What are the main types of ad fraud, and how is each one caught?
Part of MFA avoidance
The short answer
The main types of ad fraud are bot traffic, ad stacking, pixel stuffing, domain spoofing, click injection, ad injection and click farms. Each one fakes a different part of an ad transaction: the visitor, the ad's visibility, the website's identity or the credit for an action. Each needs its own detection method, from viewability geometry to seller files.
Key takeaways
- Each fraud scheme in this guide fakes one link in the chain: who saw the ad, whether it could be seen, where it ran, or who earned the credit.
- Bots range from simple crawlers in data centers to hidden browsers on infected home computers that look like ordinary visitors.
- Ad stacking and pixel stuffing produce impressions no one can see, so they surface through viewability geometry and ad size checks.
- ads.txt and sellers.json target counterfeit inventory, where a seller misrepresents the site or app; they do nothing against bots or hidden ads on an authorized site.
- No single control catches every type, so buyers layer pre-bid filters, seller file checks, measurement and log reconciliation.
What are the main types of ad fraud?
Ad fraud is any scheme that makes an advertiser pay for advertising that did not happen as sold. Measurement bodies group the activity under invalid traffic, which our invalid traffic glossary defines. This page looks at the schemes themselves: how each one works, what it fakes, and which control is designed to catch it.
| Scheme | What is faked | How it works | How it is detected |
|---|---|---|---|
| Bot traffic | The visitor | Software loads pages and ads, from servers or from infected home devices | Industry bot lists, data center IP filtering, behavior and device analysis |
| Ad stacking | Visibility | Several ads are layered in one slot so only the top one can be seen, but all are counted | Viewability geometry, ads reporting the same position at the same time |
| Pixel stuffing | Visibility | A full size ad, or a whole page of ads, is squeezed into a frame of one or a few pixels | Rendered size checks, 0x0 and 1x1 placements, measured ad dimensions |
| Domain spoofing | The website | A bid request claims a premium site or app while the ad actually runs somewhere else | ads.txt and app-ads.txt authorization, sellers.json identity, page URL checks |
| Click injection | Credit for an install | A malicious app reports a click just before another app finishes installing, to take the attribution | Comparing the click time with the time the install began |
| Ad injection | The publisher's ad slots | Adware on a device inserts or overlays ads on pages without the site's permission | Ads on the page that the publisher's ad server never sent, malware signals |
| Click farms | Human interest | Paid or directed people click and browse to inflate activity | Pattern analysis across many users and devices |
Several of these schemes appear by name in the Media Rating Council's IVT 2.0 standards (opens in a new tab), which list Z-order stacking, banner stuffing, domain laundering, app ID spoofing, hijacked devices, adware ad injection and cookie stuffing among the activities measurers must handle. Google's help page on managing invalid traffic (opens in a new tab) names clickjacking and ad stacking as examples of fraudulent ad placements.
How do bots imitate real visitors?
The simplest bots announce themselves, run from servers in data centers and are removed by lists of known crawlers and hosting IP addresses. The harder cases hide. A joint DHS and FBI technical alert published by CISA on November 27, 2018 (opens in a new tab) described an operation called 3ve that created fake versions of both websites and visitors and funneled the advertising revenue to criminals. It controlled over 1.7 million unique IP addresses when sampled over a 10 day window.
The alert describes two techniques that show how bots blend in. In one, machines in a data center used a botnet of infected home computers as a proxy to request ads, so the traffic appeared to come from residential connections. In the other, malware ran a hidden browser on the infected machine that the user could not see, loading pages and ads in the background on a real person's computer and internet connection.
The layers bots use to look human, and the check each one invites
- Borrow a real address. Traffic routed through infected or hijacked home devices defeats simple data center lists, so detection moves to device and session analysis.
- Present a normal browser. A complete user agent and a working rendering engine defeat checks for headless or non-browser clients, so measurers look at whether behavior matches the declared device.
- Behave like a reader. Scrolling, pauses and mouse movement can be simulated, so analysis looks for patterns repeated across thousands of sessions.
- Spread the volume. Many devices each send a little traffic, so the signal appears only when activity is corroborated across sites, campaigns and time.
How does ad stacking work?
In ad stacking, several ads are placed on top of each other in a single slot. The visitor sees only the top ad, but each ad in the stack fires its impression and the seller bills for all of them. The MRC lists Z-order stacking, named after the browser's front to back drawing order, among hidden or obscured ad serving that measurers must detect.
The MRC's viewable ad impression guidelines (opens in a new tab) count a display ad as viewable when at least 50 percent of its pixels are in view on an in-focus browser tab for at least one continuous second, and a video ad when the same pixel share holds for two continuous seconds. The same guidelines ask measurers to differentiate impressions served out of focus or under obstruction. A stacked ad underneath another fails that test, so stacking can show up as a cluster of placements with many impressions, near zero viewability and identical screen positions.
What is pixel stuffing?
Definition
- Pixel stuffing
- Serving an ad, or an entire page of ads, inside a frame so small that no person can see it, such as one pixel by one pixel, while the impression is still counted and billed.
Pixel stuffing differs from stacking in where the ad goes: instead of hiding under another ad, it shrinks out of sight. The MRC treats the simple form as general invalid traffic, listing invisible ad delivery and illogical ad sizes of 0x0 and 1x1 as delivered on the client side. It lists banner stuffing among sophisticated cases, where the frame is disguised. The standards also caution that the size declared in a bid request may not match the size actually rendered on the page.
Signs of pixel stuffing or stacking in a placement report
- A placement reports many impressions but almost none measured as viewable.
- The rendered ad size differs from the size declared in the bid request.
- Several ads report the same position on the page at the same moment.
- Impressions keep accruing at hours when the site's normal traffic is low.
- Clicks are close to zero across every creative served to the placement.
Which fraud types do ads.txt and sellers.json help prevent?
The ads.txt 1.1 specification (opens in a new tab) is explicit about its target: inventory offered to buyers with a misrepresented label and account during real time bidding, where typically the domain of the webpage or the ID of the mobile app has been falsified. The sellers.json specification (opens in a new tab) adds identity, allowing the final seller of a bid request to be discovered, assuming that seller is ads.txt authorized. Both are part of an effort to stop profit from counterfeit inventory. Our guides to ads.txt and sellers.json cover how to read the files.
| Scheme | Effect of ads.txt and sellers.json | What else is needed |
|---|---|---|
| Domain spoofing on the web | Strong: an unauthorized seller account claiming the domain can be refused | A check that the page URL in the request belongs to the claimed domain |
| App ID spoofing | Strong when the app publishes app-ads.txt, the companion specification | Store listing checks for the developer's site |
| Unknown intermediaries | Partial: sellers.json names the seller behind an account ID | The SupplyChain object in the bid request for the full path |
| Bots, stacking and pixel stuffing | None: they can run on a site that authorizes every seller correctly | Invalid traffic filtering, viewability and ad size measurement |
| Click injection and ad injection | None: the fraud happens on the device, not in the selling path | Attribution timestamps and malware detection |
What is click injection, and how is it detected?
Click injection is attribution fraud on Android phones. When an app advertiser pays partners for installs and credits the last click before each install, a malicious app already on the device can watch for a new install and report a fake click just before it completes. The fraudster then collects the payment for an install the advertiser's own campaign or an organic search produced.
The defense is timing. The Google Play Install Referrer API (opens in a new tab) returns the referrer for an install along with two timestamps, one for the referrer click and one for when the install began. When Google introduced the API on the Android Developers Blog (opens in a new tab) in November 2017, it called accurate install referral data vital for correctly attributing app installs, and an attribution partner quoted in the post said the data lets it detect and prevent click injection. A click recorded after the install began cannot have caused it.
What does a fraud review look like on a real campaign?
Fictional example
A lawn mower dealer in Alabama and Arkansas
A fictional dealer group selling riding mowers, string trimmers and leaf blowers runs spring display and online video in Alabama and Arkansas, beside lawn care, home and garden and rural living content. Its agency reviews delivery after the first two weeks.
Three findings come back. A placement on a gardening site shows heavy impression counts with almost no viewable impressions and identical screen coordinates for several ads, so the desk files it as suspected stacking and asks the seller for an explanation. A batch of bid requests claimed a well known home magazine, but the seller account was not in the magazine's ads.txt file, so the buying platform's authorized seller setting had already refused them. Finally, the verification partner reports a rise in sophisticated invalid traffic from residential connections on two sites, which it flags for further analysis.
The desk excludes the stacking placement pending an answer, keeps the authorized seller setting on, and holds the two flagged sites at reduced spend until the partner's analysis is complete. Each decision is logged with its evidence.
Which ad fraud risks can a planning check reduce?
Which controls should a buyer layer against ad fraud?
A layered ad fraud control set for one campaign
- Pre-bid filtering is switched on in the buying platform for invalid traffic and known bad sources.
- Authorized seller settings require ads.txt or app-ads.txt authorization for every web and app placement.
- A measurement partner reports general and sophisticated invalid traffic separately, with its decision rate.
- Viewability and rendered ad size are reviewed by placement, not only as a campaign average.
- App install campaigns compare click and install timestamps before paying for attribution.
- Served domains and apps are reconciled against the approved inclusion list after each flight.
- Contract terms say how late fraud findings are credited or made good.
Frequently asked questions
What is ad injection?
Ad injection is when adware or malware on a visitor's device inserts ads into web pages, or overlays ads on top of the publisher's own, without the site's permission. The publisher is not paid for those ads and may not know they appear. The MRC lists adware and malware that conduct ad injection and unauthorized overlays among sophisticated invalid traffic.
What is cookie stuffing?
Cookie stuffing plants tracking cookies on a device without a real interaction, so that a later purchase or sign up is attributed to a fraudster's link. The MRC's standards list cookie stuffing, recycling and harvesting, which manipulate or falsify a user's prior activity, as sophisticated invalid traffic. It harms affiliate and performance campaigns.
Can ad fraud happen inside a private marketplace deal?
Yes. A deal limits who may bid, not what happens on the device. A site in a private marketplace can still receive bot traffic or serve ads in hidden slots, and a deal can still be offered for spoofed inventory if seller authorization is not checked. Keep invalid traffic measurement and viewability reporting on deal buys too.
Are accidental clicks a type of ad fraud?
They are treated as invalid activity even when nobody intended them. Google lists accidental clicks, for example from poor ad placement, among invalid traffic, and the MRC lists tricking users into accidental clicks under manipulated activity. The difference is intent: a cramped mobile layout is a quality problem, while ads placed under buttons by design is manipulation.
Sources
- Invalid Traffic Detection and Filtration Standards Addendum (IVT 2.0), Update (Final) (opens in a new tab)Media Rating Council, June 2020
- 3ve: Major Online Ad Fraud Operation (Alert TA18-331A) (opens in a new tab)CISA, DHS and FBI, November 27, 2018
- MRC Viewable Ad Impression Measurement Guidelines, Version 2.0 (opens in a new tab)Media Rating Council, August 2015
- ads.txt Specification Version 1.1 (opens in a new tab)IAB Tech Lab
- sellers.json specification (opens in a new tab)IAB Tech Lab
- Managing invalid traffic (opens in a new tab)Google Ads Help
- Play Install Referrer Library (opens in a new tab)Android Developers
- Google Play Referrer API: Track and measure your app installs easily and securely (opens in a new tab)Android Developers Blog, November 20, 2017
Researched and written by PlacementGraph Research. Read how we research, source and update these pages. Corrections: diana@placementgraph.com.