301 vs 302 vs Canonical for AI Search: The Redirect Playbook for 2026

TLDR: AI search engines follow redirects but with two big differences from classic Googlebot: each hop adds latency that reduces chunk re-indexing probability, and most AI crawlers cache the original URL longer than Google does. The result is that messy redirect strategies tank AI citations for 2 to 6 months even when classic SEO recovers in weeks. This is the redirect playbook I follow on every site migration: when to use 301, when to use 302, when to use canonical, and the four mistakes that cost the most citations.

How AI crawlers actually handle redirects

OpenAI’s GPTBot, Anthropic’s ClaudeBot, Perplexity’s PerplexityBot, and Google’s Google-Extended AI crawlers like GPTBot follow standard HTTP redirects. A 301 transfers authority to the target URL. A 302 keeps the original URL as the canonical reference. A 307 is a temporary redirect that preserves the HTTP method. A canonical tag is a hint, not a redirect, and AI crawlers respect it but apply it inconsistently.

The difference from classic Google is in cache freshness. Google revisits high-traffic URLs every few days. AI crawlers operate on monthly to quarterly recrawl cycles for most domains. A 301 you ship today may not show up in ChatGPT or Perplexity citations for 60 to 120 days, even though Googlebot picks it up within a week.

301 vs 302: pick the right one

The rule is simple but often misapplied:

• 301 (permanent): The original URL is gone forever. Use for site migrations, URL slug changes, consolidating duplicate pages, removing seasonal content into evergreen pillars.
• 302 (temporary): The original URL will return. Use for A/B tests, temporary maintenance pages, regional redirects that may flip back.
• 307 (temporary, method-preserving): Same use case as 302 but for non-GET methods (rare in content contexts).

The most common mistake is using 302 by default because the dev tooling defaults to it (NGINX’s ‘rewrite’ directive, Express’s ‘res.redirect’ without a status code, etc.). 302s on permanent moves leave the AI crawler caching the original URL indefinitely. Audit your stack and ensure all permanent moves are 301.

When canonical tags beat redirects for AI search

Canonical tags are the right choice when the original URL must remain live and accessible (because users can land on it from external links, bookmarks, or apps) but you want AI crawlers to attribute authority to the canonical version. Three common cases:

1. Tracking parameters: ?utm_source, ?ref, etc. Always canonical to the clean URL.
2. Pagination: /blog/page/2/ canonical to /blog/ (controversial but works for AI retrieval).
3. Print or amp variants: Canonical the variant to the main URL.

Do not use canonical for a permanent move. Canonical is a hint, not a directive. AI crawlers may follow it for one query and ignore it for another. A 301 is unambiguous.

The four redirect mistakes that crush AI citations

1. Redirect chains: URL A 301s to B which 301s to C. Each hop drops citation continuity. Always collapse to a single hop (A directly to C).
2. Redirect loops: URL A 301s to B which 301s back to A. AI crawlers give up and de-cite both URLs. Audit weekly during any migration.
3. Mass redirect to homepage: All deleted blog posts 301 to /. AI engines treat this as a soft 404 signal. Use 410 for genuinely dead content.
4. Redirecting HTTPS to HTTP or vice versa: Forces AI crawlers to re-evaluate the entire domain’s security posture. Always serve consistent protocol.

Audit redirect chains using Screaming Frog’s ‘Redirects’ tab. Any chain longer than one hop should be collapsed before the next AI crawler cycle.

Migration playbook for keeping AI citations

Site migrations are the highest-risk moment for AI search visibility. Follow this sequence:

1. Map every existing URL to its new equivalent in a spreadsheet. No orphans.
2. Implement 301s at the server level (not in JavaScript or meta refresh).
3. Update internal links to point to the new URLs directly. No reliance on the redirects.
4. Submit both old and new sitemaps to GSC. Bing too.
5. Update llms.txt if you maintain one to reflect the new canonical URLs.
6. Monitor citations weekly for the first 90 days using a GEO tracker.
7. Do not remove redirects for at least 12 months. AI crawler recrawl is slow.

Sites that follow this playbook typically recover full AI citation volume within 60 to 90 days. Sites that skip steps 4, 5, or 6 routinely lose 30 to 60% of their AI presence and take 6 to 12 months to recover.

Status codes AI crawlers actually understand

Most AI crawlers handle the standard set well, but a few status codes have surprising behaviour:

• 200 OK: Index normally.
• 301 Moved Permanently: Follow, transfer authority.
• 302 Found: Follow, keep original as canonical.
• 404 Not Found: De-cite over time, slowly.
• 410 Gone: De-cite immediately. Use for confirmed dead content.
• 451 Unavailable for Legal Reasons: Suppress without de-citing other domain content.
• 503 Service Unavailable with Retry-After: Pause crawling, no de-citation.

Use 410 aggressively for content you have confirmed should never come back. It accelerates de-citation by roughly 4x compared to 404.