Blog · · 7 min read

Ask your AI to show its working

#The situation

You're going freelance in January and you need an hourly rate. So you ask an AI assistant, and you give it what you know: you want to take home about €48,000 a year, your accountant said to set aside roughly 30% for tax and social contributions, your costs are about €500 a month, you want six weeks off, and not every hour of a 40-hour week will be billable.

The answer comes back fast and sounds sure of itself: about €70 an hour is break-even, charge €75 to €85. It even has a small table.

And then you have questions it doesn't answer. Where did 70 come from? Did it take the tax off before or after the costs? How many weeks did it count? What did it assume about billable hours, given that you never said? Next March, when the accountant says 35% instead of 30%, which part do you change?

A number in a chat is an answer you have to trust. The working behind it is an answer you can check, and the place to keep the working is a file.

#Ask the AI

Paste this after the question, or as part of it. It works in ChatGPT, Claude, Gemini, or an agent that can write files directly, and for any calculation you'd want to defend later: a rate, a budget, a loan, a break-even, a pricing model.

Before you give me a number, write it up as a file I can check and keep.

Name it: <topic-slug>-<today's date YYYY-MM-DD>.md

The file:
- Frontmatter: title, date, question (the question in one line), answer (the
  result in one line, with its unit).
- A # heading, then the answer in one sentence inside a > [!note] callout.
- "## Inputs": a table of every number you used: name, symbol, value, unit,
  and where it came from (what I told you, or "assumed" with the reason). If
  I didn't give you a number you need, assume a sensible one and mark it
  "assumed". Never hide an assumption inside a step.
- "## Working": one short step per heading (### 1., ### 2., ...). Each step
  has one sentence of plain English, then the formula in symbols as display
  math ($$ ... $$), then the same formula with the numbers put in and the
  result. Keep units in every line.
- "## What moves the answer": a table showing the result when each of the
  two or three inputs that matter most goes up or down by a realistic amount.
- "## Check": redo the final result a second way (backwards from the answer,
  or a quick estimate) and say whether it matches.
- If one assumption could change the decision on its own, flag it in a
  > [!warning] callout.

Rules: math as LaTeX between $...$ (inline) or $$...$$ (display). Never use
the $ sign for money; write currency as a code or symbol (EUR, €, USD). Round
only at the last step, and say how you rounded. No advice beyond what the
numbers show.

Reply with the filename only, then the file in one code block fenced with
four backticks.

Four lines do most of the work.

"Never hide an assumption inside a step." This is the line that matters. You didn't say how much of your week is billable, so the assistant has to pick a number. Without this rule it picks one quietly, and the quiet guess ends up driving the whole answer. With it, the guess sits in the inputs table, marked assumed, where you can see it and replace it.

"The formula in symbols, then with the numbers put in." The symbols tell you what the step means; the numbers let you redo it on a calculator. Most arithmetic slips are visible the moment both are on the page.

"Check: redo the final result a second way." Working backwards from the answer catches a step done in the wrong order, like taking tax off the costs as well as the profit. It costs the assistant three lines.

"Never use the $ sign for money." In Markdown, $ is how math begins. A file that mixes dollar prices with formulas can trip over its own text, so the prompt keeps money in codes or symbols and leaves $ to the math.

#What comes back

We tried it in a fresh chat: first the freelancer's question on its own (the €70 answer above), then the prompt as a follow-up. The assistant replied with the filename, freelance-hourly-rate-2026-10-12.md, and this file, shown here in full.

---
title: Freelance hourly rate for a 48,000 EUR take-home
date: 2026-10-12
question: What hourly rate must I charge to take home 48,000 EUR a year after tax as a full-time freelance product designer?
answer: Break-even rate of about 69 EUR/hour at 60% billable time (about 76 EUR/hour with a 10% buffer), before VAT.
---

# Freelance hourly rate

> [!note]
> To take home 48,000 EUR a year with 6,000 EUR of costs, 30% tax and 1,080 billable hours, you need to charge about 69 EUR/hour, or about 76 EUR/hour with a 10% buffer.

## Inputs

| Name | Symbol | Value | Unit | Source |
|---|---|---|---|---|
| Target take-home | $N$ | 48,000 | EUR/year | You told me |
| Tax and contributions rate | $t$ | 0.30 | share of profit | You told me (your accountant's estimate) |
| Business costs | $c_m$ | 500 | EUR/month | You told me |
| Months per year | $m$ | 12 | months/year | Calendar fact |
| Weeks in a year | $w_y$ | 52 | weeks/year | Calendar fact |
| Weeks off (leave and public holidays) | $w_o$ | 6 | weeks/year | You told me |
| Weeks sick | $w_s$ | 1 | weeks/year | You told me ("a week or so") |
| Hours per working week | $h$ | 40 | hours/week | You told me |
| Billable share of working hours | $b$ | 0.60 | share of hours | Assumed: you gave no figure. Admin, sales and invoicing commonly take 30-40% of a freelancer's time, and the first year has more sales work. |
| Safety buffer on revenue | $u$ | 0.10 | share of revenue | Assumed: you gave no figure. It covers gaps between projects, overruns and late payment. |

All amounts exclude VAT.

## Working

### 1. Profit needed before tax
Take-home is profit after tax, so divide the target by the share of profit you keep.

$$P = \frac{N}{1 - t}$$

$$P = \frac{48{,}000\ \text{EUR/year}}{1 - 0.30} = 68{,}571.43\ \text{EUR/year}$$

### 2. Annual business costs
Monthly costs times twelve.

$$C = c_m \times m$$

$$C = 500\ \text{EUR/month} \times 12\ \text{months/year} = 6{,}000\ \text{EUR/year}$$

### 3. Revenue needed
Revenue must cover the profit you want plus the costs.

$$R = P + C$$

$$R = 68{,}571.43\ \text{EUR/year} + 6{,}000\ \text{EUR/year} = 74{,}571.43\ \text{EUR/year}$$

### 4. Billable hours available
Subtract the weeks off, multiply by hours per week, then keep only the billable share.

$$H = (w_y - w_o - w_s) \times h \times b$$

$$H = (52 - 6 - 1)\ \text{weeks/year} \times 40\ \text{hours/week} \times 0.60 = 1{,}080\ \text{hours/year}$$

### 5. Break-even hourly rate
Divide the revenue needed by the billable hours.

$$r = \frac{R}{H}$$

$$r = \frac{74{,}571.43\ \text{EUR/year}}{1{,}080\ \text{hours/year}} = 69.0476\ \text{EUR/hour}$$

### 6. Rate with buffer
Raise the revenue target by the buffer, then divide by the same billable hours.

$$r_u = \frac{R \times (1 + u)}{H}$$

$$r_u = \frac{74{,}571.43\ \text{EUR/year} \times 1.10}{1{,}080\ \text{hours/year}} = \frac{82{,}028.57\ \text{EUR/year}}{1{,}080\ \text{hours/year}} = 75.9524\ \text{EUR/hour}$$

Rounding: nothing was rounded until the end. The results are rounded to the nearest whole EUR/hour: 69.0476 becomes 69, and 75.9524 becomes 76. Intermediate values are shown to 2 decimals for readability only.

## What moves the answer

Each row changes one input and keeps the others as above. Rates are the break-even $r$ (no buffer), rounded to the nearest whole EUR/hour.

| Input changed | Value | Billable hours (hours/year) | Break-even rate (EUR/hour) |
|---|---|---|---|
| Billable share $b$ | 0.50 | 900 | 83 |
| Billable share $b$ | 0.70 | 1,260 | 59 |
| Tax rate $t$ | 0.25 | 1,080 | 65 |
| Tax rate $t$ | 0.35 | 1,080 | 74 |
| Working weeks (weeks off plus sick) | 44 weeks (one more week off) | 1,056 | 71 |
| Working weeks (weeks off plus sick) | 46 weeks (one less week off) | 1,104 | 68 |
| Base case | $b = 0.60$, $t = 0.30$, 45 weeks | 1,080 | 69 |

The billable share moves the result most: 10 percentage points either way changes the rate by about 14 EUR/hour in one direction and 10 EUR/hour in the other.

## Check

Work backwards from the rate. Charge 69.0476 EUR/hour for 1,080 hours:

$$69.0476\ \text{EUR/hour} \times 1{,}080\ \text{hours/year} = 74{,}571.4\ \text{EUR/year}$$

$$74{,}571.4 - 6{,}000 = 68{,}571.4\ \text{EUR/year (profit)}$$

$$68{,}571.4\ \text{EUR/year} \times (1 - 0.30) = 48{,}000\ \text{EUR/year (take-home)}$$

Quick estimate: 48,000 ÷ 0.7 is about 68,600 EUR. Add 6,000 EUR to get about 74,600 EUR, then divide by about 1,080 hours to get about 69 EUR/hour.

Both methods match the result in step 5.

> [!warning]
> The billable share ($b = 0.60$) is my assumption, not yours, and it alone decides whether the rate is about 59, 69 or 83 EUR/hour. Replace it with your own estimate before relying on the result. The 10% buffer ($u$) is also assumed, but it only affects the buffered rate.

We checked every figure by hand, including the table, and they're right.

It's a better answer than the chat one, and not because the number changed much (69 instead of "about 70"). It's better because now you can see why.

The weakest input is named. The warning says, in plain words, that the billable share is the assistant's guess and that it alone swings the rate between 59 and 83 euros an hour. In the chat, the same guess was one row of a table, with no flag that it was a guess. In the file, it's the first thing you'll want to replace with your own number.

The order of operations is on the page. Step 1 takes tax off profit; step 3 adds the costs back on top. If your accountant does it differently, you'll see exactly which line to argue with.

It knows what it doesn't know. VAT is excluded and says so. The rounding is stated. The check confirms the answer two ways instead of asserting it.

One thing to change for your own use: the frontmatter answer line carries two numbers. If you'll compare many of these files later, ask for one number there and keep the buffer in the body.

#Save it as a file

Save it as freelance-hourly-rate-2026-10-12.md in a money folder, and connect that folder in MD Flow (what browsing a folder of Markdown looks like).

#Why that matters next week

The formulas read as formulas. Open the file in a plain text editor and step 5 is \frac{74{,}571.43\ \text{EUR/year}}{1{,}080\ \text{hours/year}}. In MD Flow it's a fraction again, the way it looked in the chat: KaTeX math renders inline and on its own line, the inputs are a table, the answer and the warning are coloured callouts, and the frontmatter becomes a header with the question and the answer at the top. On a Mac, pressing Space on the file in Finder shows the same rendered preview without opening anything.

Change an input and read the result. When the accountant says 35%, open the file in your editor, or hand it back to the assistant with "t is now 0.35, update the file." MD Flow doesn't do the arithmetic, but its preview follows each save, so the new working is on screen the moment the file changes.

It survives being sent. With Pro, export the file as a PDF and the math survives the page: fractions, tables and callouts included. That's the version for an accountant, a partner or a client who has never heard of Markdown and shouldn't have to.

The folder becomes a record of your decisions. A year from now you'll want to know what you assumed when you set this rate, and whether it held. The file has the date in its name and every assumption in a table. Search by filename is free in MD Flow; with Pro, search also reaches inside the files, so billable finds every calculation that leaned on it, with the line and its number.

The chat gave you a number. The file gives you the reasoning behind it, which is the part you'll need again.