---
title: 'Website migration checklist: five things that broke when we moved our site'
description: >-
  We kept every URL when we moved our site to a new stack, and five things broke
  anyway. Our website migration checklist, built from what each one taught us.
author: elroot_labs
date: '2026-11-18'
categories:
  - digital-infrastructure
---
To migrate a website without losing search traffic, keep every URL or redirect each old one, and then check everything the old site did that no page shows: structured data, analytics, verification and key files, background jobs, and anything the server read while running. On 16 November 2026 Elroot Labs moved elrootlabs.com to a new stack and kept all 42 URLs. Five things broke anyway, and none of them showed up by clicking around the new site.

| What broke                                                         | Live until | How we found it                                           | The check that catches it now                                   |
| ------------------------------------------------------------------ | ---------- | --------------------------------------------------------- | --------------------------------------------------------------- |
| Our other brand names were missing from the site's structured data | 17 Nov     | Asking why Google rewrote our brand name                  | A unit test on the schema                                       |
| Search-engine pings looked for a key file at the old site's path   | 17 Nov     | Porting our scripts to the new site                       | The script reads the key from the new site and stops without it |
| The blog listed 25 of 49 posts, and search did nothing             | 18 Nov     | Opening the blog in a real browser                        | The release check counts every post on the blog page            |
| Analytics recorded nothing at all                                  | 18 Nov     | Accepting cookies in a real browser and watching requests | A test on the security policy's allowed hosts                   |
| The template's identity files were served under our domain         | 18 Nov     | Cleaning up the test suite                                | A test fails if any file but security.txt is served             |

![Table card of our website migration checklist findings: five things that broke after a move that kept every URL, how each was found, and the check that now catches it.](/images/blog/elroot/figures/website-migration-checklist-what-broke.png)

## Keeping every URL was the part that went right

The old site was a Node app on shared hosting. The new one is a Next.js site on Cloudflare Workers, built from an open-source company website template. Every page address stayed the same, so there was no redirect map to get wrong, and we resubmitted the sitemap on the day of the move.

Google treats this as a move without URL changes, and its [guidance for that case](https://developers.google.com/search/docs/crawling-indexing/site-move-no-url-changes) is short: lower your DNS time-to-live ahead of time, confirm Googlebot can reach the new setup with the URL Inspection tool, watch the server logs on both sides, and expect a temporary drop in crawl rate that recovers. Our pages kept being served at the same addresses. Every problem below sat underneath them.

## The things no page shows are what broke

Each of the five came from a different layer of the old site.

**Structured data.** The old site told search engines that "Elroot" and "ElrootLabs" were other names for the company, using the schema.org [alternateName](https://schema.org/alternateName) property. The new template built its own schema and left that out, and it also attached the company's profile links to the founder. We found it on 17 November while working out why Google rewrote searches for our brand to a different company's name. Restoring it took one commit; a test now fails if the names go missing again. We wrote up the whole schema check in our [schema markup audit](/blog/schema-markup-audit).

**A background job.** We tell Bing and other engines about new pages with [IndexNow](/blog/indexnow-nextjs), which needs a key file on the site. The script still looked for the file where the old site kept it, and the new site answered that path with a 404. The fix was a path, and the script now reads the key from the new site's public folder and refuses to run without it.

**An assumption about the server.** The blog index fetched its list of posts from an internal route that read the post files from disk while the site was running. That worked on a laptop. Cloudflare Workers have no file system at run time, so on the live site the route failed every request from 16 to 18 November, and the blog showed 25 of 49 posts with a spinner that never stopped. We now pass the full list into the page when it is built.

**A third-party call.** After the move, Google Analytics recorded nothing. Our new security policy allowed requests to `*.analytics.google.com`, and a wildcard like that does not match the bare `analytics.google.com` that the tracker actually posts to. It also left out a second Google host. After accepting cookies in a real browser, we watched six analytics requests get blocked and none get through. Both hosts are allowed now, and a test fails if either is removed.

**Leftovers from the template.** The template came with seven identity files in its `.well-known` folder that belonged to its original owner, including one that would have let them publish MCP servers under our domain and another that announced their company name. Our site served them for two days. We found them while cleaning up the test suite, removed them, and added a test that fails if any file other than our own `security.txt` is served from that folder.

## Test the new site the way the server runs it

The blog failure is the one to learn from, because every check before launch passed. The development server reads files from disk, the build succeeded, and the pages rendered. Only the real platform, with no disk, showed the problem. Our rules file now says it plainly: anything a route reads must be bundled at build time. It also records that the local server worked the whole time every live request failed.

The general rule for a migration: before you switch, run the new site on the actual host it will live on, at its real address if you can, and open the pages that load data rather than only the static ones. Our release check now loads every page in the sitemap on the staging copy, on the same platform as live, and fails if the blog does not list every post.

## Diff the invisible parts before and after

Four of the five were things the old site did quietly. The way to catch them is to list them on the old site before you switch, then check the same list on the new one:

- The structured data on the home page and one inner page, compared field by field.
- Every file under `/.well-known/` and any other file a service checks for, such as verification and key files.
- Every scheduled job or script that talks to the site, and the paths it expects.
- Every third-party host the site sends requests to, checked in a real browser after accepting cookies.
- The security headers, especially the content security policy, which fails silently from the visitor's side.

None of this is visible in a screenshot, which is why a "looks fine" check after launch missed all five. For the code-level fixes our AI-built site needed before it held up, see [vibe coding to production](/blog/vibe-coding-to-production).

## Website migration checklist

Before the move:

1. Export the list of every URL and decide, for each one, whether it stays or gets a redirect.
1. Record the structured data, `.well-known` files, security headers and third-party hosts of the old site.
1. List every background job, script and integration that reads from or writes to the site.
1. Lower the DNS time-to-live about a week ahead.
1. Run the new site on its real host and open the pages that load data.

On the day:

1. Switch, then resubmit the sitemap and inspect a few key URLs in Search Console.
1. Ping IndexNow once the new build is live, if you use it.
1. Load every page in the sitemap at a desktop and a phone width.

In the first week:

1. Compare the structured data and `.well-known` files against your old record.
1. Accept cookies in a real browser and confirm analytics requests reach their hosts.
1. Read the server logs for repeated errors, and watch the index coverage report.
1. Run each background job once by hand and check its result.

If you are planning a move or a rebuild and want it done without losing what the old site quietly did, [tell us about your site](/contact).
