VPN and ads: how it affects geo and filtering

The Yandex Direct and Google Ads help pages on geo delivery do not use the word VPN. It does not follow from their text that changing a device’s geolocation marks a visit as a bot. A W3C note says something else on its own: clearing cookies and using a VPN do not stop later linking by browser fingerprint. Below is what the platforms take into the geography of an impression, and what a forward proxy and a shared IPv4 do not promise about it. This is not an answer to who clicks ads.

What Direct treats as a person’s region

Direct’s geotargeting is control over where impressions are shown. In a chosen region an ad can be seen by people who live there permanently, people who are there now, and people who are interested in the product or service. The help says the system looks at more than coordinates in real time. It looks at the query context, how geolocation changes, and the regions where the person more often buys. An ad can therefore be seen by people who turned off location sharing or changed the device’s geolocation, and by someone who is in another region now but usually buys in the targeted region.

The page’s example: a Perm resident on a business trip in Moscow can see ads for furniture stores in Perm. That is an illustration from the help, not a campaign statistic and not a way to catch a bot. The page does not use the word VPN. A changed device geolocation, in that list, is one case where an impression is still possible, not a mark of an automated visit.

A region on the map is set with a radius from 0.5 to 10 km. A group holds no more than 100 addresses. The help says choosing regions on the map can narrow impressions sharply. The 0.5 km, 10 km, and 100 addresses are limits of the map setting, not a fraud threshold. A smaller circle is not a filter of “suspicious” points.

Where Google Ads takes a physical place from

Google Ads’ help says a physical place is usually derived from the IP address the provider assigns to the computer. If the device is on a Wi-Fi network, they may take that network’s address. If a mobile device leaves through a carrier proxy, they may take the carrier’s address. The page does not call the carrier proxy residential and does not use the word VPN. When location settings are on, GPS, Wi-Fi, Bluetooth, and a cell-tower database are added. The help qualifies the accuracy of each input separately, with no single figure in this retelling.

Interest in a place can be derived from search wording, past queries, and past physical places. Targeting is a best effort from settings, devices, and behavior. One hundred percent accuracy is not guaranteed. An ad for another city, inside these bounds, can be interest in a place or a past physical place, not proof of a substitution. An invalid click does not follow from this description.

A proxy and a shared address are not a definition of VPN

MDN splits proxies into forward and reverse. A forward proxy acts for a client or a group of clients and can hide their address. A reverse proxy hides the server and is used for load balancing, cache, and compression. The CONNECT method opens a tunnel so a client reaches a site over TLS through an HTTP proxy. Not every proxy accepts CONNECT; some cut it off before port 443. The Forwarded and X-Forwarded-For headers can carry the original address, so a proxy does not always hide it. The page does not define a residential proxy and does not describe that market. A forward proxy can hide an address. It does not have to, if a header carries the original.

RFC 6598 sets aside 100.64.0.0/10 for Carrier-Grade NAT on the carrier’s network. Those addresses are not forwarded past the carrier. Several subscribers share one external public IPv4. Geolocation locates the translation server, not the end device. Some sites limit simultaneous sign-ins on one such address. The console example is a class of application breakage: two subscribers with one external address cannot connect to each other. The RFC has no subscriber count and does not say “thousands of people.” A shared address explains why a geo point and “one person” diverge, without the word VPN.

A VPN does not cut a fingerprint

The W3C Privacy Working Group note of 25 September 2025 defines a browser fingerprint as a site’s ability to identify or re-identify a user, a program, or a device from settings and other observable characteristics. It can link visits inside a session and across sessions, including across sites when another origin’s cookies do not hand over an identifier. Collection often happens without an explicit sign. A typical fingerprint cannot be cleared or reset. Clearing cookies and using a VPN do not stop later linking by fingerprint.

The note treats complete elimination of fingerprinting, against an interested party and by widely deployed measures alone, as implausible. Mitigations are not a solution. The surface includes the program’s capabilities and other layers, for example TCP differences. This is guidance for specification authors, not a detector. The page does not retell uniqueness figures. Changing the address through a VPN, on this text, does not end linking by settings. It does not turn a visit into a bot, and it does not cancel a geo impression that Direct builds from interest and purchase history.

Click fraud and sklikivanie and a Yandex refund do not follow from the fact that a VPN does not erase a fingerprint. Geotargeting decides whom to show an ad. A fingerprint decides whether a program can be recognized again. Those are different questions, and the word VPN belongs only to the second.