Can a website migration break your visibility in AI answers?

A migration rebuilds the plumbing AI engines rely on, even when the copy stays identical. Before an engine can quote a page, its crawler has to pass four gates:

• Rendering. The content has to be in the HTML the crawler receives.
• robots.txt. The rule for that specific bot has to allow it.
• Network. Your host, CDN or firewall has to let the request through.
• Index and snippet eligibility. The page can't carry noindex, nosnippet or max-snippet:0.

A replatform can flip any of these. Each one fails quietly. Visitors see a working site, and no dashboard tells you an AI crawler got a 403.

The good news: these are access problems, and access is the part you control. Which pages an engine retrieves and cites after that is the engine's decision, so no migration plan can promise citations. For the full code-level list, see what stops AI search engines from indexing a page.

The most common robots.txt error treats all AI bots as one. OpenAI's crawler docs say each setting is independent: GPTBot governs training, and OAI-SearchBot governs ChatGPT search. This is a valid rule set that opts out of training but stays eligible for ChatGPT search:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

If your new platform generates robots.txt automatically, check its size and structure too. See whether a very large robots.txt can break AI crawler rules.

• Rendering. Vercel's crawler study found ChatGPT and Claude crawlers fetch JavaScript files (11.50% and 23.84% of requests) but don't execute them, so they can't read client-side rendered content. Independent testing in June 2026 reconfirmed this for OpenAI, Anthropic, Perplexity and Meta crawlers. Googlebot and AppleBot are the exceptions.
• Network. OpenAI's help center says eligibility for ChatGPT search means allowing OAI-SearchBot and confirming your host or CDN allows traffic from OpenAI's published IP addresses. Both halves count. How often firewalls block AI crawlers has not been measured by anyone.
• Index. Google states a page must be indexed and eligible for a snippet to appear as a supporting link in AI Overviews or AI Mode, with no additional technical requirements.
• Engine by engine. In Otterly's 2026 experiment, Google deindexed two test sites and its AI surfaces almost stopped citing them. Bing kept the sites, and Microsoft Copilot kept citing them. One healthy index tells you nothing about another.

CDN defaults also move under you. From 15 September 2026, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for newly onboarded domains, while Search stays allowed. Moving onto Cloudflare during a migration puts you in that group. More on this in how AI crawler access breaks without you changing anything.

Before launch, on staging

1. Fetch key pages with JavaScript disabled and compare them to the rendered page. Any price, product name or heading that appears only after rendering is missing for most AI crawlers.
2. Diff the new robots.txt against the old one, bot by bot: OAI-SearchBot, PerplexityBot, Claude-SearchBot, Meta-WebIndexer, Googlebot and bingbot.
3. Confirm staging's Disallow: / and any staging noindex won't ship to production.
4. Search templates for noindex, nosnippet, max-snippet and data-nosnippet.
5. Ask whoever owns the CDN and firewall which bot categories are blocked by default.
6. Map every old URL to its new URL with a one-to-one redirect.

Launch day and the weeks after

1. Request a sample of URLs as each AI crawler and record status codes and challenge pages. The AI Crawler Checker does this per crawler. Also confirm robots.txt itself loads, since Anthropic warns IP blocking stops it reading your rules.
2. Check index status in Google Search Console and Bing Webmaster Tools separately.
3. If you created a new Search Console property, check its generative AI control. Inclusion is the default, but child properties inherit the parent's setting.
4. Rerun the crawler checks after any CDN or firewall change.

The site migration checklist for AI crawler access turns this into a sign-off sheet.

Some things sit outside the migration entirely.

• Answers without live search. When an engine answers from memory, it uses training data frozen at a cutoff. OpenAI's model docs list 16 February 2026 for the GPT-5.6 family and 31 August 2025 for ChatGPT's Instant path. A rebrand or new pricing won't reach those answers until a later model ships.
• User-triggered fetchers. Perplexity says Perplexity-User generally ignores robots.txt, and OpenAI says robots.txt may not apply to ChatGPT-User. Blocking those agents isn't reliable.
• Unsupported tasks. Don't add llms.txt files or new schema to the launch plan as AI levers. Google says its AI features need no special files or markup. Carry existing markup over if you have it, since it does no harm. The guide to retiring unsupported GEO tasks covers the rest.

To see whether AI engines name you before and after the switch, run the free checker on this site for three buyer questions on both sides of launch. Treat each result as a snapshot, not a trend.