When a feed seems to know what you want to watch, a recommendation system is at work. Platforms rarely publish full details of how these systems work, and anyone who claims to know a secret formula is guessing. What can be said reliably is what the platforms themselves have published and what the European Union now requires them to explain. This guide summarises those sources, with their dates, and avoids guesses.
What is a recommendation system?
A recommendation system is software that decides which items to show to which person, and in what order. On a video or photo platform there are far more posts than anyone could look through, so ranking is unavoidable. The system predicts what a given viewer is likely to find worth their time and orders the feed accordingly.
“The algorithm” is shorthand, but it hides the fact that several systems, each tuned for a different part of an app, are usually involved. That is exactly what Instagram says about its own service, as the next sections show.
What does YouTube say about its recommendations?
YouTube’s vice-president of engineering, Cristos Goodrow, set out the company’s approach in a post on the YouTube blog dated 15 September 2021. The post says the system draws on many signals, including clicks, watch time, survey responses, shares, likes and dislikes. It describes the system as “built on the simple principle of helping people find the videos they want to watch and that will give them value”, and says it demotes “borderline content”, which it defines as content that “comes close to, but doesn’t quite violate” its Community Guidelines.
On viewer control, the post says people can pause, edit or delete their search and watch history, which changes the information the system uses.
What does Instagram say?
Adam Mosseri, then head of Instagram, wrote on 8 June 2021 that Instagram does not have one algorithm that oversees what people do and do not see. It uses several ranking systems, each tailored to a different part of the app. The post describes how Feed, Stories, Explore and Reels work differently:
- Feed and Stories favour accounts the person follows, using signals such as the popularity and details of a post, information about the poster, the person’s own activity and their history of interacting with that account.
- Explore recommends content from accounts the person does not yet follow, and the post says it weights how popular a post is more heavily than Feed does.
- Reels focuses on entertainment and draws on a person’s recent engagement and signals about the content, such as audio and video understanding.
The practical point is that there is no single dial to turn. A post that does well in one part of the app may be treated differently in another.
What does EU law require platforms to explain?
Under Article 27 of the Digital Services Act (Regulation (EU) 2022/2065), providers of online platforms that use recommender systems must set out in their terms and conditions, in plain and intelligible language, the main parameters used in those systems and any options users have to change or influence them. The “main parameters” must include, at a minimum, the criteria that matter most in deciding what is suggested and the reasons those criteria have the importance they do.
Very large online platforms, which are those with at least 45 million average monthly users in the EU that the Commission has designated, must also provide at least one option for each of their recommender systems that is not based on profiling (Article 38). The Commission summarises this as users on large platforms with over 45 million monthly users in the EU being able to opt for non-personalised feeds, such as chronological ordering.
These are transparency and choice requirements. They do not give anyone access to the code, but they do mean platforms operating in the EU have to explain, in public terms, the principal factors behind their recommendations.
How do the three sources compare?
| Source | Date | Main message |
|---|---|---|
| YouTube blog (Cristos Goodrow) | 15 September 2021 | Many signals, including watch time and survey responses; aims to help people find videos that give them value; users can manage history |
| Instagram (Adam Mosseri) | 8 June 2021 | No single algorithm; different ranking for Feed, Stories, Explore and Reels |
| EU Digital Services Act, Article 27 | Regulation (EU) 2022/2065 | Platforms must explain main parameters and any user options in plain language |
What does this mean for creators?
The sources support a few sober conclusions:
- There is no single “hack”. If each part of an app ranks differently, advice that promises one trick for all of them deserves scepticism.
- Viewer value is a stated goal. YouTube describes its system as built on helping people find videos “that will give them value”, so a click on its own is not the stated aim.
- History and behaviour matter. Signals about what people watch, like and interact with feed the rankings, so consistency and clarity about who a channel is for can help viewers find it.
- Reading the platform’s own explanation is worth the time. Both the platforms and, in the EU, the law now encourage public explanations of how ranking works.
Our guide to what makes a moment go viral looks at research on spread, and how brands choose influencers explains why brands look beyond reach alone.
What questions should creators ask of their own analytics?
Platform analytics are the most direct evidence a creator has about how recommendation systems treat their work. Instead of looking only at total views, it helps to ask where views came from, such as followers, recommendations or search, since the sources above show that systems rank content differently across parts of an app. It helps to look at how much of each video people watch, because YouTube says watch time is among its signals. And it helps to compare content that performed well with content that did not, looking for differences in topic, format and opening seconds.
None of this reveals the algorithm. It does show what the audience responded to, which is a more reliable guide than any claimed shortcut.
What can viewers do?
Viewers have more control than many realise. On YouTube, managing watch and search history changes what the system learns from. Where EU rules apply, users of very large platforms can switch to a feed that is not personalised. Following accounts deliberately and using the controls each app provides will shape what appears.
What are the limits of these sources?
Two cautions apply. The platform posts quoted here are from 2021, and systems change often, so they describe the approach at that time rather than today’s exact rules. And they are self-descriptions, not independent audits. The Digital Services Act adds a public record of main parameters, but readers should check each platform’s current terms and help pages for the latest wording.
The takeaway: understand the system, do not chase it
Recommendation systems rank content using many signals, differ across parts of the same app and are increasingly required to be explained. For creators and viewers alike, the useful response is to read what the platforms publish, make content that people finish and value, and be wary of anyone selling certainty.
This article summarises public statements and law for general information. Platform systems and legal requirements change; check current platform and regulator pages.
Sources: YouTube Blog (Cristos Goodrow, “On YouTube’s recommendation system”, 15 September 2021); Instagram (Adam Mosseri, “Shedding More Light on How Instagram Works”, 8 June 2021); Regulation (EU) 2022/2065, Digital Services Act, Articles 27, 33 and 38; European Commission (Digital Services Act package)
