AdTech
Malvertising and Bad Ads on Your Inventory: An Ad Quality Vendor, Your SSP's Controls, or Your Own Scanning
Where bad ads and malvertising enter the programmatic chain, what OpenRTB block lists and SSP or ad server controls can stop, and what scanning before serving and checking at render each miss. It also covers ad quality vendors, building your own scanner, the auction time limit, the cost of blocking good ads, Chrome and Google rules, publisher reports and the questions to ask.
By amBrain11 min read
In this article
Malvertising means ads that attack or cheat the people who see them: an ad that sends a phone to an app store without a tap, a fake virus warning that leads to a tech-support scam, or a fake software update. Ad quality is the wider job of keeping these ads, and offensive or very heavy ones, away from users. A supply-side platform (SSP), ad network, publisher or retail media network has three routes. It can pay an ad quality vendor that scans creatives or watches them run. It can use the controls already in its SSP or ad server. Or it can build its own pipeline that renders creatives, scores what they do and sends blocks back into the auction.
The short answer: start with what your SSP or ad server already has. Require an advertiser domain and a creative ID in every bid, block known bad domains and buyer seats, and review new creatives. These controls trust what the buyer declares, so they miss a creative that hides its behaviour. If users still report redirects or fake warnings, add a vendor that checks ads as they render on real devices. Build your own only if you run the SSP or ad network and need verdicts inside your own auction. On every route, keep your own registry of creatives and buyers.
What counts as a bad ad, and why is it hard to trace?
Some bad ads attack the user: a redirect without a tap, or a fake system warning, chat message or software update. Others are quality problems: offensive or misleading content, auto-playing video with sound, or creatives heavy enough to slow the page. Magnite, an ad platform for media owners, bans the first kind in its ad quality guidelines: ads “may not force a full-page redirect that takes users away from the seller's content without user initiation”, including “app-store redirects on mobile devices”.
Tracing is hard because the ad passes through several companies: the publisher's ad server, one or more SSPs or exchanges, and the demand-side platforms (DSPs) that bid for advertisers. The winning bid often carries code that loads more code from the buyer's servers each time it runs. A user's complaint arrives as a screenshot with no creative ID or buyer name.
Where do bad ads enter the chain?
The first way in is a demand partner or reseller, whose buyers and checks you cannot see. OpenRTB is the real-time bidding protocol of the IAB Tech Lab, an industry standards body. Its seat field names the buyer: the “ID of the buyer seat (e.g., advertiser, agency) on whose behalf this bid is made.” Seat IDs are agreed between each bidder and exchange in advance, so a seat ID has meaning only inside that connection. In June 2021 the IAB Tech Lab released buyers.json and the DemandChain object, which together let a seller see the parties involved in buying a creative. It said they help address malvertising. Ask your demand partners whether they support them.
The second is a creative that changes after approval or shows reviewers one thing and users another. Google's ads policy calls the second trick cloaking: “showing different content on your website to different people or to Google to try to hide things that might break Google Ads' rules.” Confiant, an ad security vendor, tracks cloaking kits: “The commercial services and self-hosted panels that decide whether a visitor sees the ad or the scam.” The scanner gets the clean version, and the user gets the scam.
What can your SSP or ad server already block?
OpenRTB lets the seller put block lists into each bid request: bcat for advertiser categories, badv, a “Block list of advertisers by their domains”, bapp for apps, bseat for buyer seats, and battr for creative attributes. AdCOM, a companion IAB Tech Lab specification, lists the attribute codes. Attribute 14, for example, is “Windows Dialog or Alert Style”. The bid answers with adomain, the “Advertiser domain for block list checking”, and crid, the “Creative ID to assist with ad quality checking”.
The buyer fills in those answers, so a fake virus warning will not declare attribute 14. Blocks also need enforcing. Prebid Server is open-source software for header bidding, in which several SSPs bid before the ad server picks an ad. Its blocking module “can reject bids from bidders that don't conform to the blocking lists sent in the request”, but “The default is to not block.” Magnite requires that “bid responses must include creative IDs, advertiser URLs or domains (adomain)”, and says buyers “may not use a single creative ID for multiple materially distinct ads”.
Google Ad Manager's Protections can block buyers and advertiser URLs. Its Ad review center lets you “review Ad Exchange ads that have served on your inventory”, so it acts after the first impressions, and “Ads are blocked by creative ID”. Google says a block can take up to 24 hours to work.
Controls also stop at platform borders. Microsoft says creative that reaches its Monetize platform from other platforms through Prebid Server Premium “is not registered in Microsoft Monetize the way creative from Monetize bidders is registered and audited, preventing most Monetize ad quality controls from being applied.” List which controls apply on each path your demand takes.
Scanning before serving or checking at render: what does each miss?
A scan before serving loads the creative and its landing page in a controlled browser, before or soon after the first impression. It catches broken code, oversized files, malware that shows itself and wrong click-throughs. GeoEdge, an ad quality vendor, says it can “Verify both creatives and landing pages to detect malware, cloaking, redirects, phishing, and policy violations.”
Cloaking is the blind spot of these scans. Confiant says attackers hide the harmful code from users behind proxies, virtual private networks (VPNs) or “Datacenter IPs as these are often categorized as scanners, bots, security researchers or users spoofing their real geolocation.” Vendors answer with many scan points. The Media Trust claims “24/7/365 client-side scanning from 100K+ device and user profiles in 600+ geos”, and sells “in-flight monitoring” because creatives change after launch.
A check at render runs in the visitor's browser. A vendor script wraps the ad or sits on the page and can stop a redirect or fake alert as it starts, cloaked or not. By then the auction is over, so the space stays empty unless the page refills it. The script also sees only the ads it reaches. HUMAN Security's module for Prebid.js, the open-source header bidding library, says that in one mode “Ads served outside of Prebid will not be protected”. GeoEdge's module says some video bids are “left untouched”. Prebid's pages disclose that HUMAN's and GeoEdge's modules load “external code that is not open source and has not been reviewed by Prebid.org.”
How much time does a check get inside the auction?
Very little. OpenRTB's tmax field is the “Maximum time in milliseconds the exchange allows for bids to be received including Internet latency to avoid timeout.” Rendering a creative and following its clicks takes far longer. So run scanning outside the auction, and let the auction only look up the verdict by creative ID, code hash, advertiser domain or seat. A lookup in memory costs almost nothing, while a call to an outside service on every bid adds a network round trip.
Checks at render cost page time instead. HUMAN's Prebid module has a mode with “no delay on bid rendering”, which leaves unwrapped “those bid responses, which arrived prior to successful preloading of agent script.” New creatives serve before any scan finishes unless you hold them. Google Ad Manager's manual creative review, in beta, can do that. Google warns: “Delay in reviewing creatives may impact auction performance and lead to buyer throttling.”
What does an ad quality vendor sell?
Vendors sell detection, blocking and the threat research behind both. In their own words, GeoEdge's protection for platforms “monitors your ad inventory 24/7 to detect and block any malicious activity and quality issues in real-time”. Confiant “connects ads, landing pages and the demand path to show where quality issues and security threats come from, and where you can act.” HUMAN's module page says its updates “continuously detect and block on-page malicious ad behaviors”.
The HUMAN and Confiant modules for Prebid need an agreement first: HUMAN's needs a “distribution key”, and Confiant's an “account and customer key”. Confiant and GeoEdge offer free trials, and The Media Trust asks you to request a demo. Ask each vendor for a written quote at your volume, and for detection and false positive rates measured on your own pages, not in a case study.
What does building your own scanning pipeline take?
Consider it if you run the SSP or ad network, so every creative already passes through your systems. A minimal pipeline:
- Stores each new creative from bid responses with its markup, creative ID, advertiser domain, seat and demand partner
- Loads it in a headless browser, driven by code without a screen, inside a throw-away container cut off from your network
- Follows the clicks and records the redirect chain to the landing page
- Flags redirects without a tap, new windows, downloads, fake dialogs, unknown scripts and heavy resource use, with Chrome's heavy ad limits as a first threshold
- Writes verdicts to block lists the auction checks on every bid, and scans live creatives again on a schedule
Cloaking is the hard part. Attackers recognise a scanner that runs in one cloud region. Seeing what real users see takes home and mobile networks, real device profiles, many countries and people who track new attack methods. That research is what vendors sell, and the part least worth rebuilding.
Building fits when ad quality is part of what you sell, when verdicts must live inside your auction and reports, or when vendor fees at your volume cost more than a team. One cheap safeguard works on any route: serve ads in iframes with the HTML sandbox attribute. According to MDN Web Docs, it can let a frame change the top page “only if initiated by a user gesture”. Test it first, because some legitimate creatives break inside a sandbox.
What happens after a bad ad is found?
Choose how wide to block. A creative ID block keeps the buyer's other spend, but new IDs evade it. An advertiser domain or buyer seat block fits a repeat offender, and cutting off a demand partner is the last step. Send blocks in the bid request as badv and bseat, and check every bid response against your own lists too.
Then give the demand partner enough to find the campaign: creative ID, seat, time, page, device, country, the ad code as delivered and the redirect chain with screenshots. Magnite says that once notified it “will initiate an investigation into the complaint within 24 hours”. Google Ad Manager's “Block and report” comes with a warning: “We may not review all of the ads that you report”.
What does blocking good ads cost?
Every false positive removes a bid that would have paid. Google's advice in Ad Manager is: “Use ad blocking sparingly. Allowing all ads creates the most competitive atmosphere in the ad auction, with the largest number of ads possible competing for an impression.” Microsoft warns that its blocks on outside demand “can impact publisher revenue”. Run each new rule in report-only mode first, log every block with its reason and the revenue it removed, and give buyers a way to appeal.
Which browser and Google rules already act on bad ads?
Chrome's documentation lists three limits for an ad the user has not interacted with: “The ad uses more than four megabytes of network bandwidth”, or it uses the main thread “for more than 15 seconds in any 30-second window” or “for more than 60 seconds total”. Past any of them, Chrome unloads the ad and shows a grey box reading “Ad removed”. The page that embeds the ad can receive a report.
Google's Abusive Experience Report in Search Console covers ads too: “We review ads and content, regardless of whether or not they're served from your domain.” Its list includes “Ads or content elements that auto-redirect the page without user action” and ads that “resemble chat apps, warnings, system dialogs, or other notifications”. If a site stays in failing status, “Chrome will prevent new windows and tabs from opening, and will remove all ads.”
Google's Ad Experience Report checks sites against the Coalition for Better Ads' standards, which cover the experiences “most likely to drive consumers to install ad blockers”, such as pop-up ads and auto-playing video with sound. Google's FAQ adds that when Chrome filters ads on a failing site, “it affects all visitors to the site, regardless of where they're visiting from.” Either way, the publisher pays for an ad that a buyer sent.
What should an ad quality report show the publisher?
For each blocked or flagged creative, the report should show:
- A screenshot or video, the landing domain and the redirect chain
- The advertiser domain, creative ID, demand partner and buyer seat
- The page or app, device, country and time, the reason and the revenue affected
- A way to match a user's complaint to the impression record
What should you ask an ad quality vendor or a firm that would build the scanner?
- Which threats do you catch before serving, at render, and only after a complaint?
- From which networks, devices and countries do you scan?
- Which demand paths do you cover: Prebid, Google Ad Manager, server-side auctions, in-app software development kits (SDKs), video and connected TV?
- What does your script add to page load time, and what fills the space when you block an ad?
- What is your false positive rate on traffic like ours, and how does a buyer appeal?
- Can we export every verdict with its evidence, and what do you charge at our volume and at double it?
- If you build it for us: who owns the code, the registry and the block lists, and which parts reuse your existing components?
Where does amBrain fit?
amBrain publishes this blog. It describes its AdTech work in four lines: “DSP development, real-time bidding platforms, and ad exchange engineering.” “amBrain builds supply-side platforms (SSP) for publishers.” “amBrain builds ad servers: targeting, frequency capping and reporting.” “amBrain builds event analytics pipelines for ad tech: collection, processing and reporting of impression and click events.” None of these lines names an ad quality or malvertising scanning product.
In its own words, “amBrain has been building software since 2019.” On how it works with clients, it says: “Three formats: full delivery, a dedicated team, or engineers embedded in your team.” Its line on ownership is “The client keeps full ownership of the product and the code, except our reusable components.” Ask for the list of those components by name before you sign.
This article is not a case study and describes no client work. Nothing in it says that amBrain has built an ad quality or malvertising scanning system for any client. The quotations come from published pages of the companies named above, Google, MDN Web Docs, the Coalition for Better Ads, the IAB Tech Lab and Prebid.org, read on 17 November 2026. The article claims no relationship between amBrain and any of them, recommends none of their products and gives no amBrain prices or timelines. If amBrain is on your shortlist, ask it the same questions as any other firm.
Topics
- Ad Quality
- Malvertising
- Build or Buy
- Ad Operations