Brand Safety: brand protection in digital advertising

Three types of content that brands protect themselves from

Before setting up any tool, it’s important to distinguish threats — because each requires a different level of response. There are only three of them.

• Content prohibited by law — piracy, extremism, child exploitation. Tolerance here is always zero, regardless of industry and KPIs.
• Controversial content — wars, politics, nudity, profanity. Brand response depends on audience and reputational risks: a snus manufacturer and a children’s brand need different settings.
• Content not suitable for this specific brand — a dynamic list formed individually. An automotive brand may find ads near accident news unacceptable, while an insurance company may consider this normal context.

It is precisely the mixing of these three categories into one general block that creates the problem of excessive filtering, which is described below.

How to choose the level of risk tolerance

There is no universal answer here — but there are three working models between which one should consciously choose, not default.

The mistake that people most often make is choosing a low level of 'just in case', although business by its nature is performance. This is a direct loss of reach without corresponding benefit.

Case: why vice.com and economist.com are unfairly blocked

Here is a specific example of what happens when there is thoughtless rigidity. Analysis showed that 52,8% of content on vice.com and 21% on economist.com are automatically marked by algorithms as 'unsafe'—although both resources are considered quality, verified journalism. The reason is simple: algorithms react to individual words and topics (war, crime, politics), not distinguishing between the context of an investigative article and actual extremist content.

A similar story with keywords in general: according to GroupM, 68% of all content blocks in verification systems are caused by only 10 keywords. The word 'murder' in a news summary and in a crime thriller are blocked equally—although the brand risks are different. The result is that advertisers lose access to traffic from quality publications, pay more for the remaining inventory, and receive a distorted picture of campaign effectiveness.

The conclusion is simple: blacklists based on keywords without manual audit of top 30 domains are not protection, but a self-inflicted wound to reach.

DoubleVerify, IAS and built-in platform tools: what is the real difference

Choosing a verification partner is often based on price or habit, not on the task. The difference between solutions is more significant than it appears.

Built-in platform tools are not a replacement for a third-party verifier, but a baseline. The conflict of interest is obvious: the platform is not interested in strictly limiting its own inventory. Independent verification is mandatory for a low level of risk tolerance, for a high level — built-in settings are sufficient.

Which metric actually shows ROI Brand Safety

Viewability by itself says nothing about return on investment — it simply records that the banner was visible for at least 50% of the area for 1 second (for video — 2 seconds by IAB/MRC standard). This is a hygiene minimum, not an efficiency indicator. ROAS is too general — dozens of factors influence it besides Brand Safety.

The working metric is Quality CPM, developed by GroupM: the cost of 1000 impressions that simultaneously meet three conditions — visibility standard, absence of invalid traffic (bots), and safe environment. It is this metric that allows comparing purchases between platforms and verification partners, not just by formal CPM.

Verification should be structured in three stages — Pre-bid (filtering before auction), Bid Response (checking during purchase) and Post-bid (analysis after placement). Using only one stage is a typical shortcoming, due to which budget leaks even after the ad has been shown.

A telling example — case X5 Group: transitioning to proactive verification at all three stages enabled achieving potential savings of 4–22% of the advertising budget depending on the platform by abandoning low-quality placements, while viewability increased by 76% and conversions — by 94%. This is the measurable ROI Brand Safety, not abstract 'reputation protection'.

Why automation does not replace manual audit

No algorithm can yet reliably distinguish context at 100%. An automatic system trained to block the word 'attack' reacts equally to a sports news about an attack in football and a report about a terrorist attack. This is called false-positive — false blocking of normal content, which quietly consumes reach without visible reason.

The only way to avoid this is a combination of technologies and human control:

1. Export a list of domains and placements with the highest number of blocks for the period.
2. Conduct a manual audit of the top 30; evaluate the actual context, not just formal keyword matching.
3. Adjust the list of exceptions individually for each brand category, rather than using a generic platform template.
4. Repeat the audit quarterly — placement content changes faster than campaign settings.

A separate level of task — protection against unauthorized use of the brand logo by third parties, for example, in phishing or counterfeiting. The experience of Uzum Market is telling: Meta Brand Rights Protection tool recognizes the logo using computer vision even when distorted — changed transparency or applied filters — while the NLP module separately calculates suspicious textual links typical for phishing schemes. Without such a combination of technologies, counterfeiters bypass simple filtering based on image or text in isolation.