What is brand safety, and how is it managed?
Preventing advertisements appearing alongside content that would damage the brand — and the discipline has matured from crude keyword blocking into something considerably more careful, partly because the crude version caused real harm.
The distinction that matters:
Brand safety is avoiding genuinely damaging adjacency — extremist material, graphic violence, illegal content.
Brand suitability is the finer judgement: whether content is appropriate for this brand, in this campaign. A news article about an air crash is not unsafe, and is plainly unsuitable for an airline's advertising. Suitability is brand-specific; safety is close to universal.
How it is managed:
Inclusion lists of approved publishers, which is the most reliable approach and narrows reach.
Exclusion lists of blocked domains and applications.
Keyword blocking, the traditional method and the most problematic.
Content classification, using standardised frameworks to categorise pages by topic and risk tier.
Pre-bid filtering, rejecting inventory before bidding rather than reporting on it afterwards.
Third-party verification, measuring where advertisements actually appeared.
Why keyword blocking backfired. Blunt blocklists defunded legitimate journalism at scale: blocking words like "shooting", "attack" or "death" removed advertising from news coverage, hitting exactly the reporting that most needed funding. Blocklists have also been shown to disproportionately demonetise content about minority communities and health topics, because the associated terms were blocked wholesale. Contextual analysis at sentence and page level, rather than word matching, is the correction.
The other side: ad-funded harm. Programmatic buying routinely places advertising on disinformation sites, pirated content and made-for-advertising pages, funding them. Many brands do not know where their money goes, which is the argument for supply path transparency and published inventory lists.
Practical approach: define suitability explicitly rather than inheriting a generic blocklist; review actual placement reports; exclude by domain rather than by keyword where possible; and accept that maximum safety and maximum reach are in tension.