Website migration SEO: how to keep rankings and AI visibility

Every migration checklist I have read in the last decade optimises for the same thing: do not lose rankings. Map the URLs, ship the redirects, keep the titles, watch Search Console for three weeks, breathe out.

That checklist is still correct. It is also now incomplete, because rankings are no longer the only place your visibility lives.

A growing share of your buyers ask an assistant instead of a search engine. The mechanics that keep you visible there are not the mechanics a redirect map protects.

I went through this on my own site this month. The surprise was not that AI visibility moved during the migration. It was that I would have had no way to know, because almost nobody captures the measurement that would tell them.

This guide covers website migration SEO end to end, with the AI visibility steps most checklists leave out.

Types of website migration and their risk

Not every migration carries the same risk. The more that changes at once, the more can break, for rankings and for AI citations.

If you are combining several of these, such as a rebrand with a new platform and new URLs, treat it as the highest-risk row. Spread the changes out if you can.

Why AI visibility breaks differently

Organic rankings recover after a migration because Google has an explicit, well-documented mechanism for transferring signals between URLs. You tell it a page moved, it consolidates, and within weeks the new URL inherits most of what the old one had.

AI answer systems have no equivalent guarantee, for four reasons.

They see your site on their own schedule. AI crawlers are separate from Googlebot, obey their own directives, and re-crawl far less predictably. A page can be live and correct for weeks before the system that cites you has looked at it again.

If your migration also changed robots rules or moved to a platform with different defaults, you may have quietly cut access entirely. That is a different failure, and the one I would check first.

A citation is attached to a URL, not to a domain. When an assistant cites you, it names a specific page. Redirect that page and the citation does not follow it the way authority follows a 301. It has to be re-earned by the new URL, on the new URL's own merits.

Third parties still point at the old address. Every roundup, forum reply, directory listing and newsletter archive that links to your old URLs is a source those systems read.

Your redirect works for a crawler following the link. It does nothing about the text around that link, which describes a page that no longer exists in the form it described.

Your entity data is easy to fracture. Name, title, company, bio and the profiles that corroborate them let a system be confident it is talking about you.

Migrations scatter these: a new about page, a changed job title, a bio rewritten in three places. Rankings barely notice. Entity confidence does.

Mapping migration risk to the Visibility Stack

The four layers of the Visibility Stack fail in a migration in a specific order. The order matters because the cheap failures are the ones people skip.

Access fails instantly and completely. New platform, new robots.txt, a staging block that shipped to production, a CDN rule, JavaScript rendering on a route that used to be static.

This is binary, it happens on day one, and it is the only layer where a single mistake wipes out everything above it. Check it before anything else.

Structure fails silently. Schema that lived in a plugin on the old platform and was never rebuilt. Heading hierarchies flattened by a new design system. Answer-first opening paragraphs edited into brand copy during the redesign.

Nothing errors. The page just becomes harder to extract a clean answer from.

Entity fails gradually. Your title changed on the new about page but not on LinkedIn. Your organisation name is written three ways across the new site. Nothing is wrong; nothing quite agrees either.

Citation fails last and recovers slowest, because it depends on the other three being right first and on the rest of the web catching up second.

The redirect decision AI search changes

Here is the one place where standard website migration SEO advice actively works against you.

Consolidation is good practice. Ninety thin old posts collapsing onto nine strong new pages is the right architecture, and I recommended exactly that for my own archive. For rankings it is close to free.

For citations it is not free, because assistants cite at the level of the question, not the topic.

If your old page answered "what should be on a website redesign checklist" and you redirect it into a broad pillar that mentions redesign in passing, a system looking for that answer now finds nothing specific enough to quote. The link equity arrived. The answer did not.

So apply one extra test to every consolidating redirect: does the target page contain a self-contained answer to the question the source page answered?

If it does not, you have two honest options: add that answer as a section on the target, or keep a dedicated page. What you should not do is assume the redirect covered it.

The related discipline is boring and non-negotiable: one hop, never a chain, and never a blanket redirect to the homepage.

Every retirement that genuinely has no successor should be served as a clean 410 rather than a fake 301. A redirect to an irrelevant page is a worse signal than an honest gone.

Website migration SEO checklist, with the AI steps

The checklist below combines the standard website migration SEO steps with the AI visibility steps. The AI steps are marked, because they are the ones most teams skip.

A downloadable version is in the website migration checklist. Google's own guidance on site moves is the reference for the ranking side.

What to do, in order

Before you migrate

This is the only step you cannot recover later.

Capture a baseline. Write a fixed set of prompts covering your category questions, your competitor set, your brand, and the specific questions your best pages answer. Twenty is enough; twenty-two is what I use.

For each one record three things: whether you appear at all, which exact URL is cited if any, and how you are described in the answer.

Then lock the wording and never edit it. A baseline you keep tweaking is not a baseline. Run it across each surface you care about separately, because they disagree with each other and the disagreement is informative.

Do this before the new site goes live. Once it is live, the comparison you needed no longer exists, and no analysis afterwards will reconstruct it. The method is in measuring brand visibility in AI answers.

During the build

Verify crawler access on the new platform against the actual production robots rules, not the staging ones. See are you blocking AI crawlers.

Confirm that every indexable route renders its content in the initial HTML. If a page needs JavaScript to show its own body, assume some systems will not see it. The detail is in technical SEO for AI crawlers.

Rebuild schema deliberately rather than assuming the new platform emits it. Keep the answer-first paragraph at the top of every page that had one, and resist the redesign instinct to replace it with a headline and a hero image.

Pull identity fields from one source. One file, one bio, one job title, propagated everywhere. Then make it match your LinkedIn, your YouTube channel and your X profile exactly.

After launch

Re-run the frozen prompt set on a schedule (I use day 7, day 30, day 60 and day 90) and record the same three fields.

Chart mention rate and citation rate separately from organic traffic, because they move on different timelines and averaging them hides both.

How long a website migration takes to recover

Expect rankings and AI visibility to recover on different clocks.

Rankings. Google's guidance is that a medium-sized site can take a few weeks for most pages to be processed at their new addresses, and larger sites longer. Some fluctuation is normal even when the migration is clean.

AI visibility. The recovery is uneven. Live-retrieval systems come back first. Anything reading from an older snapshot or training data takes considerably longer, so the fastest and slowest surfaces can sit weeks or months apart. See how often AI updates its training data.

Track the surfaces separately rather than averaging them. Otherwise the slow recovery hides inside the fast one and you call the migration done before it is.

Common website migration mistakes

1. No AI baseline. The one step that cannot be done afterwards.
2. Staging robots rules shipped to production. Blocks every crawler on day one.
3. Client-side rendering on pages that used to be static. Invisible to many AI crawlers.
4. Redirect chains. Old redirects pointing at URLs that now redirect again.
5. Blanket redirects to the homepage. Treated as soft 404s, and useless to a reader.
6. Consolidating away specific answers. The equity arrives; the answer does not.
7. Schema left behind on the old platform. Nothing errors, so nobody notices.
8. Identity details changed in one place only. A new title on the site, the old one everywhere else.
9. Declaring success at 30 days. AI surfaces can still be recovering.

The part most teams get wrong

They treat AI visibility as something to check after the migration. The honest answer is that after the migration there is nothing to check against.

I have watched capable marketing teams run flawless technical migrations (clean redirect map, no chains, rankings back inside a month). They were then unable to answer their CEO's question about whether the company still shows up when someone asks ChatGPT for a recommendation in their category.

Not because the answer was bad. Because they had no before.

The measurement costs an afternoon. The migration costs a quarter. Skipping the afternoon is the expensive decision.

If you are planning a redesign or replatform in the next two quarters, the baseline capture is the piece to schedule first. It is the first thing I run in an AI search diagnostic for exactly that reason.

That baseline is my side of the work. The rebuild itself is delivery, and when the move is also a platform change, my agency runs WordPress to Webflow migrations with the redirect and rendering work built in.

The takeaway

Redirects move authority. They do not move citations. Plan the website migration for both.

Verify access on day one. Preserve the specific answers your old pages gave rather than just their URLs. Keep your entity data consistent across the site and the profiles that corroborate it. And before anything else, freeze a measurement you can compare against ninety days later.

The underlying model is the four layers in SEO in the AI era. If you suspect the access layer is already the problem, start with whether you are blocking AI crawlers instead.

This article is part of the SEO in the AI Era: The Complete Guide guide.