Last updated: September 8, 2026
Spotting a supply problem before it eats an entire test spend on internet advertising platforms
A fraud filter catches the obvious cases, the datacenter IPs and the impossible click patterns, but a supply source built around low-effort real devices can pass every automated check while delivering almost nothing of value. Traffic quality is not a single number a dashboard reports, it is a pattern that only becomes visible once several signals are checked together rather than trusted one at a time. A small test budget spent deliberately on verification before scaling up saves a far larger budget from disappearing quietly into low-quality inventory on internet advertising platforms later.
Click-through rate alone tells you almost nothing about quality on internet advertising platforms
A suspiciously high click-through rate is the classic warning sign, but a suspiciously average one can hide the same problem just as easily on internet advertising platforms, since a seller aware that extreme numbers draw scrutiny will simply tune a bot pattern to land in a plausible-looking middle range instead. Treating CTR as a pass or fail threshold misses this entirely.
Post-click behaviour tells a far more honest story than the click itself does. Time on page, scroll depth and return visits within a session are harder to fake convincingly at scale, and a source with a normal click rate but near-zero engagement afterward is a far stronger warning than a raw click number could ever be by itself.
Scroll depth in particular is worth pulling separately rather than trusting a blended average across all traffic, because a source mixing genuine visitors with a smaller share of low-quality ones will show an average that looks acceptable while a segmented view reveals two very different populations sitting inside the same number.
The single cheapest test worth running before any real spend
Send a small, deliberately unattractive test campaign with no real offer attached and watch what comes back. Genuine human traffic mostly ignores an unappealing offer and moves on, while several categories of low-quality traffic interact with almost anything placed in front of them regardless of relevance, and that gap is often visible within a single day.
Keep the test creative genuinely dull rather than mildly interesting, since a middling offer can still attract a reasonable number of curious real visitors and muddy the comparison, while a genuinely unappealing one gives the cleanest possible read on which portion of a source is willing to interact with literally anything shown to it.
Run this test on a fresh account rather than an established one wherever possible, since an established account may already carry a reputation, good or bad, with the seller's own systems that quietly shapes what gets delivered, while a fresh account gives the cleanest possible baseline before any history accumulates.
Geographic mismatches reveal fraud that a fraud filter itself frequently misses on internet advertising platforms
A campaign targeted narrowly at one country that somehow receives meaningful traffic from a completely unrelated region is showing a mismatch that automated fraud detection does not always catch, because the traffic may still pass every individual technical check while simply originating from the wrong place entirely for the campaign in question.
This kind of mismatch is especially common with reseller supply chains, where a source several layers removed from the original publisher may not even know, let alone disclose, that a meaningful share of its inventory originates far outside the region its own dashboard claims to represent.
Cross-referencing delivered geography against targeted geography on a weekly basis catches this quickly and costs nothing beyond a few minutes with an export, yet it is one of the most commonly skipped checks across internet advertising platforms precisely because the dashboard reports geography as though targeting alone guarantees it.
A related check worth running alongside geography is time-zone alignment. Traffic claiming to originate from a specific country but arriving heavily concentrated during hours that make no sense for that country's local daytime is often the clearest and least ambiguous signal available, and it requires nothing more than an hourly breakdown already sitting in most reporting exports.
| Signal to check | What a mismatch usually means |
|---|---|
| Delivered geography vs targeted geography | Traffic routed through the wrong pool |
| Session duration distribution | Uniform durations suggest scripted behaviour |
| Return visitor rate | Near zero can indicate one-shot bot traffic |
| Device model diversity | Too few models can indicate emulation |
Session duration distributions expose scripted traffic better than an average ever will on internet advertising platforms
An average session duration can look entirely normal across most internet advertising platforms while hiding a distribution where most sessions last either under two seconds or suspiciously close to one exact figure, and averaging those two clusters together produces a perfectly plausible-looking number that conceals a very real problem underneath it.
Pulling a raw distribution rather than trusting a single summary statistic takes an extra export and a few minutes in a spreadsheet, and that small extra effort is usually enough to separate a genuinely engaged audience from a scripted one dressed up to resemble one on paper.
Reading a histogram instead of trusting an average
Look for a sharp spike at any single duration value, since real human sessions naturally spread across a wide range while scripted traffic frequently clusters tightly around whatever value the script was configured to simulate, and that spike is usually visible at a glance once plotted.
Building this chart takes little effort once the raw session data has been exported, and repeating it monthly for every active source turns a one-off investigation into a routine health check that catches a degrading source months before its summary metrics would ever show anything unusual.
Independent verification tools catch what a seller's own dashboard has no incentive to surface on internet advertising platforms
A seller reporting its own quality metrics has little reason to highlight a problem with its own supply, which is why a third-party verification tag, run independently of the seller's reporting, remains worth the modest cost even on a source that otherwise looks acceptable from the inside. I started requiring one on every new source after a single bad month, a decision confirmed against the verification framework described on internet advertising platforms.
The modest cost of a verification tag is easiest to justify by comparing it against the size of the budget it protects. A tag costing a small fraction of one percent of monthly spend is cheap insurance against discovering, three months in, that a much larger share of that spend never reached a real person at all.
Compare the independent verification numbers against the seller's own self-reported figures directly, because a large, consistent gap between the two is itself a signal worth acting on regardless of which individual number looks acceptable in isolation.
Keep the comparison over time rather than as a single snapshot, since a gap that widens gradually month over month tells a very different story than a gap that has stayed roughly constant since the account was opened, and only a running record makes that distinction visible at all.
| Verification approach | What it catches that self-reporting misses |
|---|---|
| Third-party viewability tag | Impressions never actually rendered |
| Independent fraud scoring | Patterns the seller has no incentive to flag |
| Manual sample audit of raw logs | Anomalies too subtle for automated summaries |
A single bad week is not proof of a bad source on internet advertising platforms, but a pattern is
One unusual week can happen to any legitimate source on internet advertising platforms for reasons that have nothing to do with fraud, a seasonal dip, a technical outage upstream, or a temporary shift in the publisher mix behind the scenes. Cutting a source after one bad week throws away sources that would have recovered on their own the following week.
Setting a fair threshold before judging a source at all
Give a new source at least three weeks and a meaningful sample size before drawing a conclusion either way, and write the threshold down before the test starts rather than deciding after seeing the results, because a threshold chosen after the fact tends to quietly match whatever answer already felt right going in. I keep this threshold written into a short checklist now, a habit built after reading the testing framework on internetadvertisingplatforms.com.
Sharing that written threshold with anyone else who approves new sources matters just as much as setting it, because a threshold known only to one person tends to drift quietly the moment that person is on holiday and somebody else makes the call instead under time pressure.
Written down or not, the threshold only works if it is actually consulted before a decision rather than after, and building that check into whatever approval process already exists is worth more than any cleverness in how the threshold itself was originally calculated.
Treat traffic quality as an ongoing check rather than a one-time gate passed at signup on internet advertising platforms, because a source that verified cleanly six months ago can degrade quietly over time as its underlying supply mix shifts without any announcement ever reaching a buyer relying on it.