Top 10 Signs Your Campaign Traffic Isn't Human

Bots leave tells. Here are ten of the clearest, straight from the metrics you already track — what each looks like, and the innocent explanation you should rule out first.

None of these is proof on its own. Bot detection from platform reports is a game of accumulating evidence: each signal below has a legitimate explanation that fires occasionally, and almost none fires repeatedly in the same placement at the same time. Two or more clustering together in one traffic source is the threshold at which it stops being coincidence.

Work through them against a single large campaign, segmented by placement and publisher rather than campaign-wide — aggregate numbers average the fraud away, which is exactly why it survives in monthly reporting.

Why your platform didn’t catch these

Most major platforms already strip general invalid traffic — declared crawlers, known data-centre ranges, published bot lists — before you ever see a report. That filtering is real, and it is why the crudest signals show up less often than they used to.

What it doesn’t catch is traffic engineered to look human: hijacked consumer devices, residential proxies, farmed handsets. Those arrive in your reports already counted as valid, which is why the tells below are behavioural rather than technical. You are looking for audiences that behave impossibly, not for traffic the platform has labelled suspicious.

1. 100% bounce with zero-second sessions

Real people linger, even briefly. Instant bounces at scale are automated hits, not curious visitors who changed their mind.

Rule out first: broken tracking on the landing page, or a redirect that fires before analytics loads.

2. Traffic spikes at 3am

Human demand follows the clock. Round-the-clock delivery, or surges between 2am and 4am local time, usually means a script running on a schedule rather than an audience.

Rule out first: genuine international audiences in other time zones, and scheduled email or push sends.

3. Impossible click-through rates

A CTR many times the channel norm isn’t a great creative — it’s a bot clicking everything it’s served. Display CTRs several points above benchmark are a red flag, not a win.

Rule out first: a small sample size, or an accidental placement in a high-CTR unit such as an interstitial.

4. Ancient operating systems

Clusters on long-obsolete OS versions usually mean emulated or spoofed devices pretending to be real handsets. Genuine audiences skew heavily to recent releases.

Rule out first: a legitimate older-device audience — some categories and markets genuinely skew old.

5. Data-centre IP addresses

Consumers don’t browse from server farms. Data-centre traffic is non-human until proven otherwise, and it is one of the easiest signals to check.

Rule out first: corporate VPN traffic, which can legitimately present as data-centre in some B2B campaigns.

6. Identical user agents at scale

Thousands of “users” sharing the exact same device fingerprint is a bot signature, not a coincidence. Real audiences produce a long tail of device and browser combinations.

Rule out first: in-app webviews, which can flatten user-agent diversity legitimately.

7. Geo that doesn’t match your targeting

Impressions from regions you explicitly excluded signal spoofing or corrupted location data leaking budget abroad.

Rule out first: VPN use by genuine customers, and known IP-geolocation inaccuracy at city level.

8. A conversion cliff

Big impressions and clicks, then almost no downstream action — the funnel collapses because there was never a human in it. Compare conversion rate by placement, not just campaign-wide.

Rule out first: a genuine landing-page or checkout problem affecting everyone equally.

9. Duplicate device IDs

The same advertising IDs appearing again and again inflate “unique reach” with repeats, recycled handsets and machines. Reach that barely moves as spend climbs is the tell.

Rule out first: a genuinely small, heavily-retargeted audience where high frequency is expected.

10. Suspiciously perfect viewability

Flawless viewability with no variance is often ad stacking or automated rendering, not attentive humans watching. Real viewability is messy and varies by placement.

Rule out first: a single premium placement with genuinely strong viewability — check the distribution, not the average.

What to do when the flags cluster

One flag can be innocent, which is why each gets a counter-explanation above. Several together, concentrated in the same placements or traffic sources, means a chunk of your budget is reaching machines — and only device-level verification can tell you how much.

Treat these as a dashboard, not a tick-box checklist — and when you want to check them systematically, work through the full audit. Any single flag has an innocent explanation, but they rarely travel alone, and clusters concentrated in particular placements are how you find the inventory quietly draining a campaign.

When you see the pattern repeat, don’t just optimise around it. Excluding a placement moves the budget; it doesn’t recover the spend, and the same junk inventory will resurface under a different domain next month. Verify at the device level and cut it off at the source.

What clustering looks like in practice

A worked example. One publisher in your placement report shows 100% viewability, a CTR four times the campaign average, 90% of sessions under a second, and a conversion rate of zero. Individually each has an excuse: a premium unit, strong creative, a tracking fault, a broken landing page. Together, in one source, they describe a placement that is serving your ads to software.

That is the judgement the list is designed to support. You are not looking for a single damning metric — you are looking for a placement where every innocent explanation would have to be true at once.

If two or more of these look familiar, we’ll verify a live campaign for you.