Invalid traffic, bots clicking ads and never buying, inflates clicks and crushes ROAS. Split general invalid traffic from sophisticated invalid traffic, filter the first, investigate the second, and never optimize a campaign on numbers the bots wrote.
The two kinds of invalid traffic
The industry splits invalid traffic into two kinds, and the split matters because the fix is different. General invalid traffic, GIVT, is the easy stuff: known crawlers, bots that identify themselves, accidental double clicks. It is recoverable waste, and ad platforms filter most of it automatically. Sophisticated invalid traffic, SIVT, is the hard kind: bots that mimic human behavior, rotate through residential proxies, and click ads in patterns designed to look real. SIVT is fraud, and no platform filter catches all of it.
The tell is the gap between the ad report and the site. Clicks rise 40 percent, sessions rise 5 percent, and conversions do not move. The clicks are landing on the site and bouncing in two seconds, or landing and doing nothing at all. A campaign where clicks and conversions diverge is not a creative problem. It is a traffic problem, and the creative team cannot fix it.
Related reading on this site: Home and Friendly Fraud: The Chargeback Where the Customer Is the Fraudster cover the neighbouring problems in more depth.
How to prove it before you spend more
The first check is the session log. A real click leaves a trail: a landing page, a scroll, a product view, a few seconds of dwell. A bot click leaves a landing page and a bounce, or a landing page and nothing. Pull the last hundred clicks from the campaign, check how many produced a session longer than ten seconds, and the share that did not is the invalid share, roughly.
The second check is the IP and device data. A hundred clicks from the same device fingerprint, or from a block of residential proxies in a country the store does not ship to, is not an audience. It is a farm. The third check is time. Bot clicks cluster in bursts, fifty in five minutes at 3 a.m., while human clicks spread across the day. A campaign with a burst pattern and a bounce rate above 90 percent is paying bots.
What to do about it
First, stop optimizing on the polluted numbers. A campaign that looks profitable because bots inflate the clicks is not profitable, and raising its budget raises the fraud with it. Pause the campaign, filter the invalid traffic, and re-measure. The real ROAS is usually lower than the reported one, and the honest number is the one to optimize against.
Second, tighten the targeting. Exclude the geographies, devices, and placements where the invalid traffic concentrates, and require a conversion signal the bots cannot fake, like a completed purchase rather than a click. Third, file for refunds. Ad platforms refund invalid clicks when the evidence is documented, and a log showing burst clicks with two-second bounces is documentation. The refund will not cover everything, but it covers enough to matter, and it puts the platform on notice that the traffic is being watched.
Frequently asked questions
What is the difference between GIVT and SIVT?
GIVT is general invalid traffic like known crawlers and accidental clicks, which platforms filter automatically. SIVT is sophisticated invalid traffic, bots that mimic humans through residential proxies, which is fraud and needs investigation.
How do I know bot traffic is hurting my ROAS?
Clicks rise while sessions and conversions stay flat. A campaign where clicks and conversions diverge, with bounce rates above 90 percent, is paying bots.
Can I get a refund for invalid clicks?
Yes, when the evidence is documented. Ad platforms refund invalid clicks, and a log showing burst clicks with two-second bounces is the documentation they need.
Should I pause a campaign with invalid traffic?
Yes, before raising its budget. Optimize against the real ROAS after filtering, not the reported number the bots wrote.