The call usually arrives three weeks after a launch everybody was pleased with. The site looks far better and enquiries have halved. Nothing is broken, nothing errors, and the traffic is simply gone.
Redesigns rarely lose traffic because of the design. They lose it in two places: the addresses changed and nobody mapped them, or the pages that earned the traffic came back shorter. Both are preventable with a list written before anyone opens a design tool.
Start with an inventory, not a moodboard
Before anything is designed, produce one table: every address on the current site, with how much traffic it receives, how many conversions it produces, and how many other sites link to it.
That table decides the project. It shows which twenty pages carry most of the value — it is nearly always a short list — and it stops the common mistake of redesigning the homepage with enormous care while quietly deleting the article that brings in a third of the visitors.
It also produces the list of pages that must survive with their content intact, which is the second half of this article.
The redirect map
If addresses change, every old one needs a permanent redirect to the closest equivalent page. Three rules make the difference between a smooth move and a bad quarter.
- One to one. Sending everything to the homepage is the classic shortcut, and search engines treat it roughly as a deletion. Point each page at its true successor, or at the nearest genuinely useful page.
- No chains. Old address to intermediate to final costs you on every request and sometimes breaks entirely. Rewrite the chain to end in one hop.
- Keep them. Not for a month — indefinitely. Those redirects carry links from other sites that you cannot ask anyone to update, and removing them in a year's tidy-up quietly deletes the value again.
Check the awkward variants while you are there: with and without a trailing slash, uppercase, old query parameters from campaigns, and the language prefixes if you have them.
Content parity is the part everyone skips
A page that ranked with nine hundred words and came back with two hundred beautiful ones has lost the reason it ranked. Redesigns are usually accompanied by an editorial instinct to tighten everything, and on the pages that earn search traffic that instinct is expensive.
Keep the substance on the pages the inventory flagged. Rewrite them if they were badly written, but do not remove the answers people were arriving for. Headings, the questions covered, the specifics — those are what the traffic came for.
What else silently breaks
- Titles and descriptions regenerated from a template, replacing ones that had been tuned.
- Structured data that existed on the old product pages and was never rebuilt.
- Internal links pointing at the old addresses, so every visit takes an extra hop.
- Image addresses changed wholesale, which drops whatever image search sent you.
- Language annotations between translated versions, if you have more than one.
- The staging block. The instruction that kept the test site out of search, shipped to production with everything else. This one is common, catastrophic and invisible for weeks — check it within the hour of going live.
What to keep from the old site besides the pages
Two assets survive a redesign only if somebody deliberately carries them over. The first is your analytics history: keep the same property rather than starting a clean one, or you lose the ability to compare before and after, which is the only way to know whether the redesign worked.
The second is the tracking and verification setup — search console ownership, ad platform tags, conversion goals. These live in odd corners of the old site and are routinely lost, and their absence is discovered a month later when somebody asks for a report that cannot be produced.
Launch day, in order
Publish, then immediately take a sample of thirty old addresses from the top of the inventory and request each one. You are looking for a single permanent redirect landing on a real page. Then confirm the site is crawlable, submit the new sitemap, and watch the server logs for addresses returning not-found — those are the redirects you missed, and the logs will tell you within hours rather than weeks.
Do not skip the cache. A page can be perfect in the code and stale everywhere else; we wrote about the five caches between you and your users after losing an afternoon to exactly that.
What normal looks like afterwards
A dip in the first weeks is ordinary. Search engines need to recrawl, follow the redirects and re-evaluate. A shallow decline that recovers over a few weeks is the expected shape.
What is not normal: a sharp fall that keeps falling, traffic loss concentrated on a handful of previously strong pages, or a rise in not-found errors. Those are faults, not patience problems, and the longer they run the more they cost. Watch the numbers from the monthly five weekly for the first month rather than monthly.
If you have already lost it
Work in this order. Export the old addresses from whatever record survives — analytics history, an old sitemap, the search console, a crawl of an archived copy — and find which of them now return not-found. Add the missing redirects, highest traffic first. Then compare the surviving pages against what they used to contain and restore the substance that was cut. Only after those two should anybody discuss the design.
Recovery is usually possible and usually slower than the loss. Which is the argument for the inventory: an afternoon of listing addresses before the project starts is the cheapest insurance in a redesign.