Names are practical, not deceptive
A name makes it referable — staff can say what handled a ticket, customers know what they were talking to, and your logs are readable. None of that requires anyone to believe it is human.
The line is implied humanity
A name paired with a clear statement that this is an AI assistant is honest. A name paired with a stock photograph, a job title, and a location is a fabricated colleague, and it crosses from convenience into deception.
Name it if it helps. Just never let anything imply a person exists who doesn’t.
Pick something that doesn’t oversell
Names suggesting seniority or expertise set expectations the system will not meet. Something neutral avoids the disappointment when it escalates — which it should, before the customer asks.
Keep the disclosure attached to it
Wherever the name appears, the description should too. That is the practical form of saying plainly what customers are talking to, and it survives people forwarding a transcript.
One name per role, not per deployment
If you run several workflows, naming each separately confuses customers and staff. Name the role, not the instance — the same reasoning as defining shared identity once when scaling.
It affects tone expectations
A heavily branded persona invites a chatty register that ages badly and grates during a complaint. A plain name supports the neutral, competent voice that holds up over hundreds of interactions.