Nanny Agency JobsAn agency may notice wording that raises a question, use a detection service or receive information about how a document was produced. Those are different kinds of evidence. A polished paragraph does not establish its authorship, and a detector label needs context before anyone treats it as a finding about a candidate.
The useful question is what the particular evidence supports. You can respond to an authorship concern honestly while asking the agency to distinguish a software prediction, a supported provenance signal and a factual problem in your application. This guide concerns that evidence, rather than whether you should disclose assistance or why an application was rejected.
Separate suspicion, classification and provenance
A recruiter saying that an answer sounds generic is expressing an observation. A classification service assigning an AI label is producing a model result. Provenance concerns information associated with a document's origin. None of those descriptions automatically tells you whether an experience claim is true or whether a particular application instruction was followed.
There is a useful historical caution: OpenAI withdrew its own text classifier in July 2023 because of low accuracy. That describes a discontinued system, not a measurement of today's detection products. OpenAI's classifier announcement.
Avoid both confident extremes: “They can always tell” and “There is no way anyone can know.” Ask which system, document and result are actually involved. If the agency has not said it uses a detector, do not turn your suspicion into a description of its process.
Read a result according to its actual scope
As checked on October 27, 2026, OpenAI's provenance guidance describes text watermarking for ChatGPT users in the EU and optional API text watermarking globally. It describes restricted text-detector access for qualifying research uses, with general business applications outside that program's scope. Those statements do not establish that an agency can run every résumé through an OpenAI text-verification service. OpenAI's provenance guidance.
The same guidance cautions that provenance does not establish accuracy, ownership or the creator's identity. A supported origin signal answers a narrower question than “Did this candidate deceive us?” Those distinctions matter when assistance involved only editing, translation or formatting and the experience itself remains the candidate's.
For any named detection product, request its relevant explanation rather than interpreting a percentage yourself. A displayed number might have a product-specific meaning. Do not assume that “80%” means 80% of the words were written by AI or an 80% probability that you broke a rule.
Use an evidence-request checklist
NAJ suggests this original checklist when an agency raises a concern. You do not need to demand a technical investigation before every interview; use the questions that address the stated issue.
- Material: Which submitted document, version and passage prompted the concern?
- Method: Was this a reader's impression, a named classifier or an identified provenance check?
- Meaning: What does the provider say the particular result means?
- Instruction: Which application rule is the agency asking you to address?
- Facts: Is any date, duty, qualification or example disputed separately?
- Response: What explanation or correction would help the responsible reviewer?
Keep the agency's answer beside your submitted version. If the answer remains incomplete, write “method not explained” in your notes. That records a limit without concluding either that the result is correct or that the agency acted improperly.
Work through a disputed result without inventing proof
Fictional worked example: Robin used an editing tool to shorten a paragraph about school pickups. The agency says a screening result flagged the paragraph. Robin has the original notes and final submitted version, both describing work Robin actually performed.
Robin first asks which passage and application instruction are at issue. The agency identifies the paragraph but does not explain the software's scale. Robin records the method as unconfirmed and responds: “I used an editing tool to shorten this paragraph. The pickup responsibility is mine, and I can explain the routine and my part in it. Please let me know which instruction or factual detail you would like me to address.”
The example does not establish how any real agency handles these questions. It shows why authorship, application instructions and experience verification belong in separate lines. Robin's notes may help explain the process; they are not a guaranteed certificate of independent authorship.
Keep your explanation accurate and proportionate
Describe the help you actually used. “Spelling corrections,” “reorganized my notes” and “generated a first draft” are different explanations. Do not minimize substantial assistance or claim that a document history proves more than it shows. Never create backdated drafts or fictional handwritten notes to manufacture evidence.
You can offer a fresh, truthful explanation of your work without exposing a former family's private details. If the agency requests confidential records as proof, clarify what information it needs and use its appropriate submission route. An authorship question is not a reason to upload the same sensitive application into several additional detectors.
If the original instructions required independent writing, address that requirement directly. A debate about a detector's limitations does not resolve whether you followed the instruction. If you made an inaccurate claim, correct the claim regardless of how it came to the agency's attention.
Choose a useful next step
Return to the evidence checklist and identify the unresolved question. You might need an explanation of the result, clarification of the rule, or a correction to an actual statement. Send one focused response to the responsible contact and retain what you sent.
Avoid rewriting your application to chase a supposedly human score. The goal is a truthful application you can explain, with a clear account of any assistance and respect for the agency's instructions. The agency still owns its review process; no checklist can promise reconsideration or selection.
Sources
OpenAI's historical classifier announcement and current provenance guidance. Provider statements support the narrow descriptions above. The checklist and fictional example are original NAJ editorial guidance.