A mid-market SaaS company redesigns its site, consolidates three subdomains into one, and rewrites half its help center for clarity. Traffic recovers within weeks. But three months later, the brand has quietly dropped out of ChatGPT answers it used to appear in, and nobody on the team connects the two events. This is what happens to AEO during a migration when nobody is watching the AI layer specifically.
Key Takeaways
- AI models cite specific pages, so a migration that changes URLs, merges content, or drops pages breaks the exact citations a brand had already earned.
- Structured data and canonical signals need to migrate with the same rigor as redirects, because retrieval-based models re-crawl and re-evaluate a domain after a URL change.
- Citations compound, which means a migration does not just cost visibility for a few weeks. It can reset the compounding effect and put a brand back near zero.
Why a Migration Hits AEO Differently Than SEO
Traditional SEO migration playbooks are built around one job: preserve rankings by mapping old URLs to new ones with 301 redirects, keeping title tags and header structure intact, and watching Search Console for crawl errors. That playbook still matters. But it was built for an index that ranks documents, not for a model that has already cited specific pages as sources for specific claims.
When Perplexity cites a page inline, or ChatGPT pulls a paragraph into an answer, it has formed a small piece of trust around that exact URL, that exact phrasing, and the surrounding entity signals on that page. A migration does not just move a document. It can quietly invalidate the reasons a model trusted that document in the first place, even when the redirect works perfectly from a technical standpoint.
This is the distinction that gets missed: passing a redirect check is not the same as passing a re-evaluation by a model that cross-references Search, YouTube, and Scholar the way Gemini does, or one that leans on high-authority publications and documentation the way Claude does. Each engine re-forms its judgment differently, and a migration disturbs that judgment in ways a 200 status code cannot fix.
What Actually Breaks When URLs Change
Four things tend to break in a migration, and they rarely show up in the same week.
- Citation continuity. If a model has cited a URL as a source and that URL now redirects or disappears, the next time that model refreshes its retrieval index, the citation can simply vanish rather than transfer to the new URL.
- Structured data parity. Schema markup, author bylines, and publish dates often get lost or simplified during a CMS switch. These are part of how models assess entity identity and content freshness.
- Content architecture. Migrations frequently consolidate three thin pages into one comprehensive page for SEO reasons. That is often correct for search rankings. But if a model was citing one of those thin pages for a narrow, specific answer, the consolidated page may no longer match the query as cleanly.
- Internal link graph. The pattern of internal links that told crawlers which pages mattered most gets rebuilt from scratch, and models that weigh entity authority partly through site structure lose that signal until it is rebuilt.
None of this shows up as a traffic drop on day one. It shows up as a slow erosion in share of citation over the following month, which is exactly why teams miss the connection.
growth in AI brand presence for NewDay USA, tracked weekly across five engines over six months, the kind of gain a migration can put at risk without deliberate handling.See the case study
Share of Citation Does Not Recover the Way Rankings Do
Search rankings recover on a predictable curve. Once redirects propagate and Google re-crawls, rankings tend to return to something close to their prior position within weeks, assuming the migration was executed cleanly. AI citation does not follow that curve, because citation is not purely a function of the current page. It is a function of accumulated agreement across sources that reference the brand consistently over time.
A migration does not just move a document. It can quietly invalidate the reasons a model trusted that document in the first place.
Citations compound: once a model cites a brand, the next citation becomes more likely, because the model has already treated that source as reliable. A migration interrupts this compounding. The brand is not starting from zero, but it is starting from a weaker position than the raw content quality would suggest, because some of the trust was tied to specific, now-changed URLs.
This is also why share of citation and share of voice can diverge sharply right after a migration. A brand can still be mentioned everywhere across the web, meaning share of voice holds steady, while its share of citation, how often models actually pull it in and name it as a source, drops because the specific pages models relied on no longer exist in their original form.
The Redirect Map Is Not Enough
A redirect map answers “where did this URL go.” It does not answer “does this new URL still deserve the trust the old one earned.” Teams that treat AEO as an afterthought in migration planning tend to make three mistakes.
First, they consolidate pages purely on search-intent overlap without checking which individual URLs have been cited by AI engines. A page with modest search traffic can carry outsized weight in AI answers if it happens to be the clearest, most current source on a narrow topic. Merging it into a broader page can dilute the specificity that made it citable.
Second, they let structured data lapse during the CMS transition, treating schema as a launch-week cleanup task rather than a migration requirement. Structured data is part of how a brand shapes what AI trusts, not a finishing touch applied after the fact.
Third, they measure success by search traffic and rankings alone, missing that a clean SEO migration can still be a poor AEO migration. The two need separate checklists, even when they share a single technical execution.
What Survives a Migration and What Does Not
Some things do transfer reliably. The underlying entity, the brand itself as a distinct thing in the world, does not reset just because URLs change. If a brand has strong training-data presence built through years of coverage, mentions, and documentation across the web, that layer is far more durable than any single page. Training-data presence is earned slowly and it does not evaporate over a weekend deployment.
Retrieval presence is the layer that takes the hit. It is earned technically, through structured, current content that models can fetch and parse right now. A migration directly touches that layer: new URLs, new markup, new internal structure. This is why a brand with deep training-data presence can still see a retrieval dip after a migration, and why the dip is temporary if the technical foundation is rebuilt correctly, rather than permanent.
What To Do Before and After a Migration
Before migrating, pull a list of every URL that AI engines have cited in the past two to three months, not just the ones ranking well in search. Cross-reference this against the new site’s information architecture and flag any that will be merged, renamed, or removed. Treat these as high-risk pages requiring individual redirect and content review, not batch handling.
During the build, carry structured data forward with the same priority as redirects. Author, organization, and article schema should exist on the new page before launch, not as a post-launch patch. Keep canonical tags consistent with the final URL from day one to avoid a period where models see conflicting signals.
After launch, run a fresh AI visibility check against the pre-migration benchmark within the first month, and again at the three-month mark. A single check right after launch will look fine because indexes take time to refresh. The real signal shows up in the second and third check, when models have had time to re-crawl and re-evaluate. If share of citation has dropped on specific queries the brand used to own, trace it back to the specific page that changed and rebuild the structured data and internal links pointing to it, rather than waiting for it to recover on its own.
FAQs
Does a Site Migration Always Hurt AEO?
Not automatically. A migration that preserves URLs where possible, carries structured data forward, and maps content consolidation against known cited pages can complete with little to no AEO impact. The damage comes from treating it as a pure SEO exercise and assuming AI citation will follow the same recovery curve as rankings.
How Long Does It Take To See AI Visibility Impact After a Migration?
Search rankings often show impact within days. AI citation impact tends to show up over three to eight weeks, as different engines re-crawl and refresh their retrieval indexes on their own schedules. This is why a single post-launch check is not enough.
Should We Delay a Migration To Protect AEO?
Usually not. The fix is not delaying the migration, it is adding an AEO-specific checklist alongside the standard SEO migration checklist, covering cited-page inventory, structured data parity, and a staged post-launch measurement plan.
Can Consolidating Thin Pages Hurt AI Citations?
Yes, if one of those thin pages was the specific source a model cited for a narrow query. Consolidation is often right for search intent, but it can remove the exact specificity that made a page citable in an AI answer. Check citation history before merging.
