Naver Band Proxies: What 26 Requests Show

Naver Band is a group app: schools, sports teams, churches, clubs and companies run their groups on it, mostly in private. A small part is public. On 6 October 2026 a public band answered 200 on the first attempt through residential proxies in the US, South Korea and Japan, but over plain HTTP the 16.6 KB response held a title and one word of body. Every post arrived through JavaScript, in a page of 3.3 to 3.6 MB. The exit country changed the interface language, not the posts. Band's robots.txt bans AI training and RAG bots by name, and its Open API only reads Bands a signed-in user has joined.

What people search for

Nobody searches for "naver band proxies" with proxy intent. Google autocomplete completes "naver band" to app, login, logo, download, pc, api, apk and "us". "naver band scraper" collapses to "band scraper" and "band scrape". "band app api" leads to "band api", "band app developer" and "band app integration". The demand is people who want to get at Band data or wire Band into something else.

The US results page for "naver band proxies" has no page about proxies or scraping on Band. It shows the 2014 press release for Band's US launch, Wikipedia, Band's own about page, a KED Global piece on Band's tenth anniversary (6.1 billion posts and more than 156 million app downloads) and a BusinessKorea report of 6.04 million monthly users in the US in November 2024. The related searches ask about free proxies and Android and iOS apps. Nothing explains what Band serves to an automated client. That is what we measured.

What Band serves to a proxied client

Band has three privacy levels for groups: secret, closed and public. Only public Bands, and public "Pages" run by organizations, show content to someone who is not a member. We tested an official public band run by Band itself (BAND Guide, band 62397730), one of its posts, the discover page and a keyword-group page. We did not request any private person's band or profile.

Request (6 October 2026)	Exit	Result	Bytes
Official public band, plain HTTP	US	200, 1st attempt, 1.2 s, 1 word of body	16,607
Same band, plain HTTP	South Korea	200, 1st attempt, 1.3 s, 2 words	16,652
Same band, plain HTTP	Japan	200, 1st attempt, 1.2 s, 1 word	16,667
Same band, rendered, no wait	US	200, 7.2 s, still 1 word	1,813,768
Same band, rendered, 4 s wait	US	200, 12.4 s, posts visible	3,589,986
Same band, rendered, 4 s wait	South Korea	200, 14.1 s, Korean interface	3,332,031
Same band, rendered, 4 s wait	Japan	200, 11.9 s, Japanese interface	3,607,203
One post of that band, plain HTTP	US	200, post text only in title and description	n/a
Same post, plain HTTP	Japan	200 after 3 attempts, 27.5 s	17,132
Same post, rendered	US	Failed twice: tunnel failure, then a 90 s timeout	n/a
/discover, no-JS vs rendered audit	US	200, 1 word both ways, canonical is the homepage	n/a
/keyword-group/Business, same audit	US	200, 1 word both ways, canonical is the homepage	n/a
/about/intro (marketing page), plain HTTP	US	200, 1st attempt, 1,163 words	95,656
robots.txt	US	200, 1st attempt	744
Band-home sitemap index	US	200, 25 gzip child sitemaps dated 5 October 2026	2,988
/policy/terms, plain HTTP	US	200, body reads "Loading"	9,884
/policy/terms, rendered	US	200, 9.5 s, full terms	1,345,814
Open API posts call without a token	US	401 JSON, result_code 300, "oauth error"	177

Three findings matter. First, Band's web app is a shell. The plain HTTP response for a public band carries the band name, its description, a canonical URL and a BreadcrumbList JSON-LD block, and nothing else. A rendered browser that loads the page and stops sees the same nothing: 1.8 MB and one word. Only after waiting a few seconds for the feed does the page fill in: the latest post (2 October 2026), its read count (740) and reactions (13), labels and notices. The rendered page is 216 times the size of the plain one.

Second, a single post leaks a little more over plain HTTP: its first 80 or so characters go into the page title and meta description, because Band wants link previews and Naver's own crawler to see them. The body is still empty. Third, the marketing site is ordinary. The about page came back server-rendered with 1,163 words, so if you only need Band's own product pages, a plain request is enough.

You can reproduce the plain-versus-rendered comparison on any URL with our ../../seo-audit/ endpoint, which fetches a page twice and returns both views and the diff.

What changes by country

Less than on most networks. The response sets content-language from the exit: en-US from the US, ko-KR from South Korea, ja-JP from Japan. The page title switches the brand name to its Korean spelling from a Korean exit. In the rendered page, menu labels, the "posts" tab and buttons follow the same language. The posts themselves were identical from all three countries, in English, because that is what the band's admin wrote.

Exit	content-language	Plain HTTP bytes	Rendered bytes	Posts
United States	en-US	16,607	3,589,986	Same
South Korea	ko-KR	16,652	3,332,031	Same
Japan	ja-JP	16,667	3,607,203	Same

So geo-targeting on Band is about seeing the interface your members see, for example when you publish announcements to a Korean or Japanese audience and want to check how the page reads to them. It does not unlock different content. Band also keeps users under 18 out of public Band search; the rendered page states that they can join only by invitation.

What Band's rules say

Band's robots.txt, fetched on 6 October 2026, is short and pointed. It disallows /n/ for everyone, blocks two SEO crawlers completely, and lets Naver's own crawler (Yeti) into public band and Page posts only. Then it names ten AI crawlers, from GPTBot and ClaudeBot to CCBot and Kakao's DaumRAGBot, and disallows all of them, under a comment that says bot access for "AI training and retrieval-augmented generation (RAG) is strictly prohibited". If you plan to feed Band content into an LLM pipeline, the site owner has told you no in writing. You can paste the file and any path into our ../../tools/robots-txt-tester/ to see the verdict per user agent.

The Terms of Use load only with JavaScript; the plain response is a page that says "Loading". Rendered, they forbid illegally collecting or disclosing other subscribers' personal information and their usage history, and using the service for profit-making purposes without the company's approval. Band is mostly made of private groups of named people, often families, school classes and children's teams, so the personal-data line is the one that matters. A public band's posts are visible; the people in it did not sign up to be a dataset.

The sanctioned route is the BAND Open API. Its guide says you register an app and get an access token, and the API lets you display the contents of a Band you have joined in your own site or app, post to your Band and build your own client. The Get Posts endpoint takes a user access token, a band key and a locale, and pages through posts with a cursor. Without a token it answers 401 with a 177-byte JSON body, result_code: 300 and "oauth error". The API reads with the permissions of the user who authorized it; it is not a door into Bands you are not in.

What a proxy is actually useful for on Band

Job	Proxy type	Notes
Check how your organization's public band or Page reads in Korea, Japan or the US	Residential, country targeted, rendered	Interface language follows the exit; about 3.3 to 3.6 MB a view
Check link previews of your public posts (title, description, image) by locale	Residential, plain HTTP	The head carries them in about 17 KB
Spot bands impersonating your brand or school	Residential, plain HTTP on known URLs	Band name and description are in the plain response
Run your own band's admin account from a stable address	Static ISP, one account per address	Consistency, not scale
Read your own Bands from your own software	None needed	Open API with your access token
Collect posts or members from bands at scale, or feed them to an AI model	None	robots.txt bans AI bots; the terms forbid collecting members' personal data

Every request above went through ../../residential-proxies/. The public pages loaded on the first attempt from all three countries; the only retries were on a single post page from Japan and on the Open API call. There is no case for a large rotating pool on Band: what matters is the country of the exit. For an admin who manages a band, ../../isp-proxies/ keeps the same address every session. We did not test mobile or datacenter exits on Band and make no claim about them, and we found no rate limit worth stating at the volumes we sent.

Errors you will see, and what each means

• A 200 with one word of body, about 16.6 KB. Not a block. It is Band's normal response to any client that does not run JavaScript. Check the title and description; that is all there is.
• A rendered page of about 1.8 MB with no posts. The browser stopped before the feed loaded. Waiting a few seconds took the same page to 3.3 to 3.6 MB with posts.
• A 200 whose canonical is the homepage. On /discover and keyword-group pages, a logged-out visitor got the homepage shell. Discovery on the website needs a login.
• 401 with result_code: 300 from openapi.band.us. No or invalid access token. Register an app and authorize a user.
• Slow responses and tunnel failures on single post pages. We saw a 27.5-second, three-attempt fetch from Japan and two failed renders of one post from the US. On pages this heavy, set timeouts above 30 seconds and treat a failure as a reason to slow down, not to add exits.

What it costs

For the jobs that make sense, the traffic is small but not trivial, because a rendered band is heavy. Say a school district checks its 10 public bands in English and Korean once a day: 20 rendered views a day, about 600 a month. At about 3.6 MB each that is roughly 2.15 GB a month, which at $0.80/GB on Residential Basic is about $1.72 a month. The same 600 views over plain HTTP, for title and description only, are about 10 MB, under one cent.

On pages this heavy, paying per page can beat paying per gigabyte. Our ../../web-scraping-api/ charges $0.001 per page with JavaScript rendering and $0.0002 without, and only for successful requests: 600 rendered views would be $0.60 a month, against about $1.72 of residential traffic for a browser you run yourself. Every account gets $2 of free API usage a month, which covers that example several times over.

On Band the real cost question is not bytes. It is whether the job is about your own groups and your own audience, where the Open API and a country-targeted check are enough, or about other people's groups, where the rules say no. For other Asian messaging networks, see our posts on ../line-proxies/ and ../zalo-proxies/.