On large social platforms, artificial intelligence mostly does two jobs: it ranks which posts appear in a feed, and it flags or labels content that was made or edited with AI tools. In the European Union, the Digital Services Act (DSA) requires every online platform to explain the main parameters of its recommender systems in plain language, and platforms with more than 45 million monthly EU users must also offer at least one feed that does not rely on profiling.
Key takeaways
- A recommender system is software that decides what content a user sees and in what order, and most such systems learn from past behavior.
- DSA Article 27 covers online platforms using recommender systems and requires disclosure of the main ranking criteria and any user options to change them.
- DSA Article 38 adds a non-profiling feed option for very large platforms, such as a chronological feed.
- Meta renamed its "Made with AI" label to "AI info" on July 1, 2024, after minor edits such as retouching had been labeled.
- Platforms do not publish full ranking formulas, so outside observers see only what disclosure rules and company posts reveal.
What AI does inside a feed
A social feed is a ranked list. When a person opens an app, software chooses from thousands of candidate posts and orders them, usually by predicting which ones the person is likely to watch, like, share, or skip. Those predictions come from machine learning models trained on past interactions. This is why two people following the same accounts can see different feeds.
A second job is moderation support, where models help flag spam, scams, or policy violations for review. A third, newer job is detection and labeling of synthetic media. Public documentation for these systems is thin and varies by company, so the most reliable facts come from laws that force disclosure and from policy posts the companies publish themselves.
What the Digital Services Act requires
The DSA is EU Regulation 2022/2065. Its obligations for in-scope online services have applied since February 17, 2024, and the European Commission describes the largest platforms as those with over 45 million monthly users in the EU. The rules below concern recommender systems specifically, and they apply to services offered in the EU, not to US or Canadian users as a matter of EU law.
| Provision | Who it covers | What it requires |
|---|---|---|
| Article 27 | Online platforms using recommender systems | Describe the main parameters in the terms and conditions, in clear language, and explain why each criterion matters |
| Article 27, ranking options | Platforms offering several ways to order content | Provide a function to choose and change the preferred option at any time, reachable from the place where content is ranked |
| Article 38 | Very large online platforms and search engines | Offer at least one option for each recommender system that is not based on profiling |
In practice, Article 38 is the reason large platforms in Europe offer a feed that is not personalized, such as one sorted by time. The Commission's overview of the Digital Services Act package summarizes this as the ability to opt for non-personalized feeds. Recital 94 of the regulation adds that such options should be directly accessible from the interface where recommendations appear.
Labeling AI-made content: one platform's timeline
Labeling is a company policy rather than a statute in most places, so it differs by platform. Meta published its approach on April 5, 2024, after recommendations from its Oversight Board. The company said it would keep most manipulated or AI-generated content online and add labels and context instead of removing it, unless the content breaks other rules. Labels are applied when Meta detects industry-standard AI image indicators or when a person discloses that content is AI-generated.
| Date | Change reported by Meta |
|---|---|
| April 5, 2024 | Policy announced, with labeling planned for May 2024 |
| July 1, 2024 | "Made with AI" renamed "AI info" because some minor AI edits were being labeled |
| September 12, 2024 | For content only edited by AI tools, the label moved to the post menu; labels for fully AI-generated content stay visible |
The label depends on signals embedded in files or on self-disclosure, so content with no such signal can go unlabeled. Meta's page does not claim that every synthetic post is caught.
What can and cannot be checked from outside
A reader who wants to understand a specific app can check a short list of things:
- Whether the terms and conditions describe the main ranking parameters, as Article 27 requires for EU services.
- Whether a chronological or non-personalized feed is offered, and where the switch sits in the interface.
- Whether the platform has a public policy on labeling AI-generated content, and what triggers the label.
- Whether the platform publishes the date of its last policy change.
What cannot be checked is the model itself. Disclosure rules require a description of the main criteria, not source code or training data. Claims that a platform amplifies one kind of content therefore rest on external audits, researcher access, or company statements, and each has limits. This article makes no claim about any specific platform's ranking outcomes.
FAQ
Does the Digital Services Act apply in the United States or Canada?
No. It is EU law and applies to services offered in the EU. Platforms sometimes roll out features globally, but the legal duty to offer them comes from the EU rules.
Is a chronological feed free of AI?
Not entirely. A time-sorted feed removes personalized ranking, but platforms still use machine learning for spam detection, moderation support, and other functions that do not change the order of posts.
Why was the "Made with AI" label renamed?
Meta said some content with only small AI edits, such as retouching, was being labeled, so it changed the wording to "AI info" on July 1, 2024.
Can a user see exactly why a post was recommended?
Only in general terms. Article 27 requires the main criteria and their relative importance to be described, but individual explanations depend on the tools each platform builds.
Related reading
- AI for social media engagement and reach
- How AI is changing the English language
- Character AI as a digital companion
- Human-centered AI design
More in AI Tools and Industry.
Last verified: October 2026
Sources
- Regulation (EU) 2022/2065 (Digital Services Act), Articles 27 and 38 and Recital 94, Official Journal of the European Union, 2022
- The Digital Services Act package, European Commission, 2026
- Meta approach to labeling AI-generated content and manipulated media, Meta, April 5, 2024, with updates through September 12, 2024
- DSA now fully in force, Slaughter and May, 2024