Here's the short version an answer engine can quote: after a replatform, AI engines re-discover you mostly through the same search indexes they already lean on, so your rediscovery speed is decided by how fast Googlebot and Bingbot re-crawl your new URLs, whether your redirects hold, and whether you told the crawlers where to look. There is no separate "AI ranking" button. There is a queue, and you can jump it.
Now the longer version, because you didn't come here for a tweet.
Imagine you've just moved apartments. Same you, same furniture, better kitchen. But your mail still goes to the old place, your friends have the old address saved, and the pizza guy is standing outside a building you left three weeks ago, calling a number that no longer rings. You didn't disappear. You just forgot to tell the people whose job it is to find you.
That's a replatform to an AI answer engine. The site is better. The engine has no idea yet.
What "the engine finds you" actually means
The thing people get wrong is picturing one big AI brain that remembers your site. That brain (the model itself) was trained on a snapshot from months or years ago and it is not going to re-learn your new URL structure. What matters is the part that runs at question time: the retrieval step, where the engine goes and fetches live pages to answer someone. And that step is powered by an index, which is powered by a crawler.
So "getting re-discovered" is really two boring, checkable events happening again: a crawler fetches your new pages, and an index files them under the questions people ask. Everything an AI engine says about you downstream of that is just quoting the filing.
Different engines wire this up differently, which decides who forgets you and for how long.
| Engine | Crawls with | Leans on |
|---|---|---|
| Google AI Overviews / Gemini (search) | Googlebot | Google's own index |
| ChatGPT search | OAI-SearchBot, ChatGPT-User | Its own index plus Bing signals |
| Perplexity | PerplexityBot, Perplexity-User | Its own crawl plus third-party indexes |
| Copilot | bingbot | Bing's index |
Read that column on the right and you'll notice the pizza guy is really only two or three people: Google, Bing, and a handful of dedicated bots. Get re-crawled by those and most of the answer engines catch up on their own schedule.
Why the forgetting drags on
Re-crawling is not instant and it is not fair. Search engines budget crawl time per site based on how important and how fast you've historically been. A big established domain gets swept often. A page nobody links to on a site that just changed every URL can sit in the old world for weeks while the crawler slowly notices the 301 and follows it.
The killers are the usual ones, and they're the same things that hurt classic search, which is the quiet good news. Broken or missing redirects mean the crawler hits a 404 at your old address and files that as "gone," not "moved." A sitemap that still lists dead URLs, or that never updates its lastmod dates, tells the crawler nothing changed. Orphaned pages with no internal links have no path for a crawler to even reach. And if someone on the migration team quietly left Disallow or the new AI-bot user agents blocked in robots.txt during staging (this happens more than anyone admits), you've locked the door and mailed no key.
How to jump the queue
Speeding this up is not exotic. It's the pre-cutover hygiene you should be doing anyway, aimed at the fact that machines, not humans, are the ones re-learning your address.
Ship clean 301s from every old URL to its true new equivalent, one hop, no chains, and confirm they return 301 and not 302 or a soft 404. Regenerate your XML sitemaps with honest lastmod timestamps and resubmit them in Google Search Console and Bing Webmaster Tools the day you cut over, because that's the single loudest "come look now" signal you get for free. Rebuild internal linking so the important pages are two or three clicks from the homepage, not stranded. And check robots.txt on the live site with your own eyes, allowing GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended and the rest if you want to appear in those engines at all.
One more lever most people skip: IndexNow. Ping it and Bing, Yandex, Seznam and Naver get told about your changed URLs within minutes instead of waiting for a crawl. That matters because Copilot and parts of ChatGPT's picture lean on Bing. Google does not use IndexNow, so this is not a magic wand, but it's a free way to speed up half the room.
What I'd skip the hand-wringing over is llms.txt. It's a proposed file for pointing models at your content, and it's fine to add, but no major answer engine has committed to using it for rediscovery. Don't build your recovery plan on a promissory note.
The unglamorous truth is that the fastest AI rediscovery is a side effect of a clean redirect map, a fresh sitemap, and a crawlable site. Get those right before launch and you don't need a special AI recovery project. Miss them, and no amount of "AEO strategy" saves you. Finding which pages have no inbound links, which redirects chain, and which files got orphaned in the move is exactly the kind of dull, essential inventory work worth doing before cutover rather than after (it's the whole reason our pre-migration crawl exists).
So what do I actually do
I treat AI rediscovery as a redirects-and-sitemap problem, not a new discipline, and I stop losing sleep over it. If I had one afternoon before a cutover, I'd spend it verifying every high-traffic old URL 301s cleanly to a live page, resubmitting sitemaps to Google and Bing, and firing IndexNow. That covers the crawlers that feed nearly every answer engine worth appearing in, and everything else is patience.
The pizza guy will find the new building. Your only job is to leave a forwarding note that a machine can read, and to not stand there wondering why nobody's knocking while your robots.txt quietly says go away.