Most teams that run a website migration have a familiar checklist: map the redirects, protect rankings, watch organic traffic for a few weeks, then move on. That checklist was built for a world where the main risk was dropping a few positions in a list of blue links. Today there is a second audience reading your site, and it is far less forgiving of broken context. AI answer engines and assistants pull facts, definitions and recommendations from the pages they can crawl and trust. A redesign that scatters those facts, hides them behind scripts or changes the URLs they were found at can leave an assistant repeating stale information about you, or leaving you out entirely.
This is an operator’s checklist for running a redesign, CMS replatform or domain change without losing the visibility you have earned in AI answers.
Why migrations hit AI visibility differently
Traditional search engines are generally good at following redirects and reassigning signals to new URLs over time. AI systems add a few complications. Some answers are generated from content an engine retrieved recently; others lean on what a model absorbed from older crawls and third-party sources. During a migration, both can drift out of sync with your new site. If your pricing page moves, your product names change and your About page gets rewritten in the same month, assistants may hold several conflicting versions of you at once.
Migrations also tend to change things that never show up in a rankings report: how much text is server-rendered, whether answers sit in plain HTML or inside tabs and accordions, which pages carry structured data, and whether long-standing explainer pages survive the content cull at all.
Before launch: inventory what AI engines actually use
1. Build a prompt baseline
Before anything changes, write down the questions customers ask assistants about your category, your brand and your products. Run them across the assistants your buyers use and save the answers along with any cited sources. This gives you a “before” picture you can compare against after launch. Without it, you will only find out about losses anecdotally.
2. List your cited and quoted pages
From that baseline, note which of your URLs are cited or clearly paraphrased. These are often not your highest-traffic pages. Old glossary entries, a detailed FAQ, a methodology page or a long-lived comparison article can matter more to AI answers than the homepage. Tag these as protected pages that cannot be cut, merged or rewritten without a deliberate decision.
3. Capture your key facts
Write a short fact sheet: company description, product names, plans, locations, leadership, founding details and anything else assistants repeat about you. During the redesign, this becomes the source of truth that every new template and rewritten page must match.
During the build: protect the structure, not just the URLs
- Keep answers in rendered HTML. New front-end frameworks often move content into client-side components. Check that the main text of each protected page is present in the initial HTML response, not only after scripts run.
- Do not hide the answer. Accordions and tabs are fine for design, but make sure the text exists in the page source and that headings still describe what each section answers.
- Carry structured data across. Organization, Product, FAQ, Article and breadcrumb markup are frequently dropped when templates are rebuilt. Audit the old templates, list every schema type in use and confirm each one is reimplemented and still matches the visible content.
- Preserve authorship and dates. Bylines, author pages, reviewed-by notes and last-updated dates often disappear in a redesign. They are part of how engines judge whether a page is trustworthy and current.
- Check crawler access on the new stack. New hosting, CDN or bot-management settings can block AI crawlers you previously allowed. Review robots.txt and firewall rules on staging before launch.
At launch: redirects with meaning
Map every old URL to the closest equivalent new URL, one to one wherever possible. Avoid sending retired explainer pages to the homepage; a generic destination breaks the link between the question and the answer. If a page is genuinely being retired, decide whether its content should live on elsewhere, and redirect to that specific section or page.
If you are changing domains or product names at the same time, update your entity footprint in the same window: business profiles, social bios, marketplace and review listings, partner pages and press boilerplate. Assistants often corroborate facts across these sources, and a half-finished rename produces confused answers for months.
After launch: re-test and correct
- Rerun the prompt baseline at regular intervals for the first few months and compare answers and citations against your “before” snapshot.
- Watch for stale facts. Old prices, discontinued products or a previous company description showing up in answers usually means an old page, PDF or third-party listing is still being read.
- Check AI referral traffic to protected pages in analytics. A sudden drop on a specific URL is a faster signal than waiting for answers to change.
- Fix sources, not symptoms. When an answer is wrong, find the page it came from and correct or redirect it rather than adding another page that contradicts it.
Make it part of the migration plan
The simplest change is organizational: add AI visibility to the migration’s definition of done, alongside redirects and analytics tags. Give one person ownership of the protected-page list and the prompt baseline, and make their sign-off a launch requirement. If your team lacks the bandwidth to run that work alongside a rebuild, it can be worth bringing in a specialist generative engine optimization team to handle the audit and post-launch monitoring.
A redesign should make your site easier to understand for every reader, human or machine. With an inventory of what AI engines already rely on and a plan to carry it across, it can.