Author: Nita Farahany Source: https://nitafarahany.substack.com/p/is-an-algorithm-speech Published: February 13, 2026
This is a class write-up by Nita Farahany, published February 13, 2026 as Class 4.3 of her Advanced Topics in AI Law and Policy course (one of roughly 30 classes). It closes Week 4 by framing the question of whether a platform's recommendation algorithm is itself constitutionally protected expression. Farahany argues that the answer determines the scrutiny standard applied to later child-safety regulation: if algorithmic curation is "speech," architectural regulation faces strict scrutiny and likely fails; if it is not, intermediate or rational-basis review opens more regulatory room. The Supreme Court, she writes, has so far refused to answer the question cleanly.
Summary of argument
The class traces the doctrinal lineage from editorial discretion to algorithmic curation. Miami Herald v. Tornillo (1974) established that editorial discretion — what a newspaper chooses to publish, where to place it, and how prominently — is core First Amendment expression that the government cannot compel. Turner Broadcasting (1994) extended that protection to cable operators under intermediate scrutiny.
The open question Farahany poses is whether automated, machine-learning-driven recommendation ranks as editorial discretion in the same sense. She lays out two competing positions. On the platform view, the platform designed the algorithm, set its objectives, and deploys it to curate, so its output is expression even if no human reviewed any individual decision. On the regulator view, the platform did not "speak" each ranking decision; the algorithm inferred it from behavioral signals, and there is no expressive intent at the granular level — if a speaker cannot say what they said, it is not speech.
Farahany frames Moody v. NetChoice (2024) as the case where the Supreme Court had the opportunity to resolve this and declined. The Court reviewed Texas (HB 20) and Florida (SB 7072) laws that constrained large platforms' ability to remove or demote content, and vacated and remanded both decisions without resolving the algorithmic-speech question. According to Farahany, Justice Kagan's plurality opinion signaled that traditional curation — whether to host, demote, or remove — is protected expression, but stopped short of holding that algorithmic ranking is expression.
She connects this to the "speech certainty" principle advanced by Austin and Levy (Stanford Law Review, 2025): speech is only speech if the speaker knows what they said when they said it. Machine-learning systems frequently cannot satisfy this, because engineers often cannot explain why a specific user received a specific recommendation at a specific time. If the speaker cannot account for the message, Farahany asks whether the message qualifies as expression; she presents the doctrine as unsettled.
The practical stake Farahany identifies concerns the architectural regulations covered later in the course (Week 9): the California Age-Appropriate Design Code Act (CAADCA), COPPA 2.0, the Kids Online Safety Act (KOSA), and Article 25 of the EU Digital Services Act (DSA). If algorithmic curation is fully First Amendment-protected, she argues, each of these likely fails strict scrutiny; if it is not, intermediate or rational-basis review opens substantially more regulatory room. She projects that the doctrine will be resolved within the next 3 to 5 Supreme Court terms, and that until then every state-level law is litigated against this uncertain backdrop.
Key claims
- Miami Herald v. Tornillo (1974) established editorial discretion as core First Amendment expression that the government cannot compel; Turner Broadcasting (1994) extended this to cable operators under intermediate scrutiny.
- Whether automated, machine-learning-driven recommendation counts as editorial discretion is contested between a platform position (the designed algorithm is expression) and a regulator position (no expressive intent exists at the granular ranking level).
- In Moody v. NetChoice (2024), the Supreme Court reviewed Texas (HB 20) and Florida (SB 7072) laws and vacated and remanded both without resolving the algorithmic-speech question; Justice Kagan's plurality signaled that traditional curation is protected but did not hold that algorithmic ranking is expression.
- The Austin and Levy "speech certainty" principle (Stanford Law Review, 2025) holds that speech requires the speaker to know what they said; many machine-learning systems cannot meet this.
- The classification of algorithmic curation determines whether CAADCA, COPPA 2.0, KOSA, and EU DSA Article 25 face strict scrutiny (and likely fail) or a lower standard (and gain regulatory room). Farahany expects resolution within 3 to 5 Supreme Court terms.
Relationships
- part-of: Nita Farahany Advanced Topics course (Class 4.3 of ~30)
- related: AI and the First Amendment
- previous: Inside My Advanced Topics Class 4.2: The Shield — Section 230 (Farahany, February 2026) next: Inside My Advanced Topics Class 5: The Perfect Friend (Farahany, February 2026)