Can ChatGPT's crawlers reach a site that only supports IPv6?

An IPv6-only site may be invisible to ChatGPT's crawlers, and the reason is plain networking, not ranking.

OpenAI publishes the IP ranges its crawlers use as JSON files. When the evidence base behind this page checked them on 22 August 2026, the counts were:

• searchbot.json (OAI-SearchBot): 35 IPv4 prefixes, 0 IPv6 prefixes, created 2 January 2026
• gptbot.json (GPTBot): 21 IPv4, 0 IPv6
• chatgpt-user.json (ChatGPT-User): 204 IPv4, 0 IPv6, dated 14 August 2026
• adsbot.json: 2 IPv4, 0 IPv6

Those lists describe where the traffic comes from. If every published range is IPv4, the sensible reading is that the crawlers connect over IPv4. A server that only answers on an IPv6 address gives an IPv4 client nothing to connect to. The request fails before it ever reaches your robots.txt or your HTML.

This failure is rare, because most hosts and CDNs are dual-stack by default. When it happens, it's total for that crawler. OpenAI describes search eligibility as two conditions: allow OAI-SearchBot in robots.txt, and confirm your host or CDN accepts traffic from OpenAI's published searchbot IP addresses. An IPv6-only origin fails the second condition by construction. The OpenAI crawlers, robots.txt and firewalls page covers the rest of that gate.

The risk sits with the hostname the crawler actually resolves, not with your internal network. Your servers can run IPv6 internally and still be fine, as long as the public edge answers on IPv4.

You're exposed if any of these is true:

• Your domain has AAAA records and no A record.
• Your apex domain is dual-stack but www, a docs subdomain or a pricing subdomain is IPv6-only.
• A CDN fronts your main site, but a separate origin hostname that serves some pages is IPv6-only.
• The host serving /robots.txt differs from the host serving content, and only one has an A record.

The last case matters more than it looks. A crawler that can't read robots.txt has to make its own decision about the site. Anthropic warns that blocking its crawlers by IP impedes its ability to read your robots.txt file, and an unreachable address has the same effect.

You're probably not exposed if a mainstream CDN fronts every public hostname. Most publish both record types for proxied hostnames. Check rather than assume, because a single DNS edit can switch that off.

Run two lookups for every hostname that serves pages you want in AI answers:

dig +short A example.com
dig +short AAAA example.com

Read the result like this:

• A returns an address, AAAA returns an address or nothing: an IPv4 client can connect. You're fine on this point.
• A returns nothing, AAAA returns an address: the hostname is IPv6-only. An IPv4-only crawler can't reach it.
• Both return nothing: the hostname doesn't resolve at all, which is a bigger problem.

Repeat for www, any subdomain that serves content, and the host behind /robots.txt. If a CNAME points to a CDN, the lookup follows it, so the answer reflects what a crawler sees.

A record alone doesn't prove reachability. A firewall can still return a 403 to the crawler from an IPv4 address. Next, request a page with each crawler's user agent and confirm a 200. The free AI crawler checker does that, and your server logs show whether real crawler requests arrive at all.

Give crawlers an IPv4 path to your public hostnames. Pick the option that fits your stack:

1. Add an A record. If your host can assign an IPv4 address, publish it alongside the AAAA record. That's the simplest fix.
2. Put a dual-stack CDN or reverse proxy in front. The edge accepts IPv4 connections and talks to your IPv6 origin behind it. Your origin stays IPv6-only.
3. Fix stray subdomains. If only one subdomain is IPv6-only, move it behind the same edge as the rest of the site.

Then re-run the DNS check, request a page as OAI-SearchBot, and watch your logs for the following week.

Two cautions. First, don't respond by building an allowlist around OpenAI's IP files alone. OpenAI publishes the ranges and warns against IP-only allowlisting, so validate on user agent plus a verified-bot programme instead. The IP allowlisting caveats and whether to allowlist OpenAI's IPs go deeper. Second, this fix removes a block. It doesn't make ChatGPT retrieve or cite you. Retrieval is per query and at the model's discretion.

Several things sit outside what the evidence covers. Be plain about them:

• The files can change. The counts above come from a check on 22 August 2026. OpenAI could add IPv6 ranges at any time. Open OpenAI's searchbot IP file yourself before you decide anything, and look for ipv6Prefix entries.
• Other crawlers aren't covered here. This evidence records OpenAI's files only. For PerplexityBot, Claude-SearchBot or Meta's crawlers, check each vendor's documentation.
• Google is a separate question. Googlebot governs AI Overviews and AI Mode. Nothing here says how it handles IPv6-only hosts.
• Robots.txt and the network are separate gates. If your DNS is dual-stack and you still never appear, see why robots.txt can allow AI bots and ChatGPT still skips you.

What stays in your control is whether a crawler can connect. What the model does after that is not.