A court left open whether chatbot output counts as speech

What happened

On 21 May 2025, Judge Anne Conway of the U.S. District Court for the Middle District of Florida filed a 49-page order ruling on the motions to dismiss in Garcia v. Character Technologies. The order, entered on the case docket as document 115, granted the motions in part and denied them in part: claims against Alphabet were dismissed without prejudice, an intentional-infliction-of-emotional-distress claim was dismissed with no leave to amend, and the remaining claims, including product liability and negligence, were allowed to proceed. This is a procedural ruling, not a verdict on what happened.

What the documents show

The order records that the defendants argued the First Amendment barred the suit entirely, on the theory that Character.AI's chat output is speech its users have a right to receive. The court agreed the company could raise its users' First Amendment interests, a step the order works through by analogy to standing rules from other listener-rights cases, but on the second and separate question of whether the chatbot's output actually qualifies as speech, the order states plainly that the court 'is not prepared to hold' that it does at this stage of the case, citing a concurring opinion on AI-driven content moderation for support. The order also allowed a jurisdictional challenge from the two individual founder-defendants to be renewed after limited discovery, and let an unjust-enrichment claim proceed on the theory that user data had commercial value to the company.

The mechanism

The legal mechanism here is the distinction between pure speech and expressive conduct under First Amendment doctrine, and the further question of whether a large language model's word selection expresses anyone's viewpoint at all. The order treats this as unresolved rather than settled, finding the defendants' analogies to video games and social-media moderation incomplete because those precedents turned on a human editorial choice behind the content, something the court says has not yet been shown for Character.AI's automated output. A motion-to-dismiss order like this one does not weigh evidence; it only asks whether the plaintiff's alleged facts, taken as true, could support a legal claim.

What it leaves open

The order explicitly reserves the free-speech question for a later stage, meaning no court has yet ruled on whether chatbot output is or is not protected speech in this case; that question could return at summary judgment or trial. It also does not resolve the underlying factual disputes about the app's design or the events preceding the death described in the complaint, since at this stage the court was required to accept the plaintiff's allegations as true rather than test them.

• Has a court actually ruled that a chatbot's output is or is not speech, or only that it declined to decide the question yet?
• Which specific claims survived a ruling, and which were dismissed outright?
• Does reporting on a ruling distinguish a procedural motion-to-dismiss stage from a trial verdict?

As a record, the order is significant less for what it decided than for what it declined to decide: a federal court left open, rather than resolved, whether generated conversation counts as speech for constitutional purposes.