Policy Advisory Opinion

A Policy Advisory Opinion is a written recommendation from an independent body to a company on how a particular rule for content or technology should be designed. The format became well known through Meta's Oversight Board, which, at the company's request, reviews entire rule sets rather than individual cases.

Large internet companies have their own rule sets governing what is allowed on their platforms and what gets removed. These rules are not state law but homegrown. Some companies seek advice on such rules from an independent body that is not part of the company. A Policy Advisory Opinion is the written result of such consultation. In doing so, the body does not review a single removed post but the rule itself, and proposes how it should be changed. The English name literally means something like: an advisory opinion on a policy.

Critiquing the rule instead of reviewing the individual case

When a user complains about a removal, usually only that single case is reviewed. That changes nothing about the underlying rule. A Policy Advisory Opinion operates one level higher. It asks: Is the rule even sensible, understandable, and fairly worded? A single opinion can therefore affect billions of posts.

For the public, the format is interesting above all because it provides insight. Such opinions are published and often contain figures and internal processes that no one else gets to see otherwise. Journalists and researchers use them as a source. Critics, on the other hand, argue that the whole thing can also serve as a fig leaf. The company seeks advice but does not have to follow it.

It is important to distinguish this from a court ruling. A court decides bindingly, whereas a body of this kind only recommends. At Meta, only the decisions on individual pieces of content are binding. For the recommendations on rule design, the company merely has to respond publicly.

From the request to the company’s response

The process almost always begins with the company itself. It poses a specific question to the body about a rule it considers difficult. The body then collects public submissions. Human rights organizations, academics, associations, and private individuals are all allowed to participate. In some proceedings, several thousand submissions have come together.

The members then deliberate in smaller groups and draft a text. In doing so, they frequently draw on international human rights standards, such as the United Nations' rules on freedom of expression. The end result is a document containing a series of numbered recommendations. Such proceedings typically take many months.

The final step is the company’s response. It must publicly address each recommendation, usually within a fixed deadline of 60 days. It can fully implement the recommendation, partially implement it, or reject it. This very list of responses shows just how much influence such a body really has.

Where such opinions turn up

The best-known example is the Oversight Board, which Meta set up in 2020 for Facebook and Instagram. Among other things, it issued opinions on the handling of posts by politicians and on an Arabic word that can be an insult or harmless depending on context. Such texts are regularly cited in tech news.

For the AI industry, the format is becoming increasingly relevant. Language models, too, need rules for which responses they refuse and which they don’t. Some providers are now working with external advisory boards or publishing their behavioral guidelines for public comment. That is not yet the same as a formal Policy Advisory Opinion, but it points in the same direction.

A common misconception is that a state regulatory authority makes the decision here. That is not true. These bodies are financed and appointed by the companies themselves. Their influence rests on publicity and reputation, not on state power. Laws such as the European Digital Services Act, by contrast, work with genuine obligations and fines.

Subscribe free. Unsubscribe the second it sucks.

High-signal news across AI, business, UX, and tech. Every morning.