IPv6 proxies are the cheapest way to collect search results at volume, on the engines that accept them: Google, YouTube, Bing, Yahoo and Yandex publish IPv6 addresses, DuckDuckGo, Baidu and Naver do not. A Google result page weighs about 57 KB on the wire, so a gigabyte at $0.20 is roughly 17,000 SERPs, and a 10,000-keyword, five-country daily tracker fits inside a $20 pack.
That is the whole argument in two sentences, and the rest of this article is the evidence: which engines and SEO tools resolve over IPv6 (measured with DNS, not assumed), what a result page actually weighs when fetched through our own infrastructure, the arithmetic per thousand SERPs, how to pace requests so that Google's per-prefix limits do not eat the savings, and the cases where IPv6 is the wrong tool for SEO work. Prices are the ones on the IPv6 proxies page: $20 for 100 GB, down to $0.083/GB on 3 TB, 30-day packs.
Which search engines accept IPv6 proxies
An IPv6 proxy can only reach a host that publishes an IPv6 address — a DNS AAAA record. On 14 September 2026 we looked up 102 major domains from three resolvers in Italy and Frankfurt, following each homepage redirect; the full table and the CSV are on the IPv6 proxies page. The rows that matter for SEO:
| Host | Role | AAAA on apex / www | Reachable over IPv6 |
|---|---|---|---|
| google.com | Search, Maps, Shopping, News | 5 / 16 | Yes |
| youtube.com | Video search | 8 / 16 | Yes |
| bing.com | Search; the index behind Copilot and DuckDuckGo | 4 / 5 | Yes |
| yahoo.com | Search (Bing-powered), portal | 12 / 4 | Yes |
| yandex.ru | Search, Russia | 2 / 2 | Yes |
| duckduckgo.com | Search | 0 / 0 | No — IPv4-only |
| baidu.com | Search, China | 0 / 0 | No — IPv4-only |
| naver.com | Search, Korea | 0 / 0 | No — IPv4-only |
| ahrefs.com | SEO platform | 4 / 4 | Yes |
| moz.com | SEO platform | 4 / 4 | Yes |
| similarweb.com | Traffic intelligence | 0 / 16 | Yes (www only) |
| semrush.com | SEO platform | 0 / 0 | No — IPv4-only |
Two details worth a second look. Bing serves the index that DuckDuckGo shows in most markets, yet duckduckgo.com itself has no AAAA record: a DuckDuckGo check has to leave from an IPv4 exit even though the underlying results come from a dual-stack engine. And Yahoo Japan is a different company on a different network — www.yahoo.co.jp publishes no AAAA record — so "Yahoo accepts IPv6" is true of yahoo.com only. Per-engine notes, sessions and pacing are on the Google, Bing, Yahoo and Yandex pages of the IPv6 by use case hub.
What a SERP costs on IPv6: the arithmetic
Rank tracking is a job made of millions of small, identical fetches, so the price of a result page is the price of the job. We measured it on our own infrastructure: seven Google result pages fetched through the QuanticData SERP layer on 19 and 20 September 2026, one page of ten organic results each, as delivered to a client that does not run JavaScript, weighed between 48.5 KB and 59.2 KB on the wire, median 56.8 KB. Call it 57 KB. From there:
| Pack | Price per GB | Result pages per GB (57 KB each) | Cost per 1,000 SERPs |
|---|---|---|---|
| 100 GB for $20 | $0.20 | ~17,500 | $0.011 |
| 500 GB for $70 | $0.14 | ~17,500 | $0.008 |
| 1 TB for $110 | $0.11 | ~17,500 | $0.006 |
| 3 TB for $250 | $0.083 | ~17,500 | $0.005 |
A tracker that checks 10,000 keywords in five countries once a day makes 1.5 million fetches a month: about 85 GB at 57 KB each, inside the $20 pack, with room for retries. The same 85 GB on a residential network at $1.00/GB is $85, at $0.80/GB at volume $68 — not ruinous, but four times the price for a job that does not need a household address, because a search engine's public result page is served to any client that behaves. Where residential earns its price is the IPv4-only engines and anything logged in, which is the next section but one.
Two caveats on the arithmetic, stated plainly because they are the useful part. The 57 KB is the page as served to a plain HTTP client with a browser User-Agent and no JavaScript; a rendered SERP with its scripts, images and the AI Overview module is several times heavier, which is one reason not to render what you only need to parse. And a challenge page also costs bytes: a tracker that gets challenged on every third query is paying for the challenge, which is why pacing matters more than the per-GB price.
Rank tracking on IPv6: pace by /64, not by address
The mistake people make with IPv6 is to treat an address as the unit of reputation. Google, and the other engines to a lesser degree, count by prefix: rotating through a thousand addresses inside one /64 subnet looks, to the rate limiter, like one client sending a thousand queries. The practical rules that follow:
- Rotate per query and spread across subnets. Our IPv6 exits are drawn from many /48 and /64 ranges precisely so that rotation changes the prefix, not just the last bits of the address. Leave the session out of the username and every query gets a fresh exit.
- Keep the pace human per exit. A query every few seconds per exit, a few requests per second across the whole job. Concurrency is unlimited on the plan; the engine's tolerance is not.
- Treat a challenge page as a stop signal. It arrives with HTTP 200 and a body that is not a result page. Detect it, back off on that prefix, do not retry it in a loop — every retry is a gigabyte you pay for and a signal you send; the 429 guide covers the backoff arithmetic.
- Match the market to the exit. Country in the username (
USER-country-de), the engine's own locale parameters in the URL (gl,hlon Google;ccon Bing), and no mixing of markets on one exit. - Do not render what you can parse. The result page as HTML has the organic list, the ads, the related searches and the People Also Ask questions; rendering it in a browser multiplies the bytes and adds a fingerprint. If you would rather have JSON than HTML, the SERP API returns parsed results billed per successful search, and you never touch the fetch layer at all.
A minimal fetch, with host and port from your dashboard:
# rotating: no session in the username, a fresh IPv6 exit per query
curl --proxy "http://PROXY_HOST:PORT" \
--proxy-user "USER-country-us:PASS" \
--connect-timeout 10 --max-time 30 \
-A "Mozilla/5.0 (Windows NT 10.0; Win64; x64) Chrome/128.0 Safari/537.36" \
"https://www.google.com/search?q=ipv6+proxies&gl=us&hl=en"
# the exit should answer with an IPv6 address in the country you set
curl --proxy "http://PROXY_HOST:PORT" --proxy-user "USER-country-us:PASS" https://api64.ipify.orgSEO tools and data sources over IPv6
SEO work is not only SERPs. Teams pull backlink and traffic data from the platforms, verify what a page looks like from another country, crawl their own and their competitors' sites, and check which pages an AI assistant cites. Of the platforms themselves, ahrefs.com and moz.com are dual-stack, www.similarweb.com is (the bare apex is not — fetch www), and semrush.com is IPv4-only. For crawling other people's sites the rule is the same as for engines: 45 of the 102 major domains in our census accept IPv6 on the host their homepage lands on, 57 do not, and a crawler that routes by AAAA — IPv6 exit when there is a record, residential or datacenter when there is not — pays the low rate on roughly half its traffic without a single failed fetch. The sector split of that census (news and media 9 of 13, dev and SEO tools 9 of 15, search engines 4 of 7, e-commerce 10 of 30) is on the IPv6 proxies page.
For AI-visibility monitoring — which pages Perplexity, Copilot or Google's AI Overviews cite for a query in a given market — the anonymous surfaces of the assistants are dual-stack too (perplexity.ai, copilot.microsoft.com, gemini.google.com all publish AAAA records), and sampling them per country from a rotating IPv6 exit is the cheap way to do it; the logged-in versions are not an IPv6 job. The AI assistants section of the use-case hub covers each one.
When IPv6 is the wrong network for SEO
We would rather sell the right $5 residential pack than the wrong $20 IPv6 one, so here is the list:
- IPv4-only engines. DuckDuckGo, Baidu, Naver, and Semrush among the tools: no IPv6 exit reaches them. Use datacenter (from $0.50/GB) or residential (from $1.00/GB) exits on the same account.
- Anything logged in. Search Console, Bing Webmaster Tools, Google Ads, a signed-in Copilot: a login wants one stable, ordinary-looking address for hours, which is a sticky residential or mobile session, not a cheap rotating one.
- Sites that reject hosting ranges by ownership. Our IPv6 addresses are datacenter-issued, on /48 and /64 subnets. A target that blocks on address ownership rather than behaviour will not be helped by a cheaper datacenter address; it needs a residential exit. Search engines block on behaviour, which is why IPv6 works for SERPs.
- Local pack and Maps results at city precision. The IPv6 network targets by country (18 countries). City-level results need the residential network's state and city targeting, or the SERP API's location parameter.
Best IPv6 proxies for SEO: what to look for, and what "unlimited" means
The searches around this topic ask for "best" and "unlimited", so here is the honest version of both. What makes an IPv6 network good for SEO is measurable: subnet diversity (rotation across /48s and /64s, not within one), country targeting on the engines' markets, rotating and static sessions on the same credentials, HTTP and SOCKS5 on one gateway, and a published list of which targets it can reach — which is why we publish the census instead of a claim. QuanticData's IPv6 network has all of that, from $0.20/GB, with the same account carrying residential, mobile and datacenter exits for the jobs IPv6 does not fit.
"Unlimited IPv6 proxies" are unlimited the way a buffet is: a fair-use policy, a speed cap or a concurrency ceiling does the limiting further down the page, and the monthly fee is due whether the tracker ran or not. We sell gigabytes instead — 1 TB for $110, 3 TB for $250, valid 30 days, unlimited concurrent sessions — and at 57 KB per result page a terabyte is on the order of seventeen million SERPs. At that point the word stops mattering.
The full ladder, the IPv4 versus IPv6 comparison and the 102-domain table are on the IPv6 proxies page; the platform-by-platform guides — Google, Bing, Yahoo, Yandex, YouTube and twenty more — are on the IPv6 proxies by use case hub.