Skip to content

HTTP/2 and HTTP/3 for crawlers: does Googlebot support them?

Google's docs list HTTP/1.1 and HTTP/2 for its crawlers, not HTTP/3. See what that means for your CDN, Alt-Svc and AI crawlers, with curl tests for each.

By Crawlito · 8 min read

In short

  • Google's crawler documentation says its crawlers and fetchers support HTTP/1.1 and HTTP/2, with HTTP/1.1 as the default, and it does not list HTTP/3.
  • Google says crawling over HTTP/2 gives no ranking advantage, so enabling a newer protocol is not an SEO lever by itself.
  • OpenAI, Anthropic and Perplexity do not document which HTTP versions their crawlers use, so keep HTTP/1.1 working as a baseline.
  • HTTP/3 is offered to clients through Alt-Svc, and clients may ignore it, so a working TCP fallback is what keeps a crawler connected.
  • To opt out of Google's HTTP/2 crawling, the server must answer Google's HTTP/2 requests with a 421 status code.
On this page

Googlebot does not appear to crawl over HTTP/3 at the time of writing. Google's crawler documentation says its crawlers and fetchers support HTTP/1.1 and HTTP/2, and it does not list HTTP/3. The AI operators OpenAI, Anthropic and Perplexity do not document HTTP versions at all. So if you have enabled HTTP/3 on a CDN, the safe setup is one where HTTP/2 and HTTP/1.1 keep working next to it, because that is the only part of the picture anyone has documented.

This guide covers what each source actually says, how HTTP/3 gets offered to clients, how Google's HTTP/2 opt-out works, and how to test every version yourself with curl.

Which HTTP versions do Google's crawlers use?

Google's overview of its crawlers and fetchers states that its crawlers and fetchers support HTTP/1.1 and HTTP/2. Three details from that page matter for your setup:

What the page saysWhat it means for you
Both HTTP/1.1 and HTTP/2 are supportedEither protocol is a valid way to serve Google's crawlers
The default protocol version is HTTP/1.1HTTP/1.1 is the baseline, so it must work
Crawlers use the version that gives the best crawling performance, and may switch between sessions based on past crawl statisticsYou cannot assume the same version on every request

Google announced HTTP/2 crawling in a 2020 Search Central post. It said Googlebot would begin crawling select sites over HTTP/2 from mid-November 2020, choosing sites that support it and where both sides benefit. The same post says only HTTPS is covered, since plaintext HTTP/2 (h2c) is not supported. It describes the main benefit as resource savings, because one connection can carry several transfers in parallel. It also says there is no explicit drawback for sites that stay on HTTP/1.1.

Does Googlebot support HTTP/3?

Not according to Google's documentation. The crawler overview lists HTTP/1.1 and HTTP/2 and says nothing about HTTP/3. The 2020 announcement does not mention it either.

There is one more data point, and it comes from a secondary source. In May 2025, Search Engine Roundtable reported that Google's Gary Illyes said on the Search Off the Record podcast that Google did not support HTTP/3 at the moment, but that it might well add it eventually, noting that HTTP/2 and HTTP/3 let a crawler reuse one connection for several requests. He was tentative, and this is an interview remark relayed by a news site, not documentation, and it may be out of date by the time you read this. Check the crawler overview page again before you make a decision that depends on it.

What about ranking? The only ranking statement in Google's documentation concerns HTTP/2 crawling: Google says there is no Google-product specific benefit to the site, with no ranking advantage in Google Search. The Google pages cited in this post do not describe HTTP/3 as a ranking input. You may have good reasons to enable HTTP/3 for visitors, but treat crawling and ranking as separate questions that Google's documentation does not link to it.

What do AI crawler operators document about protocols?

Nothing, at the time of writing. The documentation pages below describe user agents, robots.txt behavior and IP ranges, and none of them mentions HTTP/1.1, HTTP/2 or HTTP/3.

OperatorDocumentation checkedHTTP version stated?
OpenAICrawler overview (OAI-SearchBot, GPTBot, ChatGPT-User)No
AnthropicCrawler help article (ClaudeBot, Claude-User, Claude-SearchBot)No
PerplexityBots guide (PerplexityBot, Perplexity-User)No

This is a gap in the documentation, not evidence that any of these crawlers lacks HTTP/2 or HTTP/3 support, or that it has it. Do not write a rule that depends on a guess. Plan for the oldest common version, HTTP/1.1, and treat anything newer as a bonus.

Check which version your real crawlers used

If your server writes the request line to its access log, you can see the version each crawler actually used. In the common combined log format, the protocol is the third part of the quoted request line. This command counts Googlebot requests by protocol:

bash
awk '/Googlebot/ {gsub(/"/, "", $8); print $8}' access.log | sort | uniq -c

It assumes the combined format, and a user agent can be faked, so confirm the traffic is real with verify real AI crawlers before you draw conclusions. Many CDNs log the protocol in a separate field, so check your provider's log format if the request line is missing.

How does HTTP/3 get offered, and what keeps a crawler connected?

HTTP/3 runs over QUIC, which uses UDP, while HTTP/1.1 and HTTP/2 run over TCP. A client that first connects over TCP learns that HTTP/3 is available from an Alt-Svc response header, and RFC 9114 defines this discovery in section 3.1.1. A server advertises it like this:

http
Alt-Svc: h3=":443"; ma=86400

Two rules in the specifications explain why your fallback matters:

  • Clients may ignore the offer. Under RFC 7838, alternatives are optional. Clients may ignore them, and the original origin stays reachable, because an alternative is extra routing information and does not replace it.
  • Clients should fall back. RFC 9114 says that when connectivity problems such as blocked UDP prevent a QUIC connection, clients should attempt TCP-based versions of HTTP instead.

The ma parameter sets how long a client may remember the alternative, in seconds. Without it, RFC 7838 sets the default at 24 hours. Alt-Svc: clear tells a client to forget the alternatives it has stored for that origin.

The practical conclusion is short. Enabling HTTP/3 at a CDN should add a route, not remove one. After you change the setting, confirm that:

  1. HTTPS over TCP port 443 still answers with HTTP/2 and with HTTP/1.1.
  2. Both versions return the same status code, headers and body for the same URL.
  3. Your firewall does not drop TCP while allowing UDP, or the reverse.
  4. Bot rules, challenges and rate limits behave the same on every version. A rule that works on one version and not another produces results that are hard to reproduce, which is a good reason to run the curl checks below after every edge change.

On Cloudflare, HTTP/3 is a toggle under Speed > Settings > Protocol Optimization. Cloudflare says it applies only to connections between visitors and Cloudflare, and that HTTP/3 to your origin is not supported at the time of writing. Your origin therefore does not need HTTP/3 support for the edge setting to work. For the failure modes that happen before any HTTP exchange begins, see DNS and TLS errors.

How does Google's HTTP/2 opt-out with a 421 status work?

If you want Google to stop crawling your site over HTTP/2, Google's documentation says to have your server respond with a 421 status code when Google attempts to access the site over HTTP/2. The 2020 announcement gives the same instruction. If a 421 response is not feasible, the crawler overview says you can message Google's Crawling team and calls that solution temporary.

Three points keep this from backfiring:

  • It is an opt-out, not a requirement. Google's documentation gives no reason to do this unless HTTP/2 crawling causes a problem on your server. HTTP/1.1 sites face no stated penalty.
  • The 421 must reach only Google's HTTP/2 requests. Visitors and other crawlers on HTTP/2 must still get normal pages. A broad rule that returns 421 to every HTTP/2 request would break your site for browsers.
  • Unexpected 421s are worth fixing. In HTTP, 421 means Misdirected Request, and RFC 9114 section 3.3 describes servers using it when they are not authoritative for a request. Google's documentation does not say how it treats a 421 caused by a configuration mistake, so if you see 421s in your logs and did not set up the opt-out, investigate them.

The following nginx example is our own illustration, not something from Google. It returns 421 only when the protocol is HTTP/2 and the user agent contains Googlebot:

nginx
# in the http block
map "$server_protocol:$http_user_agent" $googlebot_h2 {
    default 0;
    "~^HTTP/2\.0:.*Googlebot" 1;
}

# in the server block
if ($googlebot_h2) {
    return 421;
}

This matches only user agents that contain Googlebot. Google lists other crawlers and fetchers with different tokens, so check the overview page if you need the opt-out to cover them. Test it before you ship it, and keep a note of why the rule exists so a later cleanup does not silently remove it.

How to test each HTTP version with curl

Test with GET requests that save the body, not only HEAD, because the two can differ; HEAD vs GET requests explains why. These commands print the HTTP version curl negotiated and the status code for each version:

bash
curl -s -o /dev/null -w '%{http_version} %{http_code}\n' --http1.1 https://example.com/
curl -s -o /dev/null -w '%{http_version} %{http_code}\n' --http2 https://example.com/
curl -s -o /dev/null -w '%{http_version} %{http_code}\n' --http3-only https://example.com/

Each line should show the version you asked for and the same status code. The third command needs a curl build with HTTP/3 support; run curl -V and look for HTTP3 in the feature list. If your build lacks it, the command fails and tells you so, which does not mean your server has a problem.

To see whether your server advertises HTTP/3, read the response headers from a normal request:

bash
curl -sI https://example.com/ | grep -i '^alt-svc'

No output means the server is not advertising an alternative on that response. An h3 entry means clients that honor it may try HTTP/3 next.

If you applied the opt-out, confirm it only affects the user agent you meant to target:

bash
curl -s -o /dev/null -w '%{http_code}\n' --http2 -A 'Googlebot/2.1' https://example.com/
curl -s -o /dev/null -w '%{http_code}\n' --http2 https://example.com/

The first should print 421 and the second a normal status such as 200. A spoofed user agent from your own machine says nothing about how the real Googlebot is treated, so confirm in your server logs after the change. For more curl techniques, including testing with AI crawler user agents, see how to see your page as an AI crawler with curl. If a check returns a status you do not expect, status codes and redirects explains what each one tells a crawler.

Frequently asked questions

Does Googlebot support HTTP/3?

Google's crawler documentation lists HTTP/1.1 and HTTP/2 and does not list HTTP/3. In May 2025 a Google representative said on a podcast that Google did not support HTTP/3 at the moment but might well add it eventually. Check Google's overview page for the current state.

Will enabling HTTP/3 hurt my Google crawling?

The Google pages cited in this post do not describe any such effect. Clients can ignore an Alt-Svc offer and fall back to TCP, so a correctly configured site keeps serving HTTP/2 and HTTP/1.1 as before. The risk comes from misconfiguration, such as TCP and UDP rules that differ, which is why the curl checks above are worth running.

Does HTTP/2 or HTTP/3 improve my rankings?

Google says HTTP/2 crawling carries no ranking advantage, and the Google pages cited here do not describe HTTP/3 as a ranking input. Choose a protocol for the benefit it gives your visitors and your servers, not for search results.

Which HTTP version do ChatGPT, Claude and Perplexity crawlers use?

Their operators do not say. The OpenAI, Anthropic and Perplexity documentation pages describe user agents, robots.txt and IP ranges only. If you need the answer for your own site, check your access logs for verified requests from each crawler's published IP ranges.

Check what crawlers actually receive

Run the free check on one page to see how it reads to AI crawlers, with no account needed. For protocol-level questions, keep the curl commands above. If a problem may span the whole site, an audit runs the checks across up to 500 pages and returns evidence with prioritized fixes.

Filed under Access and robots.txt. Topics: google, ai-crawlers, http-status, troubleshooting