The Best AI Prompt for Chronological Age

Yes — you can get ChatGPT, Claude, or Gemini to calculate a child's exact chronological age in years;months;days for a standardized assessment. Below is the prompt that works most reliably. But be straight about what it can't do: even this prompt still gets some real cases wrong — see where it fails.

The prompt

Replace the two dates, paste it into your AI of choice, and read its working.

You are a precise date calculator. Calculate a child's chronological age for a standardized assessment.

Birth date: [YYYY-MM-DD]
Test date:  [YYYY-MM-DD]

Use the "borrow 30" convention that test publishers (e.g. Pearson Q-global) expect:
1. Subtract day-from-day, month-from-month, year-from-year.
2. If days are negative, borrow 1 month and add 30 days.
3. If months are negative, borrow 1 year and add 12 months.
4. Do NOT round. Do NOT skip a borrow.

Show your subtraction step by step, then give the final answer in the exact format Years;Months;Days  (e.g. 5;3;14).

Why this prompt beats "how old is…"

  • It names the borrow-30 convention. Test manuals (CELF, PLS, GFTA, PPVT, Brigance) use borrow-30, not calendar-accurate math. A casual prompt makes the AI guess which one you want.
  • It forces the AI to show its work. Writing out each borrow reduces silent errors and lets you verify the result instead of trusting it.
  • It gives both dates explicitly. AIs are unreliable about "today's date," so never let it assume the test date — type it in.

Where a general calculator (or AI) gives a different number

Standardized tests don't use ordinary calendar age — they use the borrow-30 convention (every month counts as 30 days when you borrow a day), and that's the value Q-global and the manuals expect. A general calculator, or an AI doing calendar math, returns a different day count. Every date below is real and every value is computed:

BirthTestBorrow-30 (manuals)Plain calendar (AI / general)
Aug 31, 2014Sep 1, 202612;0;012;0;1
Dec 31, 2014Jan 1, 202611;0;011;0;1
May 31, 2018Jun 15, 20268;0;148;0;15
Feb 10, 2020Mar 5, 20266;0;256;0;23
Feb 29, 2024Feb 28, 20250;11;290;11;30

The gaps look tiny — a day or two — but the borrow-30 value is the one the manual expects and the calendar value isn't. Some dates break calendar math outright, and an AI is non-deterministic — ask twice, sometimes get two answers. For a number that sets a child's age band, you want the single deterministic value the manual expects.

→ The full breakdown: why AI and generic calculators get assessment age wrong

Why AI struggles with this specific math

  • It reads dates as text, not numbers. "08/15/2014," "Aug 15 2014," and "2014-08-15" are different token strings to a model; it pattern-matches rather than computes.
  • It has no scratchpad for the borrows. The calculation needs a day→month→year borrow chain; an AI generates one token at a time and tends to skip or misapply it.
  • It isn't deterministic. Ask twice, get two answers, both stated with total confidence — with no way to know which is right without doing the math yourself.

How to stay accurate

  • Double-check any birth or test date near a month-end or Feb 29 by hand.
  • Run the prompt twice and confirm you get the same answer.
  • Read the AI's borrow steps — don't just copy the final line.
  • Or skip all of that and use the free online calculator or the app — deterministic, every time.

FAQ

Can ChatGPT calculate chronological age?

Yes, but not reliably out of the box. With a carefully worded prompt that specifies the borrow-30 method and forces it to show its work, it's right most of the time — but it still errors on edge cases like leap-day birthdays and month-end dates, so its output should always be verified.

What is the borrow-30 method?

It's the convention standardized test publishers use: when the day count goes negative you borrow one month and add 30 days; when the month count goes negative you borrow one year and add 12 months. It deliberately does not use the real length of each calendar month, which is why a "calendar-accurate" AI answer can disagree with what a test manual expects.

Why does a one-month error matter?

Norm-referenced assessments score children in age bands that are often only three to six months wide. Landing in the wrong band changes the standard score and percentile, and can change whether a child qualifies for services.

What's the most reliable way to calculate it?

A deterministic calculator built for the borrow-30 method — like the free Speechies Chronological Age Calculator — gives the same correct answer every time and removes the need to verify an AI's output.

← Back to the Speechies Chronological Age Calculator