What should a team do when an ad appears beside harmful content?
Part of Brand suitability
The short answer
When an ad appears beside harmful content, the team captures evidence before anything changes, contains the placement with the narrowest pause that works, and traces the cause through settings, deals and classification. It then tells the client what is known and what is not, fixes the rule that failed and records the incident so the same path cannot repeat.
Key takeaways
- Capture comes before containment, because pausing or excluding a placement can erase the evidence of how the ad got there.
- Decision rights should be set by scope before launch: a page, a domain, a line item and a whole campaign each have a named person who may pause them.
- Six root causes are worth checking first: settings changes, deal exemptions, exclusion scope, keyword rules, label age and supply paths that hide the URL.
- The first client message states facts and the next update time; causes come in the second message, after the trace.
What should a team do when an ad appears beside harmful content?
Definition
- Brand safety incident
- A served ad that appeared beside content the brand's policy places at its floor or above its tolerance, confirmed by evidence of the placement, the ad and the time.
The first hour decides whether the team will ever know what went wrong. A screenshot from a brand manager starts an investigation: the page may change, the ad may rotate out, and a hurried exclusion can hide the setting that let the ad through. Fixed phases keep the team from skipping to blame, or to a fix that treats the symptom while the brand suitability rule that failed stays in place.
| Phase | Goal | Done when | Lead |
|---|---|---|---|
| 1. Capture | Preserve what the reader saw and how the ad was served | Page, ad, time and delivery records are saved | Whoever confirms the report |
| 2. Contain | Stop further delivery on the content at the narrowest scope | No new impressions on the page, path or domain | The person holding pause rights at that scope |
| 3. Trace | Find the control that should have stopped the ad | One root cause is named and supported by records | Campaign owner |
| 4. Tell | Give the client the facts, then the cause | The client has the facts and a time for the next update | Account lead |
| 5. Fix and close | Change the rule so the same path cannot repeat | The fix is live, tested and logged | Campaign owner, with client sign off |
What evidence should be captured first?
Capture the evidence that disappears first: the page as it looks now and the ad sitting on it. Logs and settings last longer. The IAB and MRC Guidelines for the Conduct of Ad Verification (opens in a new tab) of February 2012 name the kinds of evidence that support a finding: "(1) screen copies, (2) tag and URL strings, (3) ad log files, (4) spider/crawler log files, and (5) discussion records for inquiries." They add that if materials are changed or altered, "such as placing an ad on an illustrative web site as evidence, this should be disclosed."
Incident evidence, in the order it disappears
- A dated screen copy of the full page with the ad visible, plus the original report and who sent it.
- The full URL string, including any path and parameters, and the ad creative that appeared.
- Campaign, line item, ad group and deal identifiers that served the impression.
- Delivery log rows for the URL or domain, with the exchange and time of each impression.
- A snapshot of the suitability settings, exclusions and keyword rules in force at that time.
- The change history for the account in the days before the incident, including system changes.
- All messages about the incident, kept as the discussion record.
Who decides whether to pause a campaign?
Pause authority should be written down before launch, because an incident is the worst time to discover that the only person who can stop a line item is on a plane. TAG's Brand Safety Certified Guidelines (opens in a new tab), version 5.0 of July 2026, require certified companies to define the internal roles responsible for responding to incidents of ad misplacement, document the external resources they work with, and demonstrate "clear lines of communication across partners." The guidelines leave the job titles to each company. The table below is our own template for splitting the decision by scope.
| Scope of the pause | Who may act | When | Who must be told |
|---|---|---|---|
| One page URL | Anyone on the monitoring rota | At once, for severity 1 or 2 | Campaign owner, the same day |
| A path, domain or channel | Campaign owner | At once for severity 1; after review otherwise | Account lead and client contact |
| A line item or deal | Campaign owner | When the cause may affect more than one site | Client contact, within hours |
| The whole campaign | Client brand lead, on the agency's advice | When the cause is unknown and delivery continues | Agency leadership and client leadership |
| Restart after severity 1 | Client brand lead | After the fix is tested | Everyone on the incident thread |
Narrow pauses protect the rest of the plan. Google's help page on excluding webpages and videos (opens in a new tab) shows how scope follows the URL form: a page URL excludes only that page, a directory path excludes that directory, and a top level domain excludes the domain with its subdomains. Exclusions can be set at the campaign, ad group or account level.
How do you find the root cause?
Start from the controls, not the classifier. An ad reaches harmful content because a control was missing, mis-scoped, switched off or blind to that page, and each of those leaves a different trace. Display & Video 360's event history (opens in a new tab) records the time, date, description and user for each change, and its option to include system changes shows edits made by the platform rather than a person, which helps separate a human change from a platform action.
| Root cause | Sign in the evidence | Fix |
|---|---|---|
| Settings changed | An event history entry shortly before the incident, by a user or the system | Restore the setting and add a review step for that change |
| Deal exempted from suitability settings | Display & Video 360 lets a deal "bypass all advertiser-level brand suitability settings" (brand suitability guide (opens in a new tab)) | Remove the exemption or require client sign off for it |
| Placement removed, not excluded | The URL was dropped from targeting but other targeting still reached it | Add a true exclusion at the right level |
| Exclusion at the wrong scope | A subdomain or path was excluded while the ad ran elsewhere on the domain | Exclude at the scope the content needs |
| Keyword rule missed the page | Google says excluded content keywords "only support exact matches" (content suitability (opens in a new tab)) | Add the variants, or rely on category controls for that content |
| Label older than the content | The page changed after its last classification | Exclude the page and shorten the review interval for that section |
Two checks keep the trace honest. First, confirm the incident is real: the 2012 guidelines require human review for material errors, partly to catch false positives, where a non-compliance is flagged that does not exist. Second, check whether the fix creates a new problem. Google warns that "some words have multiple meanings," so a keyword added in haste can block responsible pages.
How should the incident be explained to the client?
The client hears twice. The first message goes out fast and holds only facts: what was seen, when, what has been paused and when the next update will come. The second follows the trace and explains the cause and the fix. Mixing them invites a guess that later turns out wrong, and the guess is what the client remembers.
Five parts of the client explanation
- State what happened in plain terms. Name the page, the ad, the time window and how the incident was found, with the evidence attached.
- Say what was contained. Describe the pause or exclusion, its scope and when it took effect.
- Separate known from unknown. List what the records confirm and what is still being checked, with the date of the next update.
- Name the cause once it is supported. Point to the record that shows it, such as a change history entry or a deal setting.
- Describe the fix and the test. Explain what changed, how the team confirmed it works and what will be watched in the coming weeks.
Some incidents belong outside the client conversation too. The 2012 guidelines encourage "closing the communication cycle with seller organizations" so the publisher can correct the problem. When a partner certified by TAG may have breached its certification, TAG's complaints page (opens in a new tab) asks for "written documentation demonstrating the relevant incident(s) of non-compliance", which the capture phase has already produced.
How does PlacementGraph help trace a placement back to the plan?
What does an incident response look like on a real campaign?
Fictional example
A candy maker's Halloween campaign in North and South Dakota
An unnamed candy maker runs a Halloween display campaign in Fargo, Bismarck and Sioux Falls on an approved list of family, recipe and local news sites. On a Saturday evening its brand manager forwards a screenshot: a candy ad beside a page hosting graphic video of a fatal crash, on a site nobody recognizes.
The agency's on call planner confirms the impression in the delivery logs, saves the page, URL, creative and line item, and excludes the domain, which the policy places at the floor. The first message to the client goes out within the hour with those facts and a promise of an update Monday morning.
The trace finds no classifier failure. Event history shows that a line item was copied the week before without the inclusion list, so it bought on the open exchange. The Monday message names that cause, shows the history entry and describes the fix: copied line items now need a second reviewer before they go live. The incident, its evidence and the fix are logged, and the site's publisher is not contacted, since the site was never meant to be on the plan.
Frequently asked questions
What if the only evidence is a screenshot from social media?
Treat it as a report, not a finding. Check the delivery logs for the domain and time shown, and look for the creative in the campaign. If the logs show no impression there, say so to the client with the records attached, since a screenshot can be cropped, old or from another advertiser.
How long should incident evidence be kept?
For verification services, the 2012 IAB and MRC guidelines say evidence tied to material errors should be retained for at least twelve months after results are reported, and records tied to a known dispute until it is resolved. A brand can set its own period for its incident files in its policy.
Should a team pause first and capture later when content is at the floor?
Capture the page and the ad in the same minute, then pause. A dated screenshot and the URL take seconds and survive the pause; the delivery records can be pulled afterward. Pausing first and capturing nothing leaves the trace with no starting point.
Does an incident mean the suitability settings failed?
Not necessarily. Google says it can't guarantee that all related content will be excluded, and some incidents trace to a copied line item or a deal exemption rather than the settings. The root cause table separates a control that failed from a control that was never applied.
Sources
- Guidelines for the Conduct of Ad Verification (opens in a new tab)IAB and Media Rating Council, February 14, 2012
- Brand Safety Certified Guidelines, version 5.0 (opens in a new tab)TAG (Trustworthy Accountability Group), July 2026
- Complaints (opens in a new tab)TAG (Trustworthy Accountability Group)
- Event history (opens in a new tab)Google Display & Video 360 Help
- Brand suitability (opens in a new tab)Google Display & Video 360 Help
- Exclude specific webpages and videos (opens in a new tab)Google Ads Help
- About content suitability (opens in a new tab)Google Ads Help
- Digital content labels in Display & Video 360 (opens in a new tab)Google Display & Video 360 Help
Researched and written by PlacementGraph Research. Read how we research, source and update these pages. Corrections: diana@placementgraph.com.