Site moves break citations you can't redirect
Crawlmind Engineering··5 min read
A site migration is any change that puts existing content on new URLs, whether that is a domain change, a platform move, or a URL restructure. The standard plan for one is a redirect map, and a redirect map is built on an assumption that has quietly stopped holding: that everything pointing at your old URLs is something you can redirect.
Your own pages are. The third-party sources that answer engines cite alongside you are not.
#Rankings recovery was already slow
Migration recovery has never been fast, and two independent datasets put numbers on it. Dan Taylor analyzed 892 domain migrations and found that on average it took 523 days for the new domain to show the same organic traffic as the old one, with 17% of the sample never returning to prior levels even after 1,000 days. The fastest recoveries in that set landed at 19, 22 and 23 days, so the spread is enormous.
SALT.agency looked at 1,052 domain migrations and reported a median recovery around 304 days, with only 27% recovering inside 90 days and 13.9% still not recovered after two years. Different samples, different methods, same shape: the median migration takes most of a year, and a meaningful minority never comes back.
Those are organic traffic numbers. The question for anyone planning a move now is whether AI citations track them, run ahead of them, or lag behind.
#Citations move with organic visibility
The best available read says they track, closely. Lily Ray examined 11 sites that lost organic visibility during a January 2026 Google update, measuring from January 20 to February 16, and found organic traffic down an average of 26.7% against an average AI citation decline of 22.5%.
The per-engine split is the useful part. ChatGPT citations fell 27.8% and Google AI Mode 23.8%, both close to the organic drop. Perplexity moved 2.9% on average and actually grew citations for 7 of the 11 sites over the same window.
That was a ranking change, not a migration, so treat it as directional rather than predictive. What it establishes is that engines leaning on Google's ranking signal inherit Google's view of you, including the temporary, degraded view it holds while a site move settles. An engine with more independent retrieval reacts differently. A migration therefore does not produce one citation curve, it produces several.
#What a redirect actually moves
Google's own guidance sets the timeline. It recommends keeping redirects "for as long as possible, generally at least 1 year", and says a small to medium site can take a few weeks for most pages to move. The Change of Address tool is narrower than teams assume: its signals run for 180 days, after which Google "does not recognize any relationship between the old and new sites" and treats the old one as unrelated if it is still crawlable.
So a redirect does two jobs. It sends a visitor or a crawler to the right page, and it passes ranking signals during a defined window. Answer engines with live retrieval follow the first job fine. An engine fetching a cited URL at answer time will land on your new page, which is exactly what you want.
What no redirect touches is the text of the reference itself. A roundup published two years ago still names your old domain in its prose. A directory record still stores the old URL as a field. A review platform still shows the old address in its profile header. Those sources keep being retrieved and quoted, and the string they carry is the one they were written with.
#The half of your citations with no redirect
This is where migration planning and citation planning diverge. Yext analyzed 6.8 million citations drawn from over 1.6 million AI responses across Gemini, OpenAI and Perplexity, and found your own website accounted for 44% of cited sources while listings accounted for 42%, with reviews and social at 8%.
Read that against a redirect map. The map covers your own pages and nothing else. The other half of your citation surface sits in records held by other companies, each with its own edit workflow, and a 301 on your server changes none of them. Google says as much in the site move documentation itself, advising you to contact linking sites and ask them to update, prioritized by the traffic each one sends.
The listings half is the tractable part, because those records are editable. You own the login. Someone simply has to go and change the URL field on every profile, which is unglamorous work that no redirect map schedules and no rank tracker flags as missing.
#A migration plan that protects citations
Before the move, inventory the sources that actually get cited for your brand, not just the sites that link to you. Link data and citation data overlap but are not the same set, and a high-authority page that no engine ever retrieves is not worth the outreach slot.
Sort that inventory by whether you can edit it. Directory and profile records go in a list with owners and due dates, because those are a pure execution problem. Editorial mentions go in a second list ranked by how often they show up in answers, because outreach capacity is finite and most of those requests will be ignored.
Keep the old domain resolving well past the point it feels necessary. Google's one year floor is the SEO minimum, and given that third-party records update on nobody's schedule, treat it as a floor rather than a target.
Flatten redirect chains to a single hop. Every extra hop is another chance for a fetcher with a tight budget to give up before it reaches your content, and a fetch that fails is a citation that silently goes to somebody else.
Finally, measure recovery per engine rather than in aggregate. A single blended citation number will average a fast recovery on one surface against a slow one on another and tell you nothing actionable. The engines that follow Google's ranking signal will move with your organic recovery. The ones with more independent retrieval will move when their own crawlers revisit and when the third-party records finally get updated, which is a different clock entirely, and the one you control least.
Related field notes
September 24, 2026 · 6 min
Google pays for grounding, not for links
Google's AI contribution pilot pays when a page shapes an answer, not when it is linked afterward. That rule says a citation count measures the wrong thing.
September 23, 2026 · 5 min
A browser agent is not a crawler
Agentic browsing runs inside the user's own session, so robots.txt, bot allowlists and crawler analytics all miss it entirely.
September 22, 2026 · 4 min
An MCP endpoint is not a discovery channel
NLWeb and MCP make your site answerable by agents that already found you. Nothing on the open web is hunting for a /mcp route yet.
Share or discuss
New posts, no spam. Roughly monthly. Unsubscribe with one click.