Nanny Agency JobsAI can help arrange the wording of two offers into a readable comparison. Give it a narrower assignment than choosing the better job: extract the stated terms, attach each term to its source and preserve unanswered questions. Keep financial calculations and the final decision outside that first pass.
A neat table can hide an assumption just as easily as a paragraph can. NIST identifies confidently presented false output as a generative-AI risk. NIST's Generative AI Profile. For an offer comparison, your practical protection is checking the proposed table against the actual documents rather than trusting its appearance.
Prepare a small, appropriate source set
Use the current written offer and any clearly identified clarification from the agency. Record the version or date of each document. Keep earlier drafts separate so the tool does not combine an old schedule with a newer compensation paragraph.
Before using an AI service, establish that its data handling and your permissions are suitable for these documents. An offer can contain private household or contact information that is unnecessary for comparison. Where appropriate, work from a carefully checked extract with names and identifying details removed. Removing a name alone does not necessarily remove identifying context.
Label the two opportunities A and B. Number the sections or lines in the extract so you can locate each statement again. If you cannot prepare an appropriate source set, make the comparison manually; uploading a confidential document is not a prerequisite for evaluating it.
Ask for extraction before interpretation
Use this original NAJ task instruction: “Using only the supplied offer extracts, list each stated term with its source label and location. Preserve conditions and units. Mark absent information ‘not stated.’ Do not calculate pay, infer benefits, decide legal compliance or recommend an offer. Flag conflicting statements separately.”
Treat the instruction as a request, not a guarantee of reliable output. Inspect each row afterward. Check that the tool retained words such as “proposed,” “subject to agreement” or “after review.” A conditional benefit should not become a confirmed entitlement in the summary.
Keep quoted wording short in your working sheet and link back to the actual document. A longer AI paraphrase can make it harder to see the original condition. Where a clause is unclear, record the uncertainty and formulate a question rather than asking the tool to settle the meaning.
Use a source-linked comparison ledger
Build one row for each decision-relevant term. This original ledger is an audit of the extraction, not a payroll worksheet.
- Term: Schedule, work location, duties, start date, stated pay wording, travel or another relevant item.
- Offer A: Exact stated information or “not stated,” with document and location.
- Offer B: The equivalent information and source location.
- Check: Verified against source, conflicting, missing or incorrectly summarized.
- Question: The specific clarification needed, addressed to the responsible contact.
- Update: Date and source of any answer; previous wording retained separately.
Do not force unlike arrangements into identical cells. One role may describe fixed weekdays while another provides a rotation. Write the arrangements as stated and identify what else you need to understand. A shared heading called “weekly hours” does not justify inventing a standard week for both.
Pay rows should preserve amount, unit, stated basis and conditions. Leave annualization, deductions, overtime treatment and package valuation out of this extraction step. Ask for an appropriate written explanation or qualified advice when those questions affect your decision.
Audit a fictional comparison
Fictional worked example: Sam has two imaginary offers. Offer A specifies a Monday-to-Friday timetable and says travel will be discussed separately. Offer B provides a rotation and includes a paragraph describing travel arrangements. An AI-generated table labels both jobs “travel included” and marks A as having the simpler schedule.
Sam checks the source locations. The first phrase is unsupported: A has not defined its travel arrangement. The second is a preference judgment disguised as a fact. Sam changes A's travel row to “separate discussion required” and removes the schedule ranking.
The ledger now contains two useful questions: “What travel arrangement would apply under Offer A?” and “Which dates and handover expectations apply to Offer B's rotation?” Sam sends each question to the relevant agency without sharing the other family's offer.
No pay arithmetic is needed to improve this comparison. The problem was the loss of uncertainty and the insertion of a preference. Once the agencies respond, Sam can update the relevant rows and decide whether the arrangements fit Sam's own commitments.
Review calculations separately if you need them
If you later need a numerical comparison, first identify the exact question and confirm the inputs. “What amount is stated for this specified period?” is different from “What will I take home?” or “Does this arrangement meet all applicable requirements?” A model-generated formula cannot supply missing contractual or legal facts.
Use a calculation method you can inspect and, where appropriate, a qualified payroll, tax or legal professional. Retain the inputs, assumptions and explanation with the result. A correct multiplication can still answer the wrong question if a package figure was treated as wages or an unstated schedule was inserted.
For the earlier task of reading advertised amounts and units, use NAJ's advertised-pay comparison guide. This workflow concerns the fidelity of an AI summary of your particular offers; it does not establish market rates or recommend a compensation target.
Make the decision from checked terms
Create a short personal list of priorities after the factual comparison is checked. You might value a defined schedule, a particular location or clearly bounded duties. Keep those preferences visibly separate from the offer text so that an AI ranking does not choose their importance for you.
Review any unresolved term that could change your willingness to accept. Request the relevant clarification before treating it as settled, and retain the actual response. If an offer changes, update its source version and recheck the affected rows rather than rerunning a comparison that mixes old and new material.
You are ready to discuss the choice when you can point to the source for each important term and name the remaining uncertainties. The table supports that conversation; it does not make the commitment on your behalf.
Sources
NIST: Artificial Intelligence Risk Management Framework, Generative Artificial Intelligence Profile. The source supports the narrow output-risk statement. The extraction instruction, ledger and fictional example are original NAJ editorial tools.