots with a browser chrome bar still in them further down. The good images were commissioned for the launch pages. No budget line ever existed for the ones after.

This takes an afternoon and a spreadsheet. It does not need a tool and it does not need us.

Read the result like this. Under about a fifth of pages off-pattern is ordinary wear and needs nothing. A third or more and the problem is the publishing path, not the pages — fixing the pages will just produce new ones. If the off-pattern pages cluster around one author, that is a training gap and it is cheap to close. If they are spread evenly across every author, the system is missing something people genuinely needed, and the missing thing is what you should build.

1. List every URL on the site and sort by publish date, oldest first. Not by traffic — date is the variable that matters here.
2. Mark the row where the redesign launched. Everything below that line is the exposed surface.
3. Open twenty of the post-launch pages at random. For each one, note a single thing: does it use a section type that existed at launch, yes or no. Count the noes.
4. Screenshot the top 1,200px of your ten highest-traffic post-launch pages, put them side by side at 25% zoom, and squint. You are not reading them. You are checking whether they look like they came from the same place.
5. Ask who published each of the twenty. The answer is usually two or three people, and they are usually not on the same team.

The cheap fixes do most of the work. We have never once found decay that needed a rebuild to prevent.

• Free, this week: write a one-page note called how to build a page here. Which section types exist, what order they usually go in, what the two spacing values are, and who to ask when none of it fits. One page. Not a design system. Most of the drift we find would have been prevented by this note.
• Free, ongoing: name one person who publishes, or who at least looks before anything goes live. Not an approver of copy — an approver of construction. Consistency survives when it belongs to a person.
• A day of build work: remove the free-text fields. If the CMS lets an author type a hex code or a pixel value, eventually one will. Replace them with a dropdown of the three legitimate options. Constraint in the editor beats documentation about constraint.
• A recurring half-day: run the census quarterly, with one rule. Every off-pattern page either gets fixed or gets promoted into a real, reusable section type. Promotion is not a defeat. A pattern used four times is a pattern.

Do not buy a second redesign for this. A site with drifted spacing and eleven stray section types has a maintenance problem. Rebuilding it produces a beautiful site with the same maintenance problem eighteen months later. We have watched that happen. A redesign is the expensive way to fix a documentation gap.

Do not buy a retainer for it either, unless the volume justifies one. If you publish two pages a month, a retainer is us charging you every month to do something a dropdown would do for free.

Some drift is fine, and chasing all of it wastes everyone’s time. Pages nobody visits do not need to be beautiful. Fix the twelve pages that carry your traffic and leave the archive alone.

If the same wrong thing has been built five times, the design was incomplete. That one is ours, not yours. The correct fix is to add the missing pattern, not to keep correcting people who needed it.

A redesign is not judged by how it looks the week it launches. It is judged by how the fortieth page someone else builds on it looks. If your census came back ugly, we will sit and run it with you.