You can migrate or redesign a site without losing AI visibility, but only if you treat it as a search migration with three extra checks: what AI assistants cite today, whether the new stack lets AI crawlers in, and whether the content that earned citations survives. AI answers depend on search indexes recrawling your new URLs, on third-party links that point to old ones, and on pages that still say the same things clearly. This post walks through the baseline, the redirect map, content parity, pre-launch checks and monitoring.
Why migrations put AI visibility at risk
Migrations tend to change several things at once, and each one can independently cut citations. The common causes:
- Broken or missing redirects. Cited URLs start returning 404s, and the index has to rediscover everything.
- Lost content. Redesigns trim "wordy" sections, which are often the FAQ answers, comparison tables and definitions assistants were quoting.
- New infrastructure, new bot rules. A new CDN, WAF or host may block or challenge AI crawlers by default. See CDNs and bot protection.
- Client-side rendering. A move to a JavaScript framework can leave crawlers that don't execute JavaScript with an empty page. See JavaScript rendering and AI crawlers.
- Dropped technical signals. New templates can lose structured data, canonicals, author information or dates.
- Old URLs everywhere else. Review sites, directories, Wikipedia, partner pages and press coverage keep linking to old URLs, and models trained on older crawls know those URLs too.
Step 1: Baseline before you touch anything
Record where you stand so you can spot losses quickly and prove recovery. At least four weeks before launch:
- Run your tracked prompt set and save mention rates, share of voice and cited URLs. See how to build a prompt set.
- Build a list of every URL on your site that AI assistants cite, from prompt tracking and citation analysis
- Export Search Console data, including the generative AI performance report's page list. See measuring AI Overviews and AI Mode in Search Console.
- Record AI crawler activity from server logs: which bots, how often, which sections. See AI crawler log analysis.
- Crawl the current site and save titles, headings, word counts, structured data and canonicals per URL
Mark cited URLs as protected. Their content and redirects get the most scrutiny.
Step 2: Build a one-to-one redirect map
Every old URL that has traffic, links or citations should redirect to its closest equivalent new URL with a permanent, server-side redirect. Google prefers permanent server-side redirects for site moves, and its guidance recommends keeping them in place for as long as possible, generally at least a year.
A redirect map is a simple table:
old_url,new_url,status,cited_by_ai
/pricing.html,/pricing/,301,yes
/product/features/reporting,/platform/reporting/,301,yes
/blog/2023/what-is-aeo,/blog/what-is-answer-engine-optimization/,301,yes
/resources/aeo-checklist.pdf,/guides/aeo-checklist/,301,no
/old-careers-page,/careers/,301,no
On Nginx, a map keeps hundreds of rules readable (the map block goes in the http context):
map $uri $redirect_target {
default "";
/pricing.html /pricing/;
/product/features/reporting /platform/reporting/;
/blog/2023/what-is-aeo /blog/what-is-answer-engine-optimization/;
}
server {
# ...
if ($redirect_target) {
return 301 $redirect_target;
}
}
Redirect rules that protect AI visibility:
- One hop only: no chains from old URL to interim URL to final URL
- No bulk redirects to the home page; search engines may treat those as soft 404s, and the specific page that was cited is gone
- Redirect PDFs and other files too, not just pages
- Preserve query-string handling for URLs that rely on parameters
- If a page is retired with no equivalent, return a 404 or 410 rather than redirecting to something unrelated
Step 3: Keep content parity
The content that earned citations has to survive the redesign, ideally word for word for the key passages. Compare old and new versions of every protected URL:
| Check | Why it matters |
|---|---|
| The answer-first opening paragraph still exists | It's often the passage assistants quote |
| FAQs, definitions and comparison tables kept as HTML text | Moving them into accordions loaded by script or images can hide them |
| Facts unchanged unless deliberately updated | Pricing, specs and claims should match your brand fact sheet |
| Structured data carried over | Organization, Product, Article, FAQ and breadcrumb markup |
| Author bylines and dates preserved | Expertise and freshness signals |
| Internal links to the page still exist | Orphaned pages are crawled less |
Then check what crawlers actually receive. curl -s https://staging.example.com/pricing/ | grep -c "Professional plan" is a quick way to confirm a key phrase is in the server-rendered HTML rather than injected later.
Step 4: Check crawler access on the new stack
Test bot access on the new infrastructure before launch, not after citations disappear. The checks:
- robots.txt on the new site is the production version; a staging
Disallow: /carried into production is a classic migration failure - No sitewide
noindexleft over from staging - CDN and WAF AI crawler settings reviewed on the new account or zone; new domains on some CDNs get stricter defaults
- Rate limits don't throttle crawlers during the post-launch recrawl surge
- Canonicals point to the new URLs, not staging or old URLs (canonicals and AI citations)
- New XML sitemaps list only new, canonical, 200-status URLs with honest
lastmoddates (XML sitemaps and lastmod) - If you publish an llms.txt file, its links are updated
Step 5: Launch-day and first-week actions
Launch is when you tell search engines about the move and start watching closely. On launch day:
- Deploy redirects and test the full map with a script that requests every old URL and checks for a single 301 to the expected target.
- Submit new sitemaps in Google Search Console and Bing Webmaster Tools.
- For a domain change, use Google's Change of Address tool (it isn't for HTTP-to-HTTPS moves) and Bing's Site Move tool, after redirects are live.
- Submit key changed URLs through IndexNow. See Bing Webmaster Tools and IndexNow.
During the first week, check server logs daily for AI crawler 404s, 403s and 5xx responses, and fix any gaps in the redirect map they reveal.
Step 6: Update the links you control off-site
Redirects catch old links, but updating your own profiles removes a hop and keeps third-party facts consistent. Update URLs on review platforms, business profiles, marketplace and integration listings, social profiles, Wikidata, partner pages, and press kits. If your brand name or positioning changed in the redesign, align those descriptions everywhere at the same time.
Example: a redesign with a URL restructure
This example is illustrative, with made-up details.
A SaaS company moves from /product/features/* to /platform/* and redesigns its marketing site. Its baseline shows 38 URLs cited by AI assistants across its prompt set, including the pricing page and six comparison pages.
What it does differently from a standard migration:
- Flags all 38 URLs as protected and assigns each an owner to sign off content parity
- Discovers that the new comparison template moves tables into image carousels, and rebuilds them as HTML tables before launch
- Finds that the new CDN zone has AI training and agent traffic blocked by default, and sets them deliberately
- Re-runs the prompt set weekly for eight weeks and compares cited URLs against the baseline
By week six, most citations point to the new URLs; two pages that lost citations turn out to have lost their FAQ sections, which are restored.
The bottom line
Migrations lose AI visibility for predictable reasons: broken redirects, trimmed content, new bot rules, client-side rendering and stale off-site links. Baseline what assistants cite, redirect every URL one-to-one with permanent server-side redirects, protect the content that earned citations, test crawler access on the new stack before launch, and monitor logs and prompts weekly until citations recover.
