Proxies for Ad Verification: Seeing the Ad Your Audience Sees

Ad verification has an awkward property: the thing you are checking is designed not to be shown to you. Campaigns target a geography, a device class and often a carrier, and the verification team sits in none of them. Checking from the office confirms that the office is not in the target segment.

You have to match the segment, not the country

An ad bought for mobile users in Munich is delivered to mobile users in Munich. A desktop browser on a corporate connection in London is outside the buy, so the ad slot fills with something else — and the screenshot shows a competitor's creative in your placement, which looks like fraud and is actually just correct targeting.

Matching the segment means matching four things at once:

• Geography, usually to city rather than country.
• Connection type, because mobile-network inventory is bought separately from fixed-line.
• Device, since the creative and often the whole placement differ.
• Carrier, on campaigns that buy per operator.

Get one wrong and you are verifying a different buy than the one you paid for.

Geography first

Rotating residential reaches 195 countries with city and ASN targeting, at $2.65 per GB down to $1.59 at 500 GB. City targeting is what makes this workable — ad delivery is bought at metro level far more often than national level.

for city in munich hamburg berlin cologne; do
  curl -x "http://user-country-de-city-$city:[email protected]:8080" \
    -s "https://publisher.example/article/4417" \
    -o "placement-$city.html"
done

Run the city list from the media plan, not from a list of large cities. If the buy is four metros, verifying twelve tells you nothing extra and costs four times as much.

Mobile inventory needs a mobile network

This is the part most verification setups skip, and it is why mobile placements come back "not serving" when they are serving perfectly well.

A large share of mobile inventory is targeted at carrier networks. A residential fixed-line address with a mobile user agent is not the same thing to an ad server — it can read the connection type, and it does. Verifying mobile creative from a home broadband exit produces a false negative.

Rotating mobile covers 60+ carriers across 38 countries at $3.40 per GB, with ASN targeting down to a single operator:

curl -x http://user-country-us-asn-21928:[email protected]:8080 \
  -s https://publisher.example/article/4417

AS21928 is T-Mobile US. If a campaign was bought against specific carriers, verify against those carriers rather than against "mobile" in general.

Hold the session for the length of the impression

Ad verification is one of the few proxy workloads that genuinely needs sticky sessions, because an impression is not a single request.

The page loads, the ad server is called, a bid happens, the creative loads from a CDN, and tracking pixels fire. If those requests arrive from four different addresses in three countries, the ad server sees an incoherent user, and either fills with a default or drops the impression. Then you record a discrepancy that exists only in your measurement.

curl -x http://user-country-de-city-munich-session-imp5518:[email protected]:8080 \
  -s https://publisher.example/article/4417

One session token per impression check. Residential sessions hold for up to 30 minutes and mobile for up to 60, both far longer than a page load, so the token is about coherence rather than duration. Release it and take a new one for the next check — reusing a token across hundreds of impressions rebuilds the concentration problem rotation was solving.

There is more on the tradeoff in sticky sessions versus rotating proxies.

Do not block the resources you are verifying

Most scraping advice says to block images and media to save bandwidth. Ad verification is the exception that reverses the rule: the creative is the thing you are measuring, and a pipeline that aborts image requests has verified nothing.

Keep images, media and scripts. Block only what is genuinely irrelevant — fonts, and stylesheets if you are not screenshotting. Then budget accordingly: verification is one of the more bandwidth-hungry proxy workloads precisely because you have to load the payload.

If cost becomes the constraint, cut the number of checks rather than the fidelity of each one. Ten complete impressions tell you more than a hundred partial ones.

What discrepancies actually mean

A verification run that reports a missing ad has four possible causes, and only one of them is a problem with the campaign:

1. The segment did not match — wrong city, wrong connection type, wrong device. By far the commonest, and it is a fault in your setup.
2. The impression was incoherent — requests split across addresses, so the ad server declined to fill.
3. No inventory at that moment — normal, and the reason single checks are worthless. Only rates across many checks mean anything.
4. A genuine delivery problem — the one you are looking for.

Work down that list before escalating. Reporting a delivery failure that turns out to be a fixed-line exit checking mobile inventory costs credibility that is expensive to rebuild.

Instrument the verifier

• Fill rate per city and per carrier, never aggregated. An aggregate fill rate hides the one metro that is failing.
• Assert the placement rendered. A page that loaded but contains no ad slot is a failed check, not a zero-fill.
• Record the exit IP, city and ASN with every capture, so a disputed result can be reproduced from the same segment.
• Screenshot with the creative loaded, not before it.

For the wider workflow, see ad verification.

A checklist

1. Take the city list from the media plan.
2. Verify mobile inventory from a carrier exit, not a residential one.
3. Match the ASN when the buy was made per carrier.
4. One session token per impression; release it afterwards.
5. Load images and scripts — they are the measurement.
6. Report rates across many checks, never single observations.
7. Store exit IP, city and ASN with every capture.