How Site Migrations and Domain Changes Affect Your AI Citations

Most migration checklists are built for Google. Preserve rankings, keep backlinks, set up redirects, resubmit the sitemap. That checklist doesn’t fully protect something else that’s harder to see going wrong: the citations an AI tool has already built up for your old domain and URLs.

A migration that scores perfectly on a traditional SEO audit can still quietly break AI visibility, because AI tools draw on your content through two different mechanisms, and a standard migration plan only protects one of them.

Two different ways AI tools know about your site

Training data. Models like ChatGPT and Claude learn from a snapshot of the web captured up to a training cutoff. Whatever they know about your domain is frozen at that point. Migrating your site, changing your domain, or restructuring your URLs does nothing to update that snapshot. The model keeps citing (or failing to cite) the version of your site it learned from, until a future training run catches up, which can take months.

Live retrieval. Tools like Perplexity, ChatGPT’s browsing mode, and Google’s AI Overviews fetch content in real time or from a more recent crawl index. This is the mechanism a migration actually affects immediately, for better or worse. Broken redirects, deleted pages, and new URL structures show up here first.

The practical result: a migration can look fine in live retrieval tools within days, while training-data models keep citing your old site’s structure and content for a long stretch afterward. How often LLMs recrawl content for citations covers why that lag exists and what it means for timing any content change.

What actually breaks during a migration

Missing or incomplete redirects. The same risk that breaks Google rankings breaks AI retrieval, often faster. If an AI tool tries to fetch a previously cited URL and gets a 404, that source drops out of its answer immediately rather than waiting for a ranking signal to decay.

URL structure or slug changes. Even a successful 1:1 redirect can lose context if the new URL doesn’t match what was cited in older training data or cached retrieval indexes. Some tools store the exact URL string alongside a citation, not just the underlying content.

Content consolidation. Redesigns often merge several thin pages into one comprehensive page. That’s usually a good move for users and for topical authority, but it can delete the exact page that was already earning a specific citation, with no guarantee the new, broader page earns that same citation back.

No accumulated trust on a new domain. Domain changes are the highest-risk migration type. A new domain has no citation history, no accumulated third-party mentions pointing to it, and no track record a model has learned to trust, even if the content is identical to the old domain’s.

Migration type	Primary citation risk	What to prioritize
Redesign, same domain and URLs	Low	Verify no accidental URL or heading changes on high-citation pages
Redesign with new URL structure	Medium	1:1 redirects, preserve slugs where possible
Content consolidation	Medium-high	Map which old pages were cited before merging them
Full domain change	High	Redirects, plus rebuilding third-party signals pointing to the new domain

How to protect citations through a migration

1. Audit which pages are already being cited before you touch anything. Run your highest-value prompts across the major AI tools and note which specific URLs come back. This is your baseline, and it tells you what’s actually at stake.
2. Keep URL slugs wherever possible. Even when changing site structure, preserving the exact slug for a previously cited page reduces the chance that the redirect confuses a retrieval system that stored the original URL alongside its citation.
3. Set up clean 1:1 redirects, not blanket redirects to a homepage or category page. A redirect that dumps every old URL onto a generic landing page gives an AI tool nothing to extract when it tries to follow a citation, which looks the same as a dead link.
4. Treat page consolidation carefully. Before merging pages, check whether any of them were individually cited. If so, make sure the merged page actually contains that specific content, not just a summary of it.
5. Update your llms.txt and resubmit to Search Console and Bing Webmaster Tools. These signals help retrieval-based tools find the new structure faster, even though they won’t affect training-data models.
6. Keep the old domain’s redirects live indefinitely. Letting an old domain expire after a short grace period cuts off any citation still pointing to it, whether from a model’s training data or a stale cached index.

Why some citations take time to come back

Even a flawless migration won’t immediately restore citations from training-data models, because those models aren’t re-learning your site in real time. This is a timing issue, not a mistake to fix. Live retrieval tools should reflect a clean migration within days to weeks, once crawlers revisit the new structure and redirects resolve correctly. Training-data models catch up on their own schedule, tied to their next release cycle, and there’s no way to accelerate that from your side.

The practical takeaway is to migrate with citation history in mind, not just rankings, and to expect the two systems to recover on different timelines. Treat the pre-migration audit as the thing that makes the whole process defensible: if citations do drop, you’ll know exactly which pages to check first instead of guessing.