Elisa answered a rebooking question in under a minute while the human agents on the same call juggled three other conversations and circled back twice for details already given. This says more about which channel, on that day, held up better than about the people on the phones handling the calls.

Lufthansa Group named its digital assistant Elisa in 2019, drawing on the name of the airline's first female flight attendant, alongside SWISS's Nelly and Austrian Airlines' Maria. A human name, chosen on purpose, deployed seven years before any state legislature wrote a rule about chatbot disclosure.

State Disclosure Laws Regulate What Chatbots Say, Not What They Are Named

Fifteen states have enacted chatbot laws as of mid-2026, and ten of those statutes were signed this year alone, according to Bloomberg Law's AI Tracker (Bloomberg Law, 2026). The wave splits into two families. A broader disclosure duty, adopted in states including Utah and New York, requires any AI system talking to a consumer to say so. A narrower, higher-stakes family targets what the statutes call "companion chatbots," systems built to sustain an ongoing, humanlike relationship with a user, and that family carries the steep statutory damages.

None of them regulate the name on the label, and the companion-chatbot laws go further than that. California's SB 243 excludes, by name, any bot used only for customer service, a business's own operations, or technical assistance (California Legislative Information, 2025). Oregon's SB 1546 follows the same companion-chatbot definition and lets a user collect one thousand dollars in statutory damages without proving any harm at all, but only once a bot crosses from transactional support into that sustained-relationship category (Orrick, 2026). Elisa closing my rebooking question in a single exchange does not, on its own, meet that bar. The line that does matter is scope, not name: a support bot that starts remembering a user's preferences and checking in between sessions can drift into the regulated category without anyone renaming it, a gap several law firms have flagged as the harder compliance question (Crowell & Moring, 2025).

The EU Drew the Line Where the US Carved It Out

Article 50 of the European Union's AI Act became enforceable today, following European Commission guidelines published in July that clarify the disclosure duty for chatbots, deepfakes, and AI-generated content (Plumb, 2026). Any system intended to interact directly with a person, including chatbots, conversational agents, and AI companions, must tell that person they are talking to AI. The exclusions run along different lines than the US carve-out: recommender systems, spam filters, authentication tools, and predictive maintenance sit outside Article 50, but a customer-facing chatbot does not (Plumb, 2026). Fines range from seven hundred fifty thousand euros to fifteen million, or as much as three percent of a company's worldwide annual revenue, whichever is higher (Plumb, 2026). The obligation follows the user rather than the company's home address, so a US or Gulf business serving EU customers is in scope the same as one headquartered in Frankfurt.

A separate duty under the same article, labeling AI-generated text, images, or audio, only applies when that content is published, informative to the public, or touches a matter of public interest. The chatbot disclosure duty carries no such condition. Lufthansa serves European customers by definition, which puts Elisa inside Article 50's chatbot rule the same day she sits outside California's and Oregon's companion-chatbot carve-out. The EU rule does not ask whether an interaction was transactional or relational. It asks only whether a person was talking to a machine, and requires the machine to say so.

Disclosure Does Little to Change What Users Think of a Chatbot

A mixed methods study published in AI & Society, combining interviews with an experiment across 194 participants, tested what changes when a customer service chatbot discloses itself versus when it does not, and found the disclosure moved almost nothing that mattered to users: perceptions of friendliness, helpfulness, and whether the interaction felt like a real social exchange showed no significant difference whether or not the chatbot announced its artificial status (van der Goot et al., 2024).

The study's own summary for business readers cuts through the naming debate on its own: what keeps users satisfied is getting help that works, delivered in a friendly tone, not whether the bot discloses its artificial status (van der Goot et al., 2024). Neither a robot-sounding name nor a human-sounding one would have changed that tone. Renaming Elisa "Unit 7" would not have made that exchange faster, and it would not have made the multitasking human agents any less stretched.

The Question Companies Keep Avoiding

shashi.co has covered what happens when the assumption behind escalation breaks in the other direction. A FedEx chatbot kept a customer named Dillon Thompson circling for four months over a missing e-bike, and not one of the four organizations he contacted handed him to a person with authority to open an investigation. An airline's own mobile app caught a lapsed TSA PreCheck credential and flagged it, while the same airline's chat assistant burned twenty minutes on generic policy links because it never saw the data the app already had.

Those two posts describe the AI channel failing while a human or a better-architected system had to compensate.

Elisa outperforming a distracted human agent is the same tension, pointed the other way, and board meetings spend their time on naming in both cases, while scope, the thing shifting under a support bot's feet as features get added, comes up in almost none of them.

Most contact center programs track the companion-chatbot carve-out at launch and few revisit it. A support bot that ships as transactional in January can pick up memory, personalization, and check-in features by summer, the exact features that move it toward the sustained-relationship definition the statutes regulate, without anyone renaming or reclassifying it along the way.