Short answer, as of August 2026: capture a citation baseline before you launch, keep the quoted passages intact in server-rendered HTML, redirect one-to-one, and re-verify crawler access on the new stack the day you go live. Standard migration checklists protect rankings well and citations badly, because they check that URLs resolve — not that the specific content an engine was quoting still exists in a readable form.
Migrations are the single largest avoidable cause of citation loss, and the losses tend to be discovered a quarter late, long after anyone would connect them to the launch.
Why citations are more fragile than rankings here
A ranking attaches to a URL, and a correct redirect carries most of its value across. That is why the classical migration checklist works.
A citation attaches to a passage. The engine selected your page because a specific part of it answered a specific question — a comparison table, a paragraph containing a number, a clearly worded definition. If the redesign moves that content behind a tab, replaces the table with an interactive component, trims it for visual balance, or renders it client-side, the URL still resolves, the ranking may hold, and the reason you were being cited is gone.
Add to that three technical shifts that arrive with almost every replatform: a new hosting stack with a different WAF, a new rendering model, and a new template that emits its own canonical and meta tags. Each is a plausible route to losing assistant visibility while the site looks perfect to humans.
Before you launch: the baseline nobody takes
The pre-launch step that matters most takes an afternoon and is almost never done.
Record which pages currently hold citations. Run your prompt set, note which of your URLs are named, and keep the verbatim answers with dates. Without this you cannot tell after launch whether something broke or was never there. If you have no prompt set yet, build one first — choosing which prompts to track covers it, and thirty to sixty real buyer questions is enough.
Extract the quoted passages. For each cited page, identify the actual sentences or tables engines are drawing on. This becomes a protected-content list that the design team must preserve verbatim, not "in spirit".
Snapshot the rendered HTML. Save the server-rendered output of the top cited pages, not the visual design. That is the artefact you will diff against after launch.
Record the current entity signals. Business name, address, phone, categories and the structured data on key templates. Migrations frequently drop schema because it lived in the old theme.
The four failure modes
| Failure | What happens | Detectable by a normal checklist? | Prevention |
|---|---|---|---|
| Passage loss | Quoted table or paragraph removed, trimmed, or moved behind interaction | No — URL and rank both fine | Protected-content list, HTML diff |
| Render regression | New stack renders body content client-side | Rarely — humans see it fine | Fetch with JavaScript disabled pre-launch |
| Access regression | New host or CDN blocks assistant agents by default | No — search crawlers unaffected | Test agents against staging and production |
| Signal loss | Schema, canonicals or metadata dropped in the new templates | Partially | Compare structured data before and after |
Access regression deserves particular attention on a replatform, because bot-management defaults differ by vendor and the new provider may ship "block AI scrapers" enabled. The test procedure is in the AI crawler access audit, and the rendering counterpart is JavaScript rendering and AI search visibility.
Launch-day sequence
Order matters more than thoroughness here; a fast, correct sequence beats an exhaustive one run the following week.
- Confirm the redirect map is one-to-one. Every old URL to its closest equivalent, not to the homepage. Bulk-redirecting a section to a hub page is the classic way to lose an entire content cluster.
- Test crawler access on production immediately. Not staging — production, with the assistant user agents, reading status codes. Do this within the first hour.
- Fetch cited pages with JavaScript disabled and confirm the protected passages are present in the raw HTML.
- Verify canonicals on a sample of templates. A shared layout emitting a single canonical is a silent catastrophe and one of the most common launch defects.
- Resubmit the sitemap and push the changed URLs through IndexNow. This measurably shortens rediscovery, particularly on the Bing side that several assistants depend on — see IndexNow and Bing indexing speed.
- Re-check structured data on the templates that carry Organization, Product, Service and FAQ markup.
- Confirm your entity data still matches what the directories and your Google Business Profile say, especially if the domain changed. Entity consistency and the knowledge graph covers the propagation problem.
The two-week watch
Rankings usually settle within a fortnight. Citations move more slowly and less legibly, so the watch period needs to be explicit or it will not happen.
Run the prompt set at launch, at one week, and at four weeks, and compare against the pre-launch baseline. What you are looking for is not a single missing citation — variance is normal — but a consistent pattern: pages that were reliably named and now never are.
If citations drop while rankings hold, work through the failure table in order. In practice it is nearly always passage loss or access regression, and both are cheap to fix once identified. If the passage was trimmed, restore it; the content that earned the citation was doing a job the redesign did not know about.
Be careful about the opposite error, too. Some post-migration citation loss is ordinary citation decay that happened to coincide with the launch. The baseline is what separates the two, which is the whole argument for taking it.
Domain changes are a different magnitude
Changing domain as well as design compounds every risk, because the entity signal moves too. Assistants that learned your business as one domain need to re-associate it with another, and third-party sources will carry the old one for months.
Two things help disproportionately. First, keep the redirects in place indefinitely rather than for the customary year — assistant retrieval will follow stale links for longer than search crawlers do. Second, update the third-party record set deliberately and early: directories, profiles, partner pages, press. That is slow work with a long tail, described in third-party sources and AI citations.
Expect a genuine dip. A domain change that costs a few weeks of citation stability is a normal outcome, not a failure, provided the baseline shows recovery beginning by around week six.
Make the baseline automatic
The reason this discipline gets skipped is that the pre-launch baseline is manual, dull, and competes with launch pressure for attention.
DigiRank's AI Visibility Tracker runs the prompt set on a schedule and stores verbatim answers and cited sources with dates, so the pre-migration baseline exists whether or not anyone remembered to take one — and the post-launch comparison is a chart rather than a research project. The site audit module re-checks access, indexing and structured data continuously, which is exactly the coverage a migration needs in its first fortnight. Tracking starts on the $99/mo Starter plan; the site audit and auto-fix pipeline, including the Git and WordPress integrations that make the fixes shippable, come with Agency at $249/mo on a 14-day trial.
Frequently asked questions
Can a site migration cost me AI citations even if my rankings survive? Yes, and this is the common case. A citation attaches to a specific passage rather than to the URL, so if a redesign trims, moves or client-side renders the table or paragraph an engine was quoting, the citation can disappear while the redirect works and the ranking holds.
What should I capture before a redesign? Which pages currently hold citations, the verbatim answers and dates, the exact passages being quoted, the server-rendered HTML of those pages, and your current structured data and entity signals. That baseline is what lets you distinguish a launch defect from normal variance afterwards.
What is the first thing to check after launch? Crawler access on production, with assistant user agents, reading status codes. New hosting and CDN stacks sometimes ship bot-blocking defaults that affect assistants without touching search crawlers, so nothing else in your checklist will surface it.
How long before AI citations recover after a migration? Rankings typically settle within two weeks; citations lag and are less legible. Check at launch, one week and four weeks against your baseline. If a domain changed as well, expect a real dip and look for recovery to begin around week six.
Should I redirect old URLs to the homepage if there is no equivalent? No. Redirect to the closest equivalent page, and if none exists, consider keeping the content live at a new URL. Bulk redirects to a hub or homepage are the classic way to lose an entire cluster of rankings and citations at once.
Do I need to resubmit anything after a migration? Yes — resubmit the sitemap and push changed URLs through IndexNow. That shortens rediscovery, and it matters more than usual for AI visibility because several assistants depend on the Bing index rather than Google's.
How long should migration redirects stay in place? Indefinitely, where practical. Assistant retrieval follows older links for longer than search crawlers do, and third-party sources carrying your old URLs update on their own schedules, sometimes over many months.
What if citations drop but I cannot find a technical cause? Compare the current rendered HTML of the affected pages against the pre-launch snapshot. Passage loss is the most common invisible cause: the content reads the same to a person while the specific quoted sentences or table are gone or no longer server-rendered.