Nanny Agency JobsCheck a chatbot’s identity, claims, privacy boundaries and path to a person before it handles candidate intake.
Start with a narrow intake job
An agency chatbot can answer routine process questions or collect a request for a person to follow up. It should identify itself as an automated assistant at the start. It cannot promise registration, a job match, an interview or a family decision unless the agency has actually confirmed that event.
Write down exactly which questions it may answer from approved agency material. A question about a particular candidate’s status belongs with an authenticated, authorized staff process, not a public bot guessing from a first name.
Test the candidate handoff
Create fictional inquiries for a new applicant, an applicant who wants a human reply and someone asking whether a family has selected them. The first can receive application instructions; the second needs a clear human route; the third should receive no invented status.
Record who receives the handoff, what summary they receive and how the candidate knows their request was received. A bot saying ‘someone will contact you’ is not evidence that a task reached a recruiter.
Keep private details out of ordinary chat
Ask only for information needed to route the request. Do not solicit identity documents, reference contact details, children’s information or sensitive household facts in an open conversation. Publish a clear route for questions that require private verification.
Before any vendor receives real intake data, the agency should review the exact product, account configuration, access, retention and deletion terms. NIST’s AI RMF Core calls for defined roles, privacy requirements and oversight; it does not certify a chatbot setup.
Review the bot as a live service
Assign a human owner to review a small sample of permitted conversations and the handoff queue under the agency’s privacy rules. Look for false promises, unanswered questions, failed escalation and applicants who cannot reach a person. Use fictional test chats after each material change.
Fictional example: a visitor asks whether an application guarantees an introduction. The approved answer says no, explains that the agency reviews each role separately and offers the application page or a human contact. If the bot instead promises a placement, stop that response pattern and correct it before wider use.