3

I have been researching different ways a web crawler might be blacklisted/blocked from a web-server and how to potentially circumvent that. One of those ways is to change the referrer header on the request. I have been going to various places trying to figure out the benefit of doing this, but I believe I am thinking about it too hard and have tunnel vision.

A couple other ways to disguise yourself from web-servers you are attempting to crawl resources from, are changing the User-Agent header on the request, or by Proxying your request through other servers thereby making the call with new public IPs each time. This makes sense since they can't tell the requests are all coming from the same machine, or using the same client agent for making the request. For all they know, it's coming from potentially thousands of machines, from 10-20 different browsers and are all unique users. Is this the same benefit of changing the referrer header in the request? Im getting hung up on how that's implemented. Would you just cycle through hundreds of randomly generated URLs and add a new one to the request headers each time ...

For Example: ref1 = www.random.com, ref2 = www.random2.com, ref3 = random3.com

  • 1
    As there are many reasons for the Referer header not to be sent, I don't think there are really than many sites that would block you based on their absence. The major reasons are keeping the default user-agent of a library (e.g. libcurl) instead of a regular browser and excessive traffic from a single IP. But it really depends on whether you're planning to crawl the same site repeatedly or crawling many different sites, and in the former case, how much effort the site puts into trying to prevent crawling. – jcaron Jan 4 at 12:21
4

The idea would be to make your requests look as much like a real browser as possible. Real browsers send referrer headers. You'd want to send referrer headers that look as much as possible like the referrer headers that a real browser sends.

A real browser never sends random referrer headers. It sends a referrer header for the previous page. Most referrers then end up being pages from the same site.

The ideal strategy would be to crawl the home page without a referrer header. This would mimic a user who types in the home page URL which is very common. As your crawler views pages on the site it would keep track of not only URLs it finds, but also which pages it found those URLs on. It would use one of the URLs where it found the link for the referrer of the new page it fetches.

  • No referrers - Gives away that you are a bot.
  • Random referrers - Gives away that you are a bot and probably pollutes analytics. That type of bot is likely to be blocked even faster than a no referrer bot.
  • Home page referrer - Always using the home page as the referrer can sometimes get around checks for no referrer and looks somewhat legitimate.
  • Linking page as referrer - The strategy I described above is most like a real browser, but even then the order in which you visit pages is likely to be different than a real visitor.
  • As @jcaron notes in his comment on the question, no referrer is not conclusive since UAs don't always send. Home page referrer is dead give-away as bot if you don't have links to referred page on home page. It's usually obvious to my bot filters what's a bot and what's not, using a combo of IP-based info (hostname, org name, geography, etc) and client-supplied info (headers, etc). Hits from server farms are almost always bots, but could be VPNs... headers usually give away bots. Vast majority of bots I see make tell-tale mistakes in individual HTTP headers or in the way they combine headers. – pseudon Jan 4 at 17:28
  • Even without more complex behavioral analysis to detect bots (like tracking the path a visitor takes through the site), other signals that contribute to bot determination are: lack of response (or incorrect response) to cookies, redirects of various kinds, JavaScript / AJAX, websockets, and other interactive client-server behaviors. – pseudon Jan 4 at 17:32
  • 2
    It is very hard to build a bot that isn't detected by some trivial heuristic. Many bots don't download CSS and images either. – Stephen Ostermiller Jan 4 at 17:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.