Invalid traffic: what it is, how to spot it and how to cut it

An exchange sits between the people who send traffic and the people who pay for it. When some of that traffic is not human, you are in the middle of the argument. Demand partners ask for credits, supply partners dispute the numbers, and your margin disappears in between. Understanding invalid traffic is not optional for an exchange operator. It is part of the job.

What counts as invalid traffic

Invalid traffic, or IVT, is any activity that should not be counted or billed as a human viewing an ad. The term is broader than "fraud" on purpose: it covers the honest crawler as well as the bot farm. The Media Rating Council's guidelines split it into two categories, and most measurement vendors follow that split.

General invalid traffic (GIVT) can be caught by routine filtering with lists and standard checks:

  • known data-centre traffic coming from servers rather than consumer devices;
  • known bots, spiders and crawlers, including legitimate ones such as search engines;
  • pre-fetch and pre-render activity, where content loads before anyone looks at it;
  • requests with non-browser user agents or other clearly invalid data.

Sophisticated invalid traffic (SIVT) is built to look human and needs deeper analysis, several corroborating signals and sometimes manual review:

  • bots that imitate human behaviour;
  • hijacked devices, adware and malware producing impressions or clicks in the background;
  • invalid proxy traffic that hides where requests come from;
  • hidden, stacked or covered ads;
  • misrepresented inventory, where a request claims a site or app it does not come from.

What it looks like in an exchange's statistics

You see traffic in aggregate, by partner, which is a good vantage point. None of these signals proves anything alone, but each is worth investigating:

  • Spikes without a reason. One supply partner suddenly sends far more requests for the same placements, with no seasonal or commercial explanation.
  • No daily rhythm. Traffic claiming to come from one region that stays flat around the clock.
  • Concentration. Large shares of requests from a narrow set of IP ranges, devices, user agents, domains or app bundles.
  • Mismatched identity. Requests claiming premium domains or apps whose ads.txt or app-ads.txt does not authorise the selling account.
  • Demand reacting. A demand partner's bid rate on one source drops sharply while it stays normal elsewhere. Buyers with their own filters often notice first.
  • Persistent discrepancies. Your counts and a partner's counts drift apart on one source only.

Look at traffic by supply partner, and inside each partner by domain or app. IVT usually hides in a corner, not across the whole exchange.

How an operator cuts it

  • Onboard carefully. Ask new supply partners where their traffic comes from, check their publishers' ads.txt and app-ads.txt, and start them on limited volume.
  • Filter the basics. Known data-centre ranges, declared bots and malformed requests should not reach demand partners at all.
  • Use independent measurement where your volume justifies it, and share findings with partners in a consistent format.
  • Pass transparency signals on. Correct schain and sellers.json entries make spoofing harder and help buyers trust the rest of your supply.
  • Put it in the contracts. Define which measurement counts, how invalid impressions are handled financially, and that persistent offenders can be suspended.
  • Act. When a source keeps showing up in the bad numbers, cut it. Short-term volume is not worth losing demand partners who trust your exchange.

Talking to partners about it

IVT conversations go badly when they start with an accusation and a spreadsheet nobody else can check. They go better when both sides already agreed how traffic is measured and what happens when a problem is found. Share the evidence by source, with dates, so the supply partner can investigate its own publishers. Give them a chance to fix it before you cut them off, unless the problem is serious. And tell demand partners what you did about it: buyers remember exchanges that act, and they remember the ones that stay quiet.

Protecting your margin

An exchange that does not manage IVT ends up paying for it twice: once in credits to demand partners, and again in lower bids as buyers learn to discount its supply. An exchange that does manage it can say, credibly, that its traffic is clean, and that is worth more than any feature list.

You will never get invalid traffic to zero, and nobody honest will promise that. The aim is to make your exchange an expensive place to send it: filter the obvious, watch the patterns, measure what matters, and remove sources that keep failing.

Want to talk about your own setup? Get in touch with the Xendiz team.