- Published
- Oct 28, 2026
- Updated
- Oct 30, 2026
- Reading time
- 14 min read
Malicious Ad Redirects: A Publisher Investigation Checklist
Turn reports of suspicious ads and automatic redirects into useful evidence. Identify the serving path, report the creative, and verify what changed.

In this guide (12 sections)
If a visitor reports that your website unexpectedly opened another page, displayed a fake security warning, or started an unwanted download, preserve the observation, identify the relevant ad delivery, and escalate through the serving provider's reporting process. A screenshot alone rarely identifies the responsible creative. A clean test on your own laptop rarely proves the problem has ended.
Start with the exact page, timestamp and timezone, device/browser, observed behavior, and any available ad-request identifier. Do not ask visitors to click suspicious ads, open downloaded files, or bypass browser warnings to produce more evidence. Your first job is to make the report actionable while reducing further exposure.
This guide is for publishers and ad operations teams investigating an observed incident. It separates a visitor's experience from the evidence needed to attribute that experience, choose a containment action, and assess whether it has stopped recurring.
Why “is the bad-ad problem fixed?” needs a better answer#
A real community question about suspicious ads asked whether an earlier problem had been resolved. The post was created on October 21, 2026, UTC. A response carrying a Community Manager flair said the particular issue had been addressed and linked the publisher's reporting process. Another reader asked about the resolution on October 25.
That thread establishes a real reader question, not an independent security assessment. It does not establish the cause of every reported redirect or whether a separate visitor's device was compromised. The useful operational lesson is that a resolution statement needs a scope: which incident, which delivery path, what action, and what subsequent observations?
The publisher's own bad-ad reporting instructions request time, page URL, location, behavior, destination URL and screenshot. Those details are useful starting evidence for any publisher designing its own intake process.
Preserve the redirect report before testing
A screenshot rarely identifies the responsible creative, and a clean local test cannot rule it out. Discuss the incident evidence available.
Discuss delivery measurementA short inquiry to scope relevant options; no campaign access is needed.
Separate the symptom from the diagnosis#
Write down what happened before deciding what caused it.
| Observation | What it establishes | What still needs investigation |
|---|---|---|
| The top-level page navigated without an intended click | An unexpected navigation was observed | Which code or interaction initiated it |
| An ad displayed a fake warning | The visible creative contained that warning | Serving source, creative identity and relevant policy violation |
| A file download started | A download was initiated | Its source and whether it was malicious; do not execute it to find out |
| A browser security interstitial appeared | The browser blocked or warned about a destination | The path that reached it and the underlying cause |
| The issue stopped with ads blocked | Behavior changed under that test condition | Which ad path was involved; whether other conditions changed |
| Only one tester sees the problem | Reproduction is conditional or incomplete | Differences in device, geography, account, extensions and delivery |
An unsuitable advertisement and a malicious advertisement require different investigations. Content adjacency concerns belong in a buyer's brand-safety and suitability workflow. An automatic redirect concerns creative behavior and site integrity. Invalid-traffic analysis answers another question: whether measured activity meets a counting policy. One result cannot substitute for the others.
Google's cloaked-ad guidance describes creatives that appear benign to its systems but change when delivered. The visible destination may consequently fail to identify the ad in its review tool.
1. Create a usable incident record#
Give the report an incident ID and one responsible owner. Keep the original visitor account separate from your interpretation.
Use these intake fields:
- Exact publisher page URL and the affected slot or screen location, if identifiable.
- Date, clock time and timezone, including whether the time is approximate.
- Browser name/version, operating system and device type.
- Country or broad region, without asking for a home address or precise location.
- Whether the visitor was signed in, and whether the behavior followed a click, scroll, page load or ad refresh.
- Screenshot or recording, original warning text, and already-observed destination URL.
- Relevant extensions, VPN, proxy, DNS filtering or other tools, recorded as context rather than presumed cause.
- Whether the same behavior was seen elsewhere, and which checks have actually been completed.
Do not make a full forensic package a condition for acknowledging a report. A visitor may have closed the tab immediately. Mark unavailable fields as missing and explain the next useful item. Never replace a missing creative ID with a guessed advertiser name based on the image.
Keep original evidence in a restricted location. Make a separate redacted copy for wider discussion. Screenshots and URLs can contain account details, identifiers or private page content; public forum posts are usually a poor place for the raw incident bundle.
2. Preserve the ad before reloading the page#
Reloading requests another auction or delivery opportunity. The replacement creative may differ, so the new page is a new observation.
For Google's cloaked-ad report, preserve an original PNG in page context, without cropping, annotation, resizing or lossy recompression. Keep it separate from any marked-up copy.
For a webpage using Google Publisher Tags, the Google Publisher Console can expose slot details and a Query ID. Use the documented bookmark or console activation approach on an already loaded page. Google's cloaked-ad instructions warn against URL-based activation that reloads and loses the relevant observation. Query IDs are an alternative for supported web display observations; do not assume the same workflow applies to mobile apps.
Record each identifier with its namespace: Query ID, creative ID, line-item ID or another provider's request ID. They are not interchangeable. Absence of a line-item or creative ID in a particular tool is not proof that no ad served.
If the tab has already closed, report what remains. Do not send a visitor back to a suspicious destination to recover the missing identifier.
3. Identify the serving route before choosing a control#
Map the affected slot's actual route. Was it a directly managed creative, an exchange transaction, a header-bidding partner, a mediation path, or another third-party script? A publisher can use several on the same page.
Create a small evidence table for the specific observation:
| Field | Record | Avoid concluding |
|---|---|---|
| Slot and page | The exact affected placement | Every slot on the site is affected |
| Serving path | Provider and transaction identifiers available | A recognizable logo proves which buyer supplied it |
| Creative evidence | Original image, ID and observed behavior | Every visual variation shares that behavior |
| Destination | URL already observed, stored privately | A matching domain alone identifies the entire chain |
| Confidence | Confirmed, suspected, or not yet linked | Suspicion is a verified root cause |
Supply authorization files can help with a separate review of who is authorized to sell inventory. They do not certify a creative's behavior. Use the ads.txt, sellers.json and SupplyChain guide for that task; do not treat a valid file as a clean bill of health for every delivered ad.
Mobile apps: separate creative behavior from SDK behavior#
For an app incident, do not substitute the web Publisher Console workflow for mobile evidence. A real developer question about a suspended app links the timing to a mediation change. The original poster later withdrew an initial code-search accusation, then reported reinstatement after conditional appeal acceptance and a compliant update. They also said Google never identified the exact trigger. Their mediation hypothesis and individual resolution establish neither SDK causation nor a general reinstatement promise.
Preserve the exact notice, affected package/version code, review date, screenshots and case references privately. Then record the released app build, resolved SDK and mediation-adapter versions, relevant dependency changes, and the remote mediation configuration in effect. “We removed a network” can describe several different changes; distinguish a configuration edit, a dependency change, a submitted build and the build actually installed in the test.
Use a mobile evidence worksheet:
| Observation | Next evidence | Conclusion to avoid |
|---|---|---|
| A specific ad opens an unexpected destination | Available ad/response identifiers, screen, format and original destination | Every ad from that source behaves this way |
| The notice identifies a library or behavior | Exact notice and affected artifact, reviewed with the SDK provider | A similar class name proves executable downloading |
| The symptom appeared after mediation changed | Before/after build and configuration inventory | The last added SDK is therefore the cause |
| A corrected test build behaves differently | Controlled conditions and the precise changes made | Every existing installation is repaired |
Google's SDK safety guidance places responsibility for included third-party code with the developer, points to SDK notices and version information, and recommends contacting the SDK provider about compliance questions. This supports a version-specific investigation rather than blaming a supplier from a broad text search.
Send the relevant provider a narrowly scoped request: identify the affected artifact or serving object, explain the observed behavior, and ask which supported correction addresses it. Keep app-code findings and creative-delivery findings separate until evidence joins them. Google's enforcement process distinguishes the action and appeal path. A technical change or a provider's assurance is not an approval or reinstatement receipt; follow the actual notice and preserve the decision separately.
4. Send the provider a report it can act on#
Follow the provider's current process rather than assuming one universal reporting form exists.
For an eligible Ad Exchange creative, Google's Ad Review Center supports locating ads and reviewing metadata. Its review instructions distinguish marking an ad reviewed from blocking it and reporting a suspected policy violation. “Reviewed” allows continued serving; it is not a containment action.
For suspected cloaking, use the specific reporting route with the original image and requested click string or supported Query ID. Follow its documented collection process without asking visitors to trigger suspicious behavior.
For a non-Google route, send the evidence to that provider's established support or incident channel. Ask for acknowledgement, a case identifier, the identified serving object, the scope of action, and the effective time. A message saying “we reported it” should remain a report status until a separate containment action is confirmed.
5. Use network traces only when they answer a specific gap#
Do not begin every incident by collecting a full HTTP archive. Google's cloaked-ad process recommends an HTTP log only when explicitly requested after the preferred methods do not apply.
If your technical investigation or provider requires a trace, assign it to an authorized technical operator in an appropriately controlled environment. Decide what question it should resolve: which request preceded navigation, which slot made a request, or whether a particular response belonged to the observation.
Chrome's Network documentation explains that logging captures requests while DevTools is open, that Preserve log retains requests across page loads, and that sanitized HAR export omits certain sensitive headers. Sanitized does not mean anonymous: inspect remaining URLs, parameters, payloads and content before sharing a file through an approved private channel.
A network trace is evidence to interpret, not an automatic malware verdict. Timing alone does not establish that the last visible request caused the navigation. Do not replay questionable requests or execute captured code merely to complete a report.
6. Contain precisely, then measure recurrence#
Choose the narrowest action justified by the evidence and the incident's severity. A verified problematic creative may support a creative-level block. Uncertain attribution may justify a temporary restriction on the affected path while specialists investigate. Your security and commercial owners should decide the scope and record the tradeoff.
Do not assume a buyer's pre-bid suitability setting blocks malicious creative behavior on the publisher's page. Pre-bid, post-bid and blocking controls operate at different points with different capabilities.
Record the control, identifier, affected products/sites, operator, time applied, expected propagation and rollback criteria. Google notes that an Ad Review Center block can take up to 24 hours to become effective. The center can also retain creatives with historical impressions. Seeing an old card there is not proof of fresh serving; inspect the actual observation time.
Check whether a recommended category control still exists
Google's current AdMob category reference and Ad Manager protection reference specify 23 October 2026 as the removal date for Downloadable Utilities, Consumer Loans, and Astrology & Esoteric. They state that after that date publishers cannot broadly block or filter creatives using those categories, or view them at creative level under those categories in Ad Review Center. The pages retain future-tense wording; this is their documented effective date, not an observation of your account's rollout.
Do not copy an older “block Downloadable Utilities” remedy into a current incident checklist. Check the actual available controls and provider reporting route for the identified delivery. Ad Manager also limits its sensitive-category protections to specified transaction types; a category setting is not a universal control over every mediated SDK or app behavior.
Retiring a category does not prove that it caused an incident, that all creative-review controls disappeared, or that a policy enforcement action will be reversed. Preserve the incident evidence and evaluate the supported action for its actual scope.
After containment, track three separate things: new visitor reports, confirmed new reproductions, and the volume of comparable observed delivery. A quieter support inbox alone cannot tell you whether exposure fell or reporting fell.
Hypothetical example: twelve reports do not mean twelve percent of visitors
Suppose a publisher receives 12 reports during one hour. Deduplication finds eight separate visitor observations; three lack a timezone and one includes a usable request identifier.
The supported statement is: “Eight distinct observations were reported; one is linked to a delivery identifier.” You cannot calculate an incident exposure rate from these reports because you do not know how many affected visitors reported, or how many comparable deliveries could have produced the behavior.
After a targeted restriction, an operator completes 40 controlled page-load checks in the affected device/region configuration and sees no recurrence. Report 0 observations in 40 checks under those conditions. Do not report “zero malicious ads across the site.” The checks do not cover every auction, device, visitor or delivery path.
If monitoring continues, preserve the same test conditions and counting unit. Forty page loads are not forty unique people, and visitor complaints are not impression counts. Record any configuration changes before comparing periods.
A practical closure checklist#
- The original observation and its uncertainty are preserved.
- The provider has a case ID and enough evidence to investigate.
- The serving route is confirmed, or the attribution gap remains explicit.
- The containment scope and effective time are recorded.
- Follow-up checks cover the affected conditions and use explicit units.
- Remaining unknowns have owners and next actions.
- The public update describes the specific incident, avoids blaming an unverified party, and says how visitors can report recurrence.
Close the operational incident according to your team's criteria while retaining a monitoring plan. “Addressed this identified creative and observed no recurrence in the stated checks” is more useful than an unlimited promise that every ad is safe.
Frequently asked questions#
Does an automatic redirect prove that the publisher or browser was hacked?
No. It establishes an observed navigation, not its cause. Preserve the page, time, environment and available delivery identifiers, then investigate the ad path and other possible sources. A screenshot or clean scan alone cannot settle attribution.
Why can I not find the bad ad using its landing-page URL?
An ad system may have recorded a different destination, particularly with cloaking. Request identifiers and provider investigation can help resolve that gap. An unsuccessful URL search does not disprove the report.
Should I collect a HAR file from every visitor?
No. Begin with the provider's requested evidence and investigate specific missing information. If a trace is needed, review it for sensitive data before sharing through an approved private channel.
How do I know whether a block worked?
Verify its scope and effective time, then review new delivery observations under the affected conditions. Distinguish historical review records from newly served ads. State the test population and remaining coverage limits rather than claiming a site-wide guarantee.
Keep campaign measurement claims within their evidence#
An ad-quality incident may also prompt buyers to review campaign delivery, measurement coverage and partner reporting. Keep that reconciliation separate from security attribution. The anonymized streaming-media delivery study illustrates a different measurement scope; it is not evidence of malicious-creative detection or incident resolution.
Discuss a scoped measurement review with AdProof, bringing the campaign window, source reports, counting definitions and available evidence. Confirm the supported measurement scope for your campaign; do not substitute a delivery metric for a malicious-creative investigation.
Preserve the redirect report before testing
A screenshot rarely identifies the responsible creative, and a clean local test cannot rule it out. Discuss the incident evidence available.
Discuss delivery measurementA short inquiry to scope relevant options; no campaign access is needed.
