Part 1: Inclusion by Design

Welcome to part 1 of The Inclusive Contextual Suitability Declarations.

Last week, Barometer published an Inclusive Brand Safety and Suitability Framework, highlighting learnings we’ve extracted from the long-standing GARM standard definitions as well as three key areas and identified areas where there is room for improvement.

In this post, I expand on these learnings as the original announcement primarily served to circulate the Inclusive Brand Safety and Suitability Framework itself.

The first thing I want to do is emphasize two positive learnings from GARM that are often overlooked or willfully ignored: the distinction between brand safety and brand suitability as well as the special definition carve-outs for news content.

First, a juxtaposition of brand safety and brand suitability.

• brand safety — A binary designation for a piece of content nodding to the Global Alliance for Responsible Media’s (GARM) brand safety floor framework. Content that is not brand safe (also known as “below the brand safety floor” or “below the floor”) is never brand safe in all cases and does not depend on the brand. The definitions for what is considered not brand safe are very specific for each GARM category and are most often crimes. All brands exclude all unsafe placements.
• brand suitability — A spectrum first defined by the Global Alliance for Responsible Media’s (GARM) brand suitability framework for categorizing content into three distinct risk levels (low, medium, high) for each GARM category. Though often confused with brand safety, brand suitability is NOT a binary. The suitability of a given piece of content is determined only in reference to a brand’s specific standards — it does not exist in absolute terms like the definitions for brand safety. To elaborate, the same piece of content may be suitable for one brand and unsuitable for another brand because each brand defines their risk tolerance slightly differently for suitability.

Notably, three categories do not have clear distinctions between the brand safety floor and high risk for brand suitability. Those categories are: Hate Speech, Debated Sensitive Social Issues and Misinformation (which was removed in the September 2023 version of the document). The conflation of these two terms results in technologies designed for brand safety being applied inappropriately in use cases of brand suitability. This leads to irrational over-blocking that harms publishers and advertisers alike. However, this was surely not as intended given GARM’s specific segmentation of the two frameworks.

Get Barometer’s stories in your inbox

Join Medium for free to get updates from this writer.

Remember me for faster sign in

Second, the GARM definitions specifically make exceptions for news content. They define breaking news or op-ed coverage as “medium risk” and news features or informative treatment of a subject as “low risk”. Though this is clearly articulated in the GARM framework, it is rarely accurately operationalized and applied. As a result, and in combination with harmful keyword-based approaches (discussed more below), news is too often inappropriately blocked even when it’s the weather, sports, or local updates. Furthermore, news is not referenced at all in the Brand Safety Floor. One possible reason for the common misinterpretation of this seemingly plain English is that GARMs’ sell-side membership is composed primarily of walled garden platforms, where there is very little “journalism” or “news” to speak of, editorial controls (common in real journalism) are nonexistent and the creators are not professionals bound to a rigorous code of ethics.

Having this common foundation as a backbone, we can proceed next to the three learnings showcasing what can be improved:

1. User Generated Content vs Premium Content — Different standards are required to evaluate content that is brand safe by design. Such content, by definition, does not contain the illegal and harmful contents that are so feared in user-generated content environments. In addition, this content could live in walled gardens, but more often lives on the open internet, where advertisers can have full transparency and control over their contextual alignment. In contrast, walled gardens, such as social media platforms, have first party and third party controls available, but ultimately the environment is not fully transparent to advertisers in the same way open internet environments are. Thus, produced premium content on the open internet ought to be evaluated with tools that are fit for purpose, rather than getting absolutely botched by unsuited solutions.
2. Keywords != Context — Even with the best intentions and the most accurate definitions, there is still a challenge to enforcing brands’ standards. Today, an unfortunate legacy of this challenge is keyword block lists. Keyword blocklists are pernicious for many reasons: they are not updated enough, always added to more than removed from, and created to solve a problem they fundamentally cannot solve because of the nuance of language. Contextual approaches are required to properly interpret whether the environment is suitable for an ad placement. However, contextual analysis is difficult and requires serious machine learning chops. Barometer is the leader of inclusive contextual suitability because of our AI’s unique exposure to massive quantities of premium, omnichannel content including nearly 200 years of transcribed audio or video and digital content from over 80k premium providers spanning 6 years and a substantial financial investment. We’ve invested in learning context from multi-modal inputs such as conversations and have developed multiple layers of contextual reasoning to overcome the limitations faced by keywords based solutions.
3. Centrality of Publishers — Though social platforms and other walled gardens are members of GARM, traditionally, publishers have not been. In fact, there has always been sort of an iron curtain between the brand safety toolsets for publishers and those for advertisers. Notably, because of GARM’s dominant focus on social media, open-internet publishers, and particularly news publishers, are not represented in regular working groups. As a consequence, publishers grow in resentment for ad-tech vendors, advertisers tolerate misrepresentation of their standards and consumers don’t hear the messages needed to drive ROI. More importantly, there is little transparency between publishers and advertisers, resulting in decreased trust. Barometer’s out-there idea is that publishers, agencies and advertisers should all have the opportunity to access the same transparent data. After all, we are all in this together.

To illustrate the importance of this, consider the following scenarios:

a. In Scenario A, an advertiser sees all of the brand suitability data of the publisher but the publisher has no visibility. The advertiser will make decisions and dictate them to the publisher and the publisher enforces those decisions. This is not really an ideal partnership because the advertiser is really missing out on the expertise of the publisher. The publisher may be able to share additional context about the programming, but why would the advertiser trust them when they are just trying to sell. This is the reality of today IF a direct relationship even exists between the advertiser and publisher.

Now consider the next scenario

b. In Scenario B, both the advertiser and the publisher have access to the same content data at the start of the planning cycle. The advertiser shares their decisions with the publisher and the publisher can actually track each of the advertisers’ assumptions. The publisher knows their content better than anybody, but now they also understand how advertisers perceive the content. They can speak the same language. The advertiser may say “I want to eliminate all content with medium and high risk for arms & ammunition”. The publisher can ask why and the advertiser may share they’re seeking to stay away from discussions of current wars. Now they are having a conversation. The publisher can then share a relevant insight like: 50% of content proposed for exclusion because of Arms & Ammunition discussion is actually History content not current events. Thus there is limited risk of discussion of current wars because the focus is historical, the content may be scripted, etc. This type of relationship, in which the advertiser gets to be an expert in their standards and the publisher gets to be an expert in their content, makes the best of both cohorts’ unique strengths.

Obviously the above scenario represents a direct relationship. There is more to be said about the power and potential of programmatic and we will get to that in the third post in this series.

For now, if you’ve read this far, thank you for taking the time to reflect deeply on the future of brand safety, and for your consideration of a contextually suitable future instead. Don’t forget to stay tuned for: Part 2 — Inclusion by definition and Part 3 — the interoperable infrastructure of inclusion!