Adwords ip filter?

Does it exists?

[djingo]

I have the problem that over 70 pct. of all clicks are the same ipīs clicking over and over again. Its like they don't bother bookmarking my page and thinks its easier just to type in the search word and clicking the ad.

And Adwords doesnt allow IP-filtering right? Is there anything I can do about it?

[gregbo]

Are you billed for these clicks? We have been led to believe that G can detect this type of traffic and filter it.

[mimmo]

We also have multiple clicks from the same IP/person. Someone looking for 'blue cars', clicks on your ad, visits your page and then if s/he wants to return to your page s/he will make the same search and click again. I think this is pretty standard.

Or someone looking at multiple ads and then clicking back to revist your page... I do not know if you are then charged multiple times in this case.

[gregbo]

Well, only G knows what they are charging for - they won't disclose on a per-click basis. The latest commentary from Ghosemajumder indicates that G can filter out "invalid" clicks, for example when someone visits a landing page, visits some other page, then revisits the landing page using their browser's "back" button.

If you're getting repeated clicks from the same IP that don't differ in other ways (user agent, cookie, other HTTP data), they may be fraudulent, or they may be some user innocently making the same request repeatedly. If this type of use is costing you too much money, and you think G isn't discounting the clicks, you have to do something like reduce your spend in accordance with the amount of repeat traffic you get.

[mimmo]

With Cookies and Javascript disabled, no way you can track the back button. Otherwise very feasible.

[gregbo]

My mistake: G would never actually see the click when someone hit the back button, because the landing page would be fetched from the advertiser's server. These duplicate entries in the advertiser's logs have been mistaken as click fraud. G provides a means of tagging that identifies when a page has been reloaded.

The "invalid click" problem comes into play when there is some repeated clicking (or otherwise requesting) of the ad. Most of these clicks are not charged for, so they say.

More info about this here [adwords.blogspot.com].

Among other things, we don't know how often/much they charge for common scenarios, such as someone resubmtting the query and the click, instead of bookmarking the landing page. Fraudsters can take advantage of this behavior, and if they don't trip any of G's filters (or manual reviews), the clicks would be charged for. With a wide spread of traffic from topologically distributed sources, such as could be generated by botnets, they could make these clicks look like what's found in common scenarios that's considered valid.

[mimmo]

When you connect with dialup (or do not have a static IP), your IP will change every connection. Even if you are on broadband, you may be connected via a proxy and then your IP can change, even for every request!

So, if you also have your cookies disabled (and maybe even javascript), there is no technical way to track... is it the same user connecting twice or two different users?

Maybe if you have 5 clicks from the same IP in 20 min, that could be tagged as repeated clicks and not charged for, but what about 5 clicks from the same IP in 1 day? What if each click is paid by the advertiser 30 cents?

I still wonder how really G marks fraud. Maybe is based on some statistical analysis for keywords, eg today 'blue cars' was clicked 3 times more than normal, for only this account.... CTR is too high, historically... or something similar...

But, at the end of the day, I believe repeated clicks and a certain level of fraud is the price of internet ads. When you bid 30 cents on a keyword, you are bidding on Pay Per Click, not Pay Per Visitor.

[gregbo]

I still wonder how really G marks fraud. Maybe is based on some statistical analysis for keywords, eg today 'blue cars' was clicked 3 times more than normal, for only this account.... CTR is too high, historically... or something similar...

How well something like this works depends on how "clean" the traffic is that is considered normal. At least from reading the Tuzhilin report, I've come to the conclusion that they did not realize how much fraud there was initially (especially the carefully concealed types I've discussed). So if there is a certain amount of fraud that's already marked as normal, spikes will only (possibly) identify the new fraud - they won't do anything about the old fraud.