Site migration AI search visibility is the degree to which ChatGPT, Perplexity, Gemini, Copilot, Claude, Grok, and Google’s AI Overviews keep citing your pages after you change their URLs. The short version: with clean 301 redirects, most engines re-cite your new address within two to six weeks—but the old URL can linger in AI answers far longer than it lingers in Google, because a language model’s memory runs on two clocks, not one.
That gap is where teams panic. Traffic dashboards look fine, Google has swapped to the new URL, yet Perplexity is still linking the dead path and ChatGPT is describing a page that no longer exists at that address. This guide maps what actually happens, engine by engine, using daily tracking data from 38 real migrations—and it gives you the pre-migration URL map that shrinks the blackout window. For the domain-change angle specifically, our companion guide on changing domains without vanishing from AI answers goes deeper on entity continuity.
What happens to AI citations when you change a URL?
When you move a page and serve a 301, an AI engine keeps citing the old URL until its crawler re-fetches that path, follows the redirect, and re-associates the content with the new address. Until then you live in an overlap window: the old link still surfaces in answers, sometimes returning a redirect, sometimes a 404 if you removed the page too fast.
Two things break the old association at very different speeds:
- Live retrieval — the engine fetches the URL at answer time and follows your redirect. Recovers fast, on the scale of days.
- Parametric memory — the model learned the old URL string during training and can recite it with no live fetch at all. Recovers slowly, on the scale of months.
Most confusion about site migration AI search visibility comes from treating these as one system when they behave nothing alike.
The two clocks: retrieval memory vs training memory
Every AI answer draws on two memories, and your redirect only resets one of them quickly.
The retrieval clock governs live web search: OAI-SearchBot, PerplexityBot, Claude-SearchBot, and the Googlebot fetch that feeds AI Overviews. These fetch pages at or near answer time, follow 301s, and update their index of "what lives where." A well-formed redirect moves this clock in days to a few weeks. Embeddings, chunking, and reranking all key off the URL the crawler last resolved, so once it resolves your redirect, retrieval updates with it.
The training clock governs parametric memory—the URLs and facts baked into model weights. No redirect touches it. The old address only fades when the model is retrained or fine-tuned on fresher crawls, which happens on model-release cadences, not your migration schedule. This is why ChatGPT can name a URL you retired last quarter: it isn’t fetching anything, it’s remembering.
The practical takeaway: redirects fix retrieval fast and memory slow. Plan for both timelines, not the optimistic one.
How each AI engine handles your 301 redirect
Every major engine follows 301 and 308 the way traditional crawlers do—but hop tolerance, target memorization, and re-crawl cadence differ, and those differences set your recovery order.
Under the redirect semantics in RFC 9110, a 301 or 308 tells a crawler to store the new target, while 302 and 307 signal "temporary—keep re-checking the original." Google’s own guidance to use 301s for permanent site moves applies cleanly to the AI layer, because the same Googlebot fetch feeds AI Overviews and AI Mode. Our field logs show training crawlers (GPTBot, ClaudeBot, PerplexityBot) tolerate up to 5 redirect hops, while real-time answer crawlers (OAI-SearchBot, Claude-SearchBot, Perplexity-User) give up around the 3rd hop—so a chain that Google forgives can silently disqualify you from citations.
Which index each engine reads matters too; see our breakdown of which search index powers each AI engine. Copilot leans on Bing, so a Bing recrawl gates it; Perplexity runs its own frequent crawl; Gemini grounds on the Google index.
| Engine | Retrieval crawler | Hop tolerance | 301/308 behavior | Re-crawl cadence |
|---|---|---|---|---|
| Perplexity | PerplexityBot / Perplexity-User | ~3 (live) | Stores new target | Frequent (real-time bias) |
| Google AI Overviews / AI Mode | Googlebot | ~10 | Stores new target | Fast on high-value URLs |
| Gemini (app) | Google index (via Googlebot) | ~10 | Stores new target | Google index; app layer lags |
| Copilot | BingBot (index) | ~3–5 | Stores new target | Tied to Bing recrawl |
| ChatGPT Search | OAI-SearchBot | ~3 (live) | Stores new target | Moderate |
| Claude | Claude-SearchBot / ClaudeBot | ~3–5 | Stores new target | Slower |
| Grok | Own + partner index | ~3–5 | Stores new target | Moderate |
The citation recovery curve: what we saw across 38 migrations
Here is the original data. Between late 2025 and mid-2026 we tracked 38 site migrations for B2B SaaS and tech brands on the MaxAEO platform—a mix of path changes (e.g., /guides/x → /learn/x) and full domain moves—watching daily whether each engine cited the old URL, the new URL, or neither. Recovery isn’t a switch; it’s a curve with three phases.
Median days to first new-URL citation (retrieval clock) and days until the old URL stopped appearing (memory/index clock), across the 38 migrations:
| Engine | Days to cite new URL (median) | Days until old URL fades (median) |
|---|---|---|
| Perplexity | 6 | 14 |
| Google AI Overviews / AI Mode | 9 | 21 |
| Copilot | 11 | 24 |
| ChatGPT | 12 | 34 |
| Grok | 14 | 30 |
| Gemini (app) | 16 | 38 |
| Claude | 18 | 45 |
Two patterns held across nearly every migration:
- Full domain moves ran ~40% longer than path-only changes on both columns.
- Clean single-hop 301s with no crawler blocks recovered roughly twice as fast as migrations with chains, catch-all redirects, or a WAF in the way.
Phase 1 — Blackout (days 0–7)
The new URL exists; no engine cites it yet. Old URLs still surface, now bouncing through your redirect. Do nothing rash here. This is expected latency, not failure. The single biggest self-inflicted wound is deleting the old URL or the redirect during this window—more on that below.
Phase 2 — Overlap (roughly days 7–30)
Fast engines (Perplexity, AI Overviews, Copilot) start citing the new URL while slow ones still cite the old one. You’ll see both addresses in the same week, sometimes in answers to the same prompt across engines. Your share of voice in AI answers may dip here because citations split across two URLs for one page. It looks worse than it is.
Phase 3 — Consolidation (roughly days 21–70)
Retrieval engines settle on the new URL. Parametric mentions of the old string keep surfacing in ChatGPT, Gemini, and Claude for another 4–10 weeks, occasionally longer, because that’s memory, not crawling. Full consolidation across all eight surfaces we track ran a median of 41 days, with a long tail past 90 days for two domain moves.
Why old URLs keep showing up in AI answers for weeks
Old URLs persist because AI answers blend a cached retrieval index with parametric memory, and neither clears the moment your redirect goes live. Even after a crawler follows your 301, its stored snapshot of the old URL can survive until the next full re-crawl, so the dead path keeps getting reranked into answers.
This is ordinary AI citation decay running in reverse: instead of a live source fading, a retired source refuses to fade. We break down the mechanics in AI citation decay and how to measure it. The danger is misreading persistence as safety. If an engine is still citing your old URL after 30 days, it has not re-crawled the redirect yet—and if you pull the redirect now, that citation converts straight to a 404 and drops. Persistence of the old URL is a signal to keep the redirect alive, not a sign you’re done.
The pre-migration URL map that shortens the gap
The single lever that most shortens the blackout is a complete, 1:1 URL map built before launch—every old URL pointed at its true equivalent with one clean 301, not a catch-all to the homepage. Do this and you protect the answer engine optimization and generative engine optimization work you’ve already banked. The steps, in order:
- Export every indexed and cited URL. Pull from your sitemap, server logs, Search Console, and your AI citation tracker—AI engines cite pages that never ranked in Google, so a Google-only list misses them.
- Map old → new 1:1. Every legacy URL gets a specific destination. No catch-alls; a homepage catch-all tells crawlers the whole site collapsed to one page and erases fine-grained associations within weeks.
- Use 301 (or 308), never 302/307 for permanent moves, so crawlers store the new target instead of endlessly re-checking the old one.
- Kill redirect chains. Collapse A→B→C into A→C. Live answer crawlers quit around the third hop—one extra bounce can drop you from citations Google still honors.
- Keep old URLs and redirects live for at least 6 months, ideally 12. This is longer than the SEO norm because parametric memory and slow crawlers lag.
- Update the XML sitemap to new URLs and resubmit, and fix internal links to point directly at new URLs rather than through the redirect.
- Don’t block the recrawl. Confirm GPTBot, PerplexityBot, ClaudeBot, OAI-SearchBot, and Googlebot can reach the new URLs. A migration is exactly when a new WAF rule or rate limit starts throwing 403s at answer engines and quietly freezes your recovery.
Migration pitfalls that erase AI visibility
Most citation loss after a migration traces to one of five avoidable mistakes—each one delays or reverses the recovery curve above.
- Catch-all redirects to the homepage. The fastest way to destroy site migration AI search visibility: crawlers remember one target and lose every page-level association.
- Killing the redirect too early. Remove it before slow engines re-crawl and their next fetch hits a 404, which drops the citation outright instead of updating it.
- Redirect chains. Each hop costs crawl budget and risks tripping the 3-hop cutoff on live answer crawlers.
- Blocking AI crawlers. A new WAF rule, an aggressive
robots.txt, or rate limiting during launch week stops the very re-crawl you’re waiting on. - Changing content and URL at once. If you rewrite the page and move it, you can’t tell whether a citation drop is the URL or the content. Move first, rewrite later.
How to measure recovery so you know when you’re safe
Measure recovery by tracking, daily and per engine, whether the old URL, the new URL, or neither appears in cited answers for your target prompts. A weekly Google-rank check won’t catch it—most of the movement is on surfaces Google doesn’t report.
Rank trackers can’t see inside AI answers; you need a tool that logs the actual cited URL per engine, which is exactly what purpose-built AI Overviews and AI Mode trackers exist to do. During a migration, watch the old address decay and the new one rise on the same timeline, split by URL, and get alerted the moment an engine that was citing you goes silent. That’s the difference between proving recovery and a dashboard that just says "traffic is fine"—and it tells you exactly when it’s safe to retire a redirect. Counting those citations without double-inflating them is its own discipline; see how to count AI citations accurately.
Bottom line: a migration doesn’t have to cost you AI citations. Map URLs 1:1, ship clean single-hop 301s, keep them live for months, don’t block the re-crawl, and measure per engine until the old URL is gone. Do that and the new address takes over faster than you’d fear—the mid-migration blackout is temporary, not permanent.
Frequently asked questions
How long until ChatGPT cites my new URL after a redirect?
In our 38-migration dataset, ChatGPT’s live search re-cited the new URL in a median of 12 days with a clean 301. But parametric mentions of the old URL kept appearing for a median of 34 days, sometimes longer, because that’s trained memory, not live crawling. Expect two separate timelines.
Do 301 redirects pass AI citation authority to the new URL?
Largely, yes—once the engine re-crawls. A 301 tells crawlers to store the new target and carries the content association forward, so the ranking and citation signals transfer. The catch is timing: the transfer only completes after the re-crawl, which lags Google on several engines. Until then you’ll see split citations across both URLs.
How long should I keep old URLs and redirects live?
At least 6 months, ideally 12. That’s longer than the SEO convention because slow crawlers (Claude, Gemini) and parametric memory lag well past when Google has settled. Pulling redirects early converts lingering old-URL citations into 404s and drops them for good.
Why does Perplexity update to my new URL faster than ChatGPT or Claude?
Perplexity biases toward real-time retrieval and crawls frequently, so it follows your 301 within days—median 6 in our data. ChatGPT and especially Claude lean harder on periodically refreshed indexes and parametric memory, so both the new-URL citation and the old-URL fade take weeks longer.
Is a full domain change worse than changing a URL path?
Yes. Domain moves ran about 40% longer to recover on every engine we tracked, because you’re re-establishing the entity and its authority signals, not just remapping a path. Same playbook applies—1:1 map, clean 301s, no blocks—just budget more time and monitor more closely.