Ad Verification: The Complete Guide

Digital advertising has a trust problem that has never fully gone away. You buy an impression targeted at a specific audience in a specific place on a specific kind of page, and a chain of intermediaries stands between your money and the eventual render. Somewhere in that chain, ads get placed on sites you'd never approve, served to audiences you didn't buy, counted for views nobody saw, or simply not delivered at all.

Ad verification is the practice of checking, independently, whether what you paid for is what happened. And it has a specific technical requirement that makes it unlike other web data work: you cannot verify an ad from a datacenter address, because most ad networks won't serve ads to hosting ranges at all. You'd see an empty slot and conclude something was broken, when the emptiness was correct behaviour for the address you used.

This guide covers what verification actually checks, why appearing as the targeted audience is the entire technical challenge, how to build collection that produces evidence rather than impressions, and the specific problems — geographic misdelivery, brand safety, competitive conquesting, and fraud — that verification exists to catch.

What Ad Verification Actually Checks

Six distinct categories, often conflated and requiring different approaches.

Delivery verification. Did the ad actually serve? Campaigns underdeliver, creative fails to load, and slots go unfilled. The simplest check and one that frequently finds problems.

Geographic accuracy. Did the ad serve in the markets you bought? Geo-targeting misconfiguration is common, and money spent showing ads to the wrong country is entirely wasted.

Placement quality and brand safety. What kind of page did the ad appear on? Adjacency to content you'd never associate your brand with is the risk that keeps marketing directors awake, and programmatic buying makes it structurally hard to control.

Viewability. Was the ad in a position where a human could have seen it? Ads render below the fold, in stacked slots, in tiny iframes, or on pages that never scroll. Note that industry viewability standards define specific thresholds of pixels and duration, and your own measurement will not match a vendor's unless you implement the same definition — worth deciding explicitly rather than inventing a threshold.

Creative correctness. Was the right creative served, in the right language, with the right offer, undamaged by the container it rendered in?

Competitive intelligence. Who else is advertising in your placements, on your keywords, and against your brand terms. A useful by-product of the same collection.

Fraud indicators. Ad stacking, pixel stuffing, domain spoofing, and placement on sites whose traffic is fabricated. Detecting these thoroughly is a specialist discipline, but the surface indicators are visible from ordinary verification data and are worth flagging even when you can't investigate them fully.

Understanding the Delivery Chain

Verification makes more sense once you know what sits between a media buy and a rendered ad, because the problems cluster at the joins.

The simplified chain: advertiser, demand-side platform, ad exchange, supply-side platform, publisher, and finally the browser that renders the creative. Real chains are longer, with resellers and intermediaries at several points.

Where things go wrong:

Targeting configuration, at the buying platform. Misconfigured geography, device, or audience settings send impressions to the wrong people. The most common and most easily fixed category.
Exchange and resale. Inventory bought at one point may be resold, so the site an impression lands on can be several steps removed from anything you approved.
Publisher implementation. Slots placed below the fold, stacked, or in containers too small to be seen. Sometimes accidental, sometimes not.
Creative rendering. The right ad served into a container that crops, scales, or breaks it.
Fraud injection. Fabricated traffic, spoofed domains, and invisible placements engineered to register impressions nobody could see.
Measurement discrepancy. Different parties counting differently, so the numbers in two dashboards legitimately disagree.

Why the chain matters for verification design. Each of these fails differently and leaves different evidence. A geographic misdelivery shows up as ads serving where they shouldn't. A resale problem shows up as unfamiliar domains in your placement data. A viewability problem is invisible in delivery counts and obvious in a screenshot. Designing verification around what you're actually trying to catch produces far better results than checking everything shallowly.

Industry transparency mechanisms exist — publisher-declared seller files and supply chain declarations among them — and checking them is a low-cost complement to direct observation. They tell you who a publisher says is authorised to sell their inventory, which makes unauthorised reselling detectable without any rendering at all.

Why This Requires Residential and Mobile Addresses

The technical constraint that shapes everything, and the one most likely to waste a team's first attempt.

Ad networks exclude datacenter traffic. Hosting ranges are heavily associated with bots, and bots are the industry's largest cost. So most networks simply decline to serve ads to addresses that classify as datacenter. From such an address you observe an empty placement, which is indistinguishable from underdelivery and is actually the network behaving correctly.

Targeting is the thing you're verifying. An ad bought for mobile users in Manchester is only observable by something that looks like a mobile user in Manchester. Verifying from anywhere else tells you what a different audience saw, which is not the question.

The implications:

Residential addresses are the baseline requirement for any meaningful verification
Geographic granularity must match the campaign's targeting. If you bought city-level targeting, country-level verification cannot confirm it
Mobile proxies are necessary for verifying mobile placements, both because carrier networks are the targeting criterion and because mobile ad delivery differs substantially from desktop
Device profile must match too. A mobile-targeted campaign verified from a desktop browser profile on a mobile IP is an inconsistent signal, and you may see desktop creative
Datacenter addresses have essentially one use here, which is confirming that the page itself loads

This is the rare use case where the cheap tier isn't a starting point to test. It's structurally unable to do the job, and starting there produces a confidently wrong conclusion that your campaign isn't delivering.

> Tip: Before investigating a suspected delivery problem, confirm the address you verified from was residential or mobile. A surprising proportion of "our ads aren't serving" investigations end here.

Appearing as the Targeted Audience

Beyond the address, verification requires resembling the audience segment the campaign was bought for. Otherwise you observe a different ad and draw conclusions about a campaign that never ran to you.

Geographic identity. IP location, browser locale, timezone, and language headers should all agree and match the target market. A German IP with a US locale is an inconsistent signal that may exclude you from segments.

Device identity. User agent, viewport, screen dimensions, touch support, and device pixel ratio, all consistent with the device class the campaign targeted. Mobile and desktop campaigns are separate buys with separate creative.

Contextual signals. Some targeting is contextual — the page you're on. Verifying placement on a specific site means visiting that site, not a proxy for it.

Behavioural and audience signals. Some targeting uses inferred interests built from browsing history. A fresh browser profile has no history and may therefore fall outside the audience entirely. This is a genuine limitation of verification: for behaviourally-targeted campaigns, you may not be able to observe what the actual audience saw.

Consent state. Where consent frameworks apply, whether you accepted or declined determines whether personalised ads serve. Verifying under one consent state tells you nothing about the other, and both are real user experiences worth checking. In markets with strict consent regimes a substantial share of your audience may be seeing non-personalised inventory, which is a materially different campaign from the one described in the media plan.

Session persistence. Some frequency capping and sequencing depends on session and cookie state. A verification client that discards cookies sees the first impression every time.

The honest limitation: verification observes what an audience-approximating client sees. For broadly-targeted campaigns that's very close to the truth. For narrowly behaviourally-targeted ones, it's an approximation, and saying so is better than overclaiming.

Building a Verification Programme

Define what you're checking. Delivery, geography, placement quality, viewability, creative correctness, or competitive presence. Each needs different collection and different evidence.

Enumerate the placements. Which sites, which sections, which page types. Programmatic buying means your ads may appear anywhere in a network, so verification often starts with discovering where they appeared rather than checking a known list.

Enumerate the markets. Every geography the campaign targets, at the granularity it was bought.

Enumerate the device profiles. At minimum mobile and desktop; more where the campaign distinguishes further.

Calculate the volume:

The multiplication is substantial, and verification is expensive per check because it requires premium addresses and usually full page rendering.

Sample rather than exhaust. Complete verification of a large programmatic campaign is not achievable. Statistical sampling across placements, markets, times, and device profiles produces defensible coverage at feasible cost. Design the sample deliberately rather than checking whatever's convenient.

Vary timing. Ad delivery differs by hour and day. Verification concentrated in one window produces a partial picture, and dayparting is often part of the buy.

Capture evidence, not just observations. Screenshots, full page HTML, the ad's own markup, the requested and served URLs, and the network requests involved. A verification finding without evidence is an assertion, and you'll be asserting it to an agency or a network that has commercial reasons to disagree.

Implementation Notes

The practical mechanics, since ad verification has requirements that ordinary scraping doesn't.

Rendering is usually mandatory. Ads are injected by JavaScript, often several layers deep through nested iframes. Position, viewability, and visual context only exist after render. This makes verification a headless browser workload, with the bandwidth and latency cost that implies.

Wait for the ad, not for the page. Ad calls frequently complete after the page load event. Waiting on network idle or on the ad container becoming non-empty produces far better results than a fixed delay.

Capture the full frame tree. Ads commonly render inside nested iframes, sometimes several deep. A verification client that only inspects the top-level document will find nothing.

Record network activity. The chain of requests an ad call generates reveals which intermediaries were involved and where the creative came from. This is the evidence for supply chain problems and it's discarded by default.

Screenshot at two scopes. The ad element alone, for creative verification, and the full viewport, for context and adjacency. Brand safety findings are visual arguments and need the second.

Determine above-the-fold status properly. Element position relative to viewport at render time, with the viewport set to a realistic size for the device profile. An ad "above the fold" on a 1440p desktop viewport may be well below it on a phone.

Never click. Configure the client so accidental interaction is impossible. Clicks cost the advertiser money and, at scale, are indistinguishable from fraud.

Handle consent gates deliberately. Where a consent dialogue blocks the page, decide whether you're verifying the accepted or declined state, and be consistent. Both are real user experiences and they produce different ads.

Set device profile and locale coherently with the proxy's location. A mismatch may exclude you from the segment you're trying to observe.

Randomise timing and page selection rather than hitting the same URLs on a fixed schedule, which produces a partial and predictable picture.

Sampling Design

Complete verification is impossible for any campaign of scale, so the quality of your sample determines the quality of your conclusions.

Stratify rather than randomise blindly. Your sample should deliberately cover each dimension that matters: market, device profile, placement type, publisher tier, time of day, and day of week. Random sampling across a skewed placement distribution will over-sample your largest publishers and miss the long tail where problems concentrate.

Weight by spend for some questions and by incidence for others. If you're asking "how much of my budget landed badly," weight by spend. If you're asking "how widespread is this problem," weight by placement count. These produce different numbers and both are legitimate answers to different questions.

Over-sample the tail deliberately. Brand safety problems cluster on small, obscure sites that represent a tiny share of impressions and a large share of risk. A spend-weighted sample will barely touch them.

Cover the times you didn't buy. Dayparting failures — ads running outside their scheduled window — are only detectable if you check then.

Sample non-targeted markets. A small but non-zero share of checks should come from geographies you excluded, because overdelivery is silent and costly.

Refresh the sample frame regularly. Programmatic placement lists change constantly. A sample drawn from last quarter's placements misses this quarter's new domains, which are exactly the ones warranting scrutiny.

Record the sample design with the findings. A percentage without its denominator and its stratification is not a finding, and the first competent person to read it will say so.

Size for the claim you want to make. Detecting a problem affecting 1% of placements needs substantially more checks than detecting one affecting 20%. Decide the resolution you need before deciding the volume.

What to Capture

Verification data needs to survive challenge, which means capturing more than a pass or fail.

Environment record: exit IP, detected location, proxy type, device profile, user agent, viewport, locale, timezone, consent state, and timestamp. Without this, any finding is unreproducible.

Page context: URL, page title, page category, surrounding content, and where in the page the ad appeared.

Ad record: creative content, advertiser identity where determinable, ad markup, dimensions, position in the page, whether it was above the fold, and the chain of redirects the ad call passed through.

Screenshot evidence: both the ad in isolation and the ad in page context. Context is what demonstrates a brand safety problem; the ad alone demonstrates creative correctness.

Network trace: the ad calls made, which domains were involved, and what responded. This is where supply chain problems become visible.

Negative results explicitly. No ad served is a finding, and it needs the same environment record as a positive one to be interpretable.

Consistency across checks. The same fields captured every time, so results are comparable and aggregatable.

Brand Safety and Placement Quality

The category with the highest stakes and the most difficult verification.

What you're looking for:

Ads adjacent to content categories you've excluded
Placement on sites of poor quality or dubious provenance
Appearance alongside competitor advertising in ways that undermine positioning
Placement on pages whose content contradicts the campaign message
Ads on sites that spoof legitimate publishers
Placement in cluttered environments where the ad has no practical visibility

Why programmatic makes this hard. Inventory is bought through exchanges, resold, and passed through intermediaries. A single impression may traverse several hops, and the site you end up on may not resemble anything on your approved list. Exclusion lists help and are always incomplete, because new sites appear faster than lists update.

What verification can establish: that ads did appear on specific sites, with evidence. That's the basis for a conversation with an agency or network, and without evidence the conversation doesn't go anywhere.

Practical approach:

Sample broadly across the network rather than only checking approved sites, since the problems are by definition where you didn't expect them
Capture full page context in screenshots, since adjacency is a visual and contextual judgement
Classify page content to identify category violations at scale
Track newly-appearing domains in your placement data, since these warrant review before they accumulate spend
Check the supply chain, including seller identity declarations where publishers implement them, which reveals unauthorised reselling

Geographic Verification

The most straightforward category and one of the most commonly violated.

What goes wrong:

Campaigns targeted at one country serving in others
City or region targeting ignored, delivering nationally
Language mismatched to market
Currency or offer inappropriate for the region
Regional legal requirements not met in the creative

How to check it:

Verify from addresses genuinely in each target market, at the campaign's targeting granularity
Verify from markets you did not target, since overdelivery outside the buy is the failure that costs money silently
Check that the creative served matches the market — language, currency, offer, and legal disclosures
Record the detected location alongside every observation, since a proxy that silently resolved elsewhere invalidates the finding
Compare delivery rates across markets, since a market with substantially lower observed delivery may have a targeting problem

The out-of-market check is the one people skip, and it's frequently where the waste is. Targeting configurations fail open more often than they fail closed.

Display verification dominates the discussion, but two other surfaces have their own requirements.

Search advertising. Ads on search result pages are highly localised and highly competitive, and verification here overlaps with SEO measurement.

Geography is decisive. Search ads are targeted at fine geographic granularity, so verification must match it. This is where city-level residential targeting earns its cost.
Competitive conquesting is visible and actionable. Competitors bidding on your brand terms show up immediately, and it's one of the few verification findings that leads directly to a response.
Ad extensions and formats vary by device and query, so checking that yours render as intended across contexts is worthwhile.
Ad position and the number of ads above organic results determine actual visibility, and both fluctuate.
A managed search results API handles much of this without maintaining a scraper, and returns paid results alongside organic ones as a matter of course.

Social and in-app advertising. Harder to verify systematically.

Most inventory sits behind authentication, which means verification requires logged-in access and raises terms-of-service questions that display verification doesn't.
Feed placement is algorithmic and personalised, so what any client sees is a poor proxy for what the audience saw.
Mobile-first delivery means desktop verification observes something different from what most users experience.
In-app inventory frequently isn't reachable by a browser-based client at all.

The practical position: search ad verification is tractable and valuable; social verification is partial at best, and programmes that claim comprehensive social verification are usually overstating what's technically possible.

Fraud Indicators Worth Watching

Full fraud detection is a specialist discipline, but several indicators are visible from ordinary verification data and worth flagging.

Domain spoofing. The domain declared in the bid request differs from where the ad actually rendered. Comparing your placement reports against observed rendering location surfaces this.

Unauthorised reselling. Inventory sold by parties the publisher hasn't authorised, detectable by checking publisher-declared seller files against the sellers in your supply path.

Ad stacking. Multiple ads layered in one slot, where only the top one is visible but all register impressions. Visible in the frame tree and invisible in delivery counts.

Pixel stuffing. Ads rendered in containers a fraction of their nominal size. Detectable by comparing rendered dimensions against expected creative size.

Implausible site characteristics. Sites with enormous ad density, minimal or auto-generated content, no discernible audience purpose, or recently registered domains carrying high traffic claims.

Placement on domains that appeared suddenly. New domains in your placement data warrant review before they accumulate spend, since fraudulent inventory frequently rotates domains.

Geographic implausibility. Traffic patterns inconsistent with the claimed audience, such as a local publisher whose impressions arrive predominantly from elsewhere.

None of these are proof on their own. They're indicators that justify investigation, and evidence gathered through verification is what makes that investigation possible.

Competitive and Creative Intelligence

The same collection supports analysis beyond compliance.

Competitor presence. Who else advertises in your placements, on your keywords, and against your brand terms. Conquesting — competitors bidding on your brand — is visible in search ad verification and is directly actionable.

Creative benchmarking. What competitors' ads say, how they're formatted, what offers they promote, and how that changes seasonally.

Share of voice. Across a set of placements or keywords, how often does each advertiser appear. A visibility measure that's more meaningful than any single observation.

Message consistency. Whether your own creative is rendering as designed across environments, or being truncated, mis-cropped, or served in the wrong language.

Landing page verification. Where the ad actually leads, whether the page loads, and whether the offer matches the ad. Broken and mismatched landing pages waste the entire cost of the click, and this is one of the cheapest checks available since it needs no rendering and no premium address — just a request to the destination URL from the right market, confirming it resolves, returns a healthy status, and carries the offer the creative promised.

Cost and Scale Management

Verification is the most expensive per-check discipline in this field, because it requires premium addresses and full page rendering. Managing that cost without gutting the coverage is a real design problem.

Where the cost comes from:

Residential and mobile addresses, which are required rather than optional
Headless rendering, which multiplies bandwidth by an order of magnitude over plain fetching
Ad-heavy pages, which are among the heaviest on the web precisely because of the inventory you're verifying
Wait times, since ad calls complete late and each check occupies a browser for longer than an ordinary fetch

Controls that preserve validity:

Sample rather than exhaust, with a deliberate stratified design
Tier by spend. Major placements verified frequently, the long tail on a rotating schedule
Don't block resources you're verifying. The usual bandwidth optimisation of blocking images and scripts destroys the thing you're measuring. Block only what's demonstrably irrelevant, and accept that this workload is heavy
Reuse browser contexts within a session where the verification design permits, rather than launching fresh for every check
Separate cheap checks from expensive ones. Supply chain file checks and landing page reachability need no rendering and no premium addresses. Run those broadly and reserve rendered verification for what requires it
Cap check duration. An ad call that hasn't completed in a reasonable window is itself a finding, and waiting indefinitely wastes capacity
Track cost per finding. If a category of check has never produced an actionable result, its frequency should drop

Coordinate volume with your agency. Verification impressions consume inventory and appear in reporting. Unannounced volume creates discrepancies that someone has to reconcile, and it's a straightforward courtesy to flag it.

Common Mistakes

Verifying from datacenter addresses. The defining mistake in this field. Networks decline to serve, you observe empty slots, and you report underdelivery that doesn't exist.

Not recording the environment. A finding without the exit IP, detected location, device profile, and timestamp is unreproducible and therefore unusable in a dispute.

Only checking targeted markets. Overdelivery outside the buy is where silent waste lives, and it's only visible if you check where you didn't target.

Fixed waits instead of waiting for the ad. Ad calls complete after page load, so a fixed delay captures empty containers roughly at random.

Ignoring nested iframes. Ads render several frames deep, and a top-level-only inspection finds nothing.

Mismatched device profile and IP. A desktop browser on a mobile carrier IP is an inconsistent signal that may serve you the wrong inventory or none.

Screenshotting the ad without context. Brand safety is about adjacency, and an isolated creative demonstrates nothing about the page it sat on.

Clicking, accidentally or otherwise. Costs money and looks like fraud at volume.

Verifying only approved sites. The problems are, by definition, where you didn't expect them. Sampling should cover the network, not the allowlist.

Overclaiming precision on behavioural targeting. A fresh profile may not be in the audience. Say so rather than presenting an approximation as a measurement.

Concentrating checks in one time window. Delivery varies by hour and day, and dayparting is often part of the buy.

Not coordinating volume with the agency. Verification impressions are impressions, and unannounced volume creates discrepancies someone will have to explain.

Reporting Findings

Verification output frequently becomes the basis of a commercial conversation with an agency or network, which changes what a good report looks like.

Lead with evidence, not conclusions. Screenshots with environment records are difficult to dispute. Summary percentages without them invite methodological argument.

Quantify with a stated sample design. "Twelve percent of 4,000 checks across six markets and two device profiles between these dates" is defensible. "Twelve percent of impressions" is not, because you didn't observe impressions.

Separate categories. Delivery failure, geographic misdelivery, brand safety, and viewability are different problems with different owners and different remedies. A single blended figure prevents anyone from acting.

Attribute where you can, and don't where you can't. Verification observes outcomes, not causes. Whether a misdelivery originated at the buying platform or downstream is usually not determinable from observation alone, and asserting it weakens the credible parts of the report.

Rank by spend impact. A problem affecting a small placement matters less than one affecting a major line, regardless of incidence rate.

State the limitations plainly. Sample coverage, behavioural targeting approximation, and consent-state scope. A report that names its own boundaries is trusted more, and it survives the first sharp question.

Track findings over time. A verification programme's value compounds when it shows whether reported problems were actually fixed. One-off audits get acknowledged and forgotten.

Legal and Ethical Considerations

Verifying your own campaigns is legitimate. You're checking what you paid for.
Ad impressions have a cost to someone. Verification traffic that triggers impressions consumes inventory and may affect advertiser billing or campaign metrics. Understand this before running verification at volume, and coordinate with your agency where appropriate.
Don't click ads during verification unless you specifically intend to and understand the billing consequence. Click fraud is a serious matter and accidental clicks at scale are indistinguishable from deliberate ones.
Publisher terms of service may prohibit automated access. Contractual risk rather than criminal in most cases, but real.
Data protection law applies to any personal data incidentally collected.
Rate-limit as a matter of conduct. Verification traffic on publisher sites should be proportionate.
Be accurate about limitations. Verification observes what an audience-approximating client sees, and for behaviourally-targeted campaigns that's an approximation. Overclaiming precision to a client is a professional problem.

Not legal advice, and jurisdictions differ. Where verification informs commercial disputes with agencies or networks, involve counsel early.

Frequently Asked Questions

Why don't I see any ads when I check?

Most likely you're checking from a datacenter address, and the network declined to serve. This is the single most common cause of false "underdelivery" findings. Verify from residential or mobile addresses.

Do I need mobile proxies specifically?

For verifying mobile placements, yes. Mobile ad delivery differs from desktop, carrier networks are frequently part of the targeting, and a desktop profile on a residential IP will see desktop inventory.

Can I verify behaviourally-targeted campaigns?

Partially. A fresh browser profile has no browsing history and may fall outside interest-based segments entirely. This is a genuine limitation and worth stating rather than working around.

How many checks do I need?

Enough for statistical confidence across placements, markets, devices, and times. Complete verification of a large programmatic campaign isn't achievable; a well-designed sample is.

Should I use headless browsers?

Usually yes. Ads are rendered by JavaScript and much of what you're verifying — position, viewability, visual context — only exists after render. This makes verification bandwidth-heavy, which is worth planning for.

Won't my verification traffic distort campaign metrics?

It can. Impressions served to verification clients are impressions. Discuss volume with your agency, and never click.

How do I prove a brand safety problem?

Screenshot evidence showing the ad in full page context, with a complete environment record. Assertions without evidence don't move conversations with networks.

Should I build verification in-house or buy it?

Established measurement vendors offer accredited verification with tag-based integration, which is the standard approach and carries weight in disputes precisely because it's independent. Building your own makes sense as a complement — for competitive intelligence, for spot-checking vendor findings, and for questions a vendor's product doesn't answer. It rarely makes sense as a replacement, since the vendor's independence is part of what you're buying.

How often should I verify?

Continuously at low volume rather than periodically at high volume. Problems appear when placements change, which happens constantly in programmatic buying, and a quarterly audit catches them a quarter late.

Can I verify ads on platforms I don't advertise on?

You can observe what ads appear, which is competitive intelligence rather than verification. Different purpose, same collection.

Getting the Collection Layer Right

Ad verification is the use case where proxy selection isn't an optimisation — it's the difference between meaningful data and none. ProxyScrape's residential proxies provide the consumer ISP classification that ad networks will actually serve to, with country, state, and city-level targeting so verification can match the granularity a campaign was bought at. For mobile placements, where carrier networks are frequently part of the targeting criteria and delivery differs substantially from desktop, their mobile proxies route through 3G, 4G, and 5G connections that present as real subscriber traffic. Their ad verification documentation covers the setup, and note that their datacenter plans — usually the sensible first test elsewhere — are structurally unsuited to this particular job.

→ Compare residential and mobile options for campaign verification

Ad verification is unusual among web data disciplines in that the cheap option doesn't merely underperform — it produces confidently wrong answers, because an empty ad slot from a datacenter address looks exactly like underdelivery and is actually the network working as designed. Everything else follows from getting that right: appear as the audience the campaign was bought for, capture evidence rather than observations, check the markets you didn't target as well as the ones you did, and be honest about what a verification client can and can't see. The findings are only worth as much as the environment record attached to them.