Nanny Agency JobsAI can organize reference-check preparation and help draft a summary of information actually collected. It cannot turn an unanswered request into a reference, establish a person's identity from a polished paragraph or supply firsthand knowledge of a nanny's work. The central question is not whether the summary sounds convincing. It is what evidence supports each statement.
For an agency, that means keeping the source conversation or response distinguishable from the later interpretation. This guide proposes a practical evidence record for that work. It does not replace the agency's reference process, applicable requirements or professional judgment about a candidate.
Separate the claim from the person who can address it
Begin with the specific work-related question. Are you checking employment dates, understanding responsibilities or asking for an example of communication? These are different tasks. A person may be able to confirm a date without having observed the nanny's daily work.
The US Office of Personnel Management describes reference checking as collecting information from people who know an applicant's work, and discusses structured, job-related questions. Its guidance concerns federal hiring; it does not establish nanny-agency requirements. OPM: Reference Checking.
For each proposed source, record their stated relationship and the period they can discuss. Keep “candidate supplied this contact” separate from “agency established the contact's identity through its process.” Neither a familiar email signature nor a detailed answer automatically resolves identity.
Give automation preparation and organization tasks
A bounded use might be turning an approved question list into a consistent worksheet, identifying missing fields in collected notes or drafting a summary with links back to the recorded answers. Test those tasks with fictional material before using any real reference information.
Do not ask a model to write the reference you expect to receive. Do not manufacture quotations, reconstruct a call that never happened or strengthen a cautious answer into an endorsement. An unanswered question stays unanswered even if the résumé makes an answer seem likely.
Contact methods require their own decision. A tool capable of placing calls or sending messages should not be enabled merely because drafting worked. Decide who authorizes contact, how participants are informed, which channel is appropriate and what information may be recorded or processed. An agreed reference conversation does not automatically authorize every additional technology.
Use an evidence ledger for each material statement
This original ledger is a review tool, not a reference scoring system. Create one line per claim that may appear in the final account:
- Question or issue being checked.
- Source person's stated relationship and relevant period.
- Contact and identity-check status under the agency's procedure.
- Date, channel and location of the actual response.
- What was explicitly reported, preserving qualifications.
- Whether the note is a direct quotation or a paraphrase.
- Unresolved discrepancy and the person assigned to follow up.
For example, “reported that the nanny often arrived early” and “verified punctuality throughout employment” are not equivalent. The latter can imply a breadth of checking the source never provided. Keep the wording proportionate to what was actually said.
Store the underlying material according to the agency's information-handling rules. The ledger should point authorized reviewers to the evidence, rather than creating uncontrolled copies of phone numbers, household details or recordings.
Review summaries for missing limits and invented meaning
Read the proposed summary beside the source record. Check names, dates, duties and qualifiers. Look for a change from “sometimes” to “consistently,” from “I did not observe that” to “there were no concerns,” or from a question left blank to a positive conclusion.
Do not infer dishonesty from a discrepancy alone. Different people may remember dates differently or describe the same responsibility from different perspectives. Record the conflict and seek clarification through the agency's process before presenting a resolved account.
Avoid using sentiment, enthusiasm, response speed or an AI-generated personality label as a substitute for job-related evidence. A short reply does not tell you why the person was brief. A delayed response does not establish anything about the candidate's conduct. The record should show the actual answer and its limits.
Work through a fictional ambiguous answer
Fictional example: An agency asks former employer Morgan whether candidate R18 regularly handled school transport. Morgan replies that R18 occasionally collected the children when the parents were delayed, but that another adult normally handled the school run.
An automated summary says, “R18 managed daily school transport independently.” The reviewer rejects it. The corrected summary says that Morgan reported occasional collections and that the usual arrangement involved another adult. The agency retains a question about the scope of R18's transport experience instead of treating it as settled.
A second requested reference has not responded. The system must not combine the first answer with résumé claims and label both references complete. The record shows one response collected and one outstanding request, with any identity checks separately documented.
If the family-facing account needs a concise version, the recruiter decides what is appropriate to share. Shortening the account must not remove the distinction between occasional and regular duties. The example illustrates evidence discipline, not a recommendation to accept or reject R18.
Define completion before marking the check finished
Choose the completion conditions in advance: the required contact process, the questions addressed, the evidence retained and the review performed. A generated summary can be finished while the reference check remains incomplete. Use separate fields for those states.
When an issue cannot be resolved, name it. The agency's responsible person must decide the appropriate next step under its process, rather than letting a model silently fill the gap or assign an unfavorable score. Keep any candidate clarification distinguishable from the reference's own account.
Before relying on the final summary, a reviewer should be able to answer: who supplied this information, what did they actually observe, where is the response and what remains unknown? If those answers are missing, better prose will not repair the evidence.
Sources
OPM: Reference Checking, checked October 27, 2026. Used for a narrowly attributed description of reference evidence and structured questions. The ledger, review process and fictional case are original NAJ editorial tools.