How to Ensure Brand Safety in Digital Ad Campaigns

To ensure brand safety in a digital ad campaign, define a brand-specific suitability policy, apply pre-bid and platform controls, verify where measurable ads appeared, separate unsafe from unknown inventory, and maintain an incident process. No blocklist, keyword list, platform promise, or verification tag can guarantee perfect protection across every site, app, social feed, and CTV stream.

Brand safety and brand suitability are different#

Brand safety aims to prevent advertising next to content that is broadly harmful or unacceptable for monetization, such as severe violence, explicit sexual content, or terrorism-related material.

Brand suitability applies the advertiser's own risk tolerance beyond that floor. A news report about alcohol may be suitable for one brand and unsuitable for another. A family brand may avoid moderate profanity, while an entertainment brand accepts it. A financial advertiser may restrict speculative-trading content even when the content is lawful and professionally produced.

This distinction matters because "brand safe" is not a complete campaign rule. Your team still needs to decide:

• which topics and risk levels are unacceptable;
• whether the rule changes by product, market, language, or creative;
• whether news, live streams, user-generated content, games, or mature themes are allowed;
• how unknown or unclassified inventory is handled;
• which exceptions require human approval; and
• how much reach the brand will trade for tighter controls.

The IAB Tech Lab Content Taxonomy 3.1 provides a common language for describing content. IAB Tech Lab lists contextual targeting and brand safety as typical uses. A taxonomy helps publishers, platforms, buyers, and verification providers exchange category labels, but each advertiser still has to map those labels to its policy.

Write a measurable brand-suitability policy#

A useful policy is detailed enough to configure and audit. It should contain six parts.

1. The safety floor

List content that is never acceptable for the campaign. Define the categories and severity thresholds, plus any legal or contractual restrictions. Avoid vague language such as "bad content" or "negative news."

2. Brand-specific suitability rules

Describe categories that require different treatment by context. For example:

• allowed in factual news but not graphic footage;
• allowed at low risk, excluded at medium or high risk;
• allowed for one product line but excluded for another;
• allowed when the article discusses prevention or education;
• allowed only on an approved publisher list; or
• sent to manual review when the classifier has low confidence.

3. Channel rules

Web pages, mobile apps, social feeds, online video, games, podcasts, and CTV do not expose identical context. State which evidence and controls are required for each channel.

4. Unknown-inventory treatment

Decide whether unclassified inventory is blocked, allowed, capped, or routed to a review queue. Unknown does not mean safe, but it does not automatically mean unsafe either. Reporting it as a separate bucket exposes the risk decision.

5. Accountability

Name the owner who approves the policy, the media operator who configures it, the measurement owner who checks it, and the communications or legal contact for incidents.

6. Evidence and escalation thresholds

Define what counts as an exposure, how much evidence is required, the time allowed to investigate, and what triggers a placement block, publisher pause, campaign pause, or executive escalation.

Once approved, convert the policy into a campaign control sheet. Keep the version and effective date with the media plan.

The four-layer brand-safety workflow#

Brand safety is a control system, not a single vendor toggle. Use prevention, supply validation, in-flight measurement, and post-campaign audit together.

Layer 1: choose inventory and apply pre-bid controls

Start with the buying platform's inventory settings. Depending on the channel, these can include:

• inventory type or sensitivity level;
• sensitive-topic and content-category exclusions;
• placement, channel, domain, app, and publisher exclusions;
• allowlists for tightly controlled campaigns;
• excluded content keywords or themes;
• age rating or digital content labels where supported;
• live-stream, embedded-video, app, or below-the-fold exclusions;
• language and market restrictions; and
• third-party pre-bid contextual or suitability segments.

Google's current content suitability guidance describes Maximum, Moderate, and Limited inventory, plus sensitive-content, placement, theme, and keyword controls. It also states important limitations: settings vary by inventory type and campaign, and content exclusions cannot guarantee that every related item will be excluded.

Check the current help documentation for the exact campaign type. An account-level setting may not apply to every format. Google notes, for example, that some controls do not apply to App campaigns or YouTube Sponsorship campaigns. A team can configure the right control in the wrong scope and believe it has protection that never applied.

Treat keyword exclusions carefully. Exact terms can miss synonyms, spelling variations, images, audio, sarcasm, or breaking news. Broad words can also block legitimate journalism and useful content. Use keywords as one signal, then review false positives and false negatives.

Layer 2: validate the supply path

Unsafe adjacency and counterfeit inventory are different risks, but both belong in a media-quality program. Supply transparency helps determine whether the claimed publisher and seller relationship is plausible.

For programmatic web, app, and CTV inventory, check:

• ads.txt or app-ads.txt authorization;
• seller identity in sellers.json;
• intermediary nodes in the OpenRTB SupplyChain object;
• direct versus reseller relationships;
• app bundle, store URL, domain, and publisher consistency;
• missing, incomplete, or contradictory declarations; and
• unexplained supply paths or reseller depth.

IAB Tech Lab says ads.txt and app-ads.txt let publishers declare authorized sellers, while sellers.json and the SupplyChain object help buyers identify direct sellers and intermediaries. These standards improve transparency and make counterfeit inventory harder to sell. They do not classify the meaning or tone of the surrounding content.

Use the supply-path transparency guide for a field-by-field workflow.

Layer 3: verify measurable delivery and context

Pre-bid controls describe the intended purchase. In-flight and post-bid evidence show what could actually be observed.

For each supported measured impression, retain the minimum fields required for audit:

• event ID and UTC timestamp;
• campaign, line item, creative, and placement IDs;
• publisher domain, page URL, app bundle, channel, or program identifiers where available and permitted;
• seller and supply-path fields where available;
• content category, risk level, confidence, and classifier version when a supported contextual provider supplies them;
• whether the content was classifiable under the method;
• viewability or render evidence supported by the environment;
• geo, device, and environment signals;
• invalid-traffic and source-quality results kept separate from content suitability; and
• policy version and the rule that produced the result.

Do not promise page-level context when the environment does not expose it. Closed social platforms may provide only aggregated placement or suitability reports. CTV and server-side ad insertion can limit device-side JavaScript and page-style URL evidence. In-app inventory may identify an app bundle but not the exact screen or adjacent content. The SSAI measurement guide explains why server-side delivery requires different evidence paths.

Open Measurement, often called OM SDK or OMID, supports standardized measurement communication in supported app and video environments. It does not by itself infer the meaning, sentiment, or suitability of the surrounding content. Treat measurement access and contextual classification as separate capabilities.

Layer 4: review, optimize, and preserve evidence

Review the first meaningful delivery window before scaling. Then monitor by source and inventory type throughout the campaign.

Look for:

• concentrated unsafe or unknown delivery from one publisher, app, seller, placement, or exchange;
• newly appearing domains or apps;
• sharp changes after a campaign-setting or inventory change;
• disagreement between platform and independent classifications;
• high invalid traffic alongside opaque or misrepresented inventory;
• unusually low measurement coverage;
• excessive reseller depth or incomplete supply paths; and
• false positives that block reputable journalism or contextually appropriate content.

Preserve the raw result, classifier and policy versions, evidence timestamps, and subsequent action. A screenshot can help explain an incident, but it should not replace structured campaign records. Collect screenshots or page content only when the environment, contract, and privacy policy permit it.

The metrics a brand-safety report should include#

A useful report shows denominators and coverage behind any badge that says "safe."

Measurable context coverage

context coverage = valid measured impressions with sufficient context / valid measured impressions

If context coverage is 65 percent, the suitability result describes that measurable portion. It does not prove the other 35 percent was safe.

Suitable, unsafe, and unknown exposure rates

Use separate buckets:

1. suitable under the active policy;
2. unsuitable or unsafe under the active policy;
3. unknown or insufficient context;
4. measurement failure; and
5. excluded invalid traffic, disclosed separately.

A transparent measurable-context version is:

unsuitable exposure rate = unsuitable valid impressions / valid impressions with sufficient context

Also report the share of all valid measured impressions. The first rate describes classified inventory; the second shows the total measured campaign impact.

Blocked versus exposed

A pre-bid or blocking vendor may report unsafe opportunities it prevented. A post-bid verifier reports impressions that were served or measured. Do not add blocked opportunities and exposed impressions into one undifferentiated count.

Use labels such as:

• evaluated opportunity;
• blocked before bid or serve;
• served and measured;
• classified suitable;
• classified unsuitable; and
• unclassified or measurement unavailable.

False positives and false negatives

Sample classifications for human review. Record:

• false-positive rate, where suitable content was blocked or labeled unsafe;
• false-negative rate, where unsafe content was labeled suitable;
• disagreement rate between providers;
• low-confidence rate; and
• appeal or reclassification outcomes.

Automated classifiers can disagree on context, sentiment, language, satire, and news. A review sample helps determine whether a low exposure rate reflects strong controls or an overly broad block policy.

Time to detect and resolve

Track:

• time from exposure to detection;
• time from detection to triage;
• time from triage to block or corrective action;
• repeat exposure after the action; and
• amount of spend and impressions affected.

Fast, documented response limits harm even when prevention is imperfect.

A pre-launch brand-safety checklist#

Use this checklist with the campaign owner, agency, buying platform, publisher, and verification provider.

Policy

• Is there an approved safety floor and brand-specific suitability policy?
• Are rules defined by category, risk level, language, market, product, and channel?
• Is the treatment of unknown inventory explicit?
• Are news, politics, live streams, user-generated content, games, and mature themes addressed?

Configuration

• Are platform controls applied at the correct account, campaign, and line-item scope?
• Are inventory types, categories, themes, keywords, placements, domains, apps, and channels configured?
• Are automated expansion and optimization settings understood?
• Have settings been independently reviewed and exported with timestamps?

Supply

• Are publishers and apps identified?
• Are ads.txt or app-ads.txt, sellers.json, and SupplyChain declarations checked where relevant?
• Is reseller depth acceptable?
• Are private marketplace or direct-deal terms consistent with the observed supply path?

Measurement

• Which channels support impression-level measurement?
• Which contextual fields are available in web, app, social, video, and CTV?
• What classifier, taxonomy, confidence threshold, and policy version will be reported?
• How are invalid traffic, viewability, suitability, and unknowns separated?
• What is the expected measurement and classification coverage?

Operations

• Who reviews alerts and placement reports?
• What threshold pauses a source or campaign?
• How quickly must the vendor and publisher respond?
• What evidence is retained, for how long, and under which privacy rules?
• Who approves exceptions and policy changes?

How to handle a brand-safety incident#

1. Preserve the evidence

Record the exact timestamp, campaign, creative, source, placement, domain or app, program or page context where available, seller path, classification, policy rule, and measurement method. Preserve an authorized screenshot or contextual record if permitted.

2. Contain the exposure

Block or pause the smallest reliable unit that stops the problem: placement, app, domain, channel, publisher, seller, inventory package, or campaign. A broad campaign pause may be justified for severe or widespread exposure.

3. Confirm what happened

Check whether the evidence represents a delivered impression, a blocked opportunity, a preview, or a classification-only alert. Reproduce the policy decision. Review the content in full context when lawful and possible. Determine whether the problem was a settings error, classification error, newly changed content, unsupported environment, or inventory misrepresentation.

4. Notify the right parties

Follow the approved escalation path. Provide bounded evidence to the buying platform, publisher, agency, verification provider, legal team, or communications owner. Avoid circulating personal data or raw identifiers beyond what the investigation requires.

5. Fix and verify

Update the relevant exclusion, allowlist, threshold, supply rule, or partner setting. Verify that exposure stopped, then monitor for recurrence and displacement into another source.

6. Complete a root-cause review

Document the cause, affected impressions and spend, detection lag, corrective action, coverage limitation, and policy improvement. A post-incident review turns one bad placement into a stronger control system.

Channel-specific limits to understand#

Open web

Page URLs and contextual content may be accessible, but dynamic pages, infinite scroll, redirects, frames, privacy controls, and rapid content changes complicate classification. The content seen at review time may differ from the content adjacent to the impression.

Mobile apps

App bundle and store metadata help identify the app. Exact screen-level or adjacent-content context may be unavailable. Check app-ads.txt, store ownership, SDK support, and whether the chosen verifier supports the app environment.

Social and user-generated content

Closed platforms control most delivery and context data. Use their placement and suitability controls, approved third-party measurement where available, and platform-specific reports. Do not assume an open-web tag can see the surrounding feed or video.

Online video and CTV

Program, channel, genre, rating, and content identifiers can support suitability when publishers and platforms pass them. IAB Tech Lab's CTV Programmatic Guide describes Content Taxonomy 3.1, supply transparency, and other signals that can help buyers evaluate inventory. Availability still varies by platform and transaction.

Search and commerce media

The concern may be the search query, product adjacency, marketplace seller, or sponsored placement rather than a page article. Define the unsafe context for that environment and confirm which query or placement data the platform exposes.

Common brand-safety mistakes#

Treating a blocklist as permanent

Domains, apps, pages, and ownership change. Refresh lists and review whether they still reflect the policy. A stale allowlist can create false confidence too.

Blocking all news

Broad news exclusions can reduce reach and defund credible journalism while still missing unsafe content elsewhere. Use category, risk, source, and context controls that match the brand's actual tolerance.

Confusing fraud with unsafe content

An impression can be human and viewable beside unsuitable content. It can also be invalid traffic on a contextually safe page. Report invalid traffic, viewability, source authenticity, and brand suitability as distinct dimensions.

Ignoring measurement coverage

A 99 percent suitable rate with 30 percent context coverage is not a 99 percent campaign-wide guarantee. Always put coverage beside the result.

Assuming one vendor covers every channel

Confirm support for web, in-app, online video, social, CTV, and server-side delivery separately. Ask which fields are observed, modeled, supplied by the publisher, or unavailable.

Where AdProof fits#

AdProof can help a media team bring supported delivery events, source and placement identifiers, geo and device evidence, invalid-traffic results, reach, frequency, and other measurement signals into one campaign evidence trail. That makes it easier to see where quality or suitability concerns concentrate and to prepare a bounded discrepancy ticket.

Brand-safety verification still depends on the context exposed by each channel and on the contextual provider or platform used for classification. AdProof should not be treated as a guarantee that every page, app screen, social post, or CTV program was classified or blocked. A scoped pilot should define the supported inventory, expected context coverage, policy, evidence fields, and escalation process before launch.

Explore AdProof campaign verification, review how to verify digital ad delivery, or talk to the team about a measured pilot.

Frequently asked questions#

What is the best way to ensure brand safety in digital advertising?

Use a documented suitability policy, platform and pre-bid controls, supply-path checks, independent post-bid evidence where supported, measurement coverage reporting, human review samples, and a tested incident process. No individual control covers every channel or failure mode.

What is the difference between a brand-safety blocklist and an allowlist?

A blocklist excludes named sources or placements. An allowlist permits only approved inventory. Allowlists can provide tighter control but reduce scale and require maintenance. Neither approach replaces contextual review, supply validation, or monitoring.

Can keyword exclusions guarantee safe placements?

No. Keywords can miss visual, audio, multilingual, or newly emerging context and can block legitimate content when a word has several meanings. Combine them with taxonomy-based classification, placement controls, source review, and post-bid measurement.

Should unknown content be counted as unsafe?

That is a policy decision. Strict brands may block unknown inventory, while others cap or monitor it. In reporting, keep unknown separate from suitable and unsafe so stakeholders can see the coverage gap and the decision taken.

Does ads.txt prove that a placement is brand safe?

No. Ads.txt indicates which sellers a publisher authorizes to sell its inventory. It helps with supply authenticity, not the meaning or suitability of the content beside the ad.

Does OMID measure brand safety?

OMID supports standardized measurement communication in supported environments. It does not independently understand the surrounding content. Contextual metadata or a separate classification method is still needed for a suitability decision.

How often should brand-safety controls be reviewed?

Review them before every campaign, during the first material delivery window, after any channel or settings change, and at a cadence proportionate to spend and risk. High-risk or fast-moving campaigns may require daily monitoring and immediate alerts.

← Back to all field notes