Website Redesign SEO: Pass Every Check, Still Vanish From AI Answers

Google’s documentation is calmer but points the same direction. Its site-move guidance says a medium-sized site can take a few weeks for most pages to move in the index, larger sites longer, and that ranking fluctuations are normal while everything is recrawled. It also confirms that 301 and other permanent redirects don’t cause a loss in PageRank, which is true, and also not the reassurance it sounds like, because redirects are the one part of a redesign teams reliably get right.

The damage usually comes from everything else that changes on the same day.

One important nuance: the SALT numbers describe full domain moves, the heaviest kind of migration. A template redesign on the same URLs carries less risk. But BrightEdge’s migration guidance is blunt that a redesign or code overhaul belongs on the same risk register as a domain change, because it alters content hierarchy, page templates and load behaviour: the exact things both search engines and answer engines read.

Why can you keep your rankings and still lose AI citations?

Because ranking and citation are no longer the same contest. Ahrefs examined 863,000 keywords and four million AI Overview URLs and found that only 38% of cited pages also ranked in the top 10 for the same query: down from 76% in its July 2025 study. The rest of the citations split almost evenly between pages ranking 11–100 and pages beyond position 100. A separate BrightEdge analysis published in February 2026 put top-10 overlap at roughly 17% on its own dataset. The methodologies differ and Ahrefs is open that better citation parsing explains part of the drop, but every serious measurement now lands in the same territory: most of what AI Overviews cite is not what ranks first.

Part of the mechanism is query fan-out. When an AI Overview is generated, Google can split the search into several related sub-queries and pull citations from pages that perform across them: pages you may never have tracked a ranking for. Your redesign scorecard, watching twenty head terms in a rank tracker, is watching the wrong game.

This is why rank retention can read 100% while citation retention quietly fails. BrightEdge’s migration guide lists the specific ways a move breaks AI trust: previously cited content gets merged or removed without keeping its identity, schema gets dropped in the new templates, content moves behind JavaScript, redirects loop or soft-404, and internal link structures that established topical authority get flattened. None of those show up in a rank tracker in week one.

All of them show up in who gets cited.

What do AI crawlers actually see on your new site?

The raw HTML your server sends, and nothing your JavaScript builds afterwards. Vercel and MERJ analysed crawler traffic across Vercel’s network and found that none of the major AI crawlers render JavaScript: not GPTBot, not ClaudeBot, not PerplexityBot. They fetch JavaScript files: GPTBot in 11.5% of requests, ClaudeBot in 23.8%, but never execute them. The one exception is Gemini, which inherits Googlebot’s rendering infrastructure. So if your redesign moves product specs, FAQs or key copy into client-side rendering, Google may still see it and the AI engines simply won’t. We’ve covered what GPTBot and ClaudeBot actually do with JavaScript in depth: a redesign is where that stops being theoretical.

The same study exposed how badly AI crawlers cope with the URL churn a redesign creates. ChatGPT’s crawler burned 34.8% of its fetches on 404 pages and another 14.4% following redirects; ClaudeBot hit 404s on 34.2% of fetches. Googlebot’s figures on the same network were 8.2% and 1.5%. Google has spent two decades learning to navigate moved URLs gracefully. AI crawlers demonstrably haven’t. They keep requesting outdated assets and stale paths long after a change. Every broken path and redirect chain your launch leaves behind costs proportionally more with the crawlers that feed AI answers than it does with the one crawler your checklist was written for.

And the volume at stake is no longer a rounding error. In the measured month, GPTBot generated 569 million fetches on Vercel’s network and Anthropic’s crawler 370 million: together about a fifth of Googlebot’s volume. If your redesign brief still treats crawler behaviour as “whatever Googlebot needs,” it is designing for one audience and shipping to several. That’s the reason we argue your next visitor is an AI agent, and launch day is when you find out whether your new build agrees.

What belongs in a redesign checklist that the old lists miss?

An AI-visibility layer on top of the redirect work: benchmarked before launch, verified after. The mechanical core hasn’t changed: full URL inventory, 1:1 redirect map, 301s only, staging crawl, sitemap resubmission. What’s new is that BrightEdge’s current migration guidance tells teams to benchmark AI Overview citations alongside rankings before the move, and to set an explicit retention KPI. Its example: preserve 95% of AI citations and 90% of organic traffic within 60 days of launch. Treating citations as a KPI with a number attached, rather than a vibe, is the single biggest upgrade to the standard checklist.

In practice the additions look like this. Inventory which pages carry schema and confirm every type survives the new templates: dropped structured data is one of BrightEdge’s named ways migrations break AI trust. Diff the raw served HTML of old and new templates with JavaScript off, because that raw response is the entire input for every major AI crawler. Check that the content AI engines were citing keeps its identity: merging three cited pages into one uncited mega-page is a loss the redirect map will happily paper over. And keep the redirects up for at least twelve months, which matters twice as much for crawlers that demonstrably struggle with moved URLs.

None of this is exotic. It’s the same discipline the industry learned for rankings, pointed at a second scoreboard. The searches teams run before a redesign (“website redesign seo” at 1,900 a month, “website migration checklist” at 880 (Semrush, US database, August 2026)) show demand for the playbook is real. The playbook most of those searches find just stops at 2019.

How long does recovery take after a redesign or migration?

For full domain moves, the median is about ten months, and the tail is long. In SALT.agency’s 1,052-migration dataset, 5% of sites were back to baseline within 30 days, 22.8% within 90, 59.5% within a year and 83.5% within two. The mean, dragged out by the worst cases, was 489 days. Lighter migrations, template redesigns on stable URLs, sit well inside those numbers, which is consistent with Google’s “few weeks for a medium-sized site” guidance for the reindexing mechanics alone.

Here is the uncomfortable part: that curve exists for traffic because thousands of teams measured traffic. No equivalent public curve exists for AI-citation recovery after a redesign: how long it takes ChatGPT or an AI Overview to resume citing a moved page, or whether it ever does. BrightEdge’s guidance to track AI visibility monthly after launch exists precisely because engines re-evaluate on their own schedules. Until someone publishes that curve, the only version of it you can act on is the one you build for your own site, which requires a baseline taken before launch day, not after the first uncomfortable board question.

How do you baseline AI citations before a redesign?

Record who gets cited, for what, by which engine, while the old site is still live. That inventory is the redesign deliverable nobody ships. A workable version has four steps. First, build a prompt set: the 30–50 questions your buyers actually ask, phrased the way people ask assistants, not the way keyword tools truncate them. Second, run that set across the engines that matter for your market, AI Overviews, ChatGPT, Perplexity, and log which of your URLs get cited, which competitors appear, and for which prompts you’re absent. Third, snapshot the technical substrate of every cited page: its schema, its raw served HTML, its internal links. That’s the identity the new build has to preserve. Fourth, set the retention KPI in the project plan next to the ranking one, so that “we kept our rankings” can no longer be the whole definition of success.

After launch, re-run the same prompt set on a schedule and compare against the baseline. Citation churn is normal; what you’re watching for is your pages dropping out of answers they used to anchor while rankings hold: the signature of a redesign that broke the second scoreboard. This is the discipline our whole website strategy practice is built around: treating the site as something AI systems read, quote and act on, not just something Google ranks.