Ask ChatGPT "will my money last if I retire at 55?" and you will get a confident, well-written answer. What you will not reliably get is a calculation. Out of the box, ChatGPT reasons about retirement the way a well-read friend would: it picks a withdrawal rule of thumb, assumes a return, and works the arithmetic out in its head. It can write a Monte Carlo script in its code tool if you push it, but it has to invent the return model and the tax rules on the spot, and the next conversation starts from scratch.
That is the gap the Retirement Lab app in ChatGPT fills. You still ask in plain language. The difference is that ChatGPT hands your inputs to Retirement Lab's simulation engine, the same one behind the website, which runs a 10,000-path Monte Carlo simulation with bracket-based tax modeling for your country of tax residence and returns what the simulation shows: a failure probability, a range of ending balances, and year-by-year percentile bands. The result is saved to your Retirement Lab account, so you can open it on the website later and keep working on it.
This guide covers how to connect the app, prompts you can paste, and what the app does and does not do.
Why ChatGPT on its own is not a retirement calculator
A large language model is very good at explaining the 4% rule, the difference between a Roth and a traditional IRA, or why sequence-of-returns risk matters. It is much weaker at the things a retirement projection actually depends on:
- Arithmetic over decades. A 40-year projection compounds returns, inflation, contributions, withdrawals, and taxes every year. Doing that "in its head," a language model can drift, round inconsistently, or quietly drop a step.
- Randomness done properly. A single average return hides the risk that matters most: a bad decade early in retirement. A Monte Carlo simulation answers that by running thousands of different market paths and counting how many run out of money. A chat reply that says "historically, portfolios like this have usually lasted" is a summary, not a simulation.
- Tax tables. Withdrawals are taxed differently in the US, the UK, Portugal, or Spain, and the rules change. ChatGPT does not carry a maintained, bracket-by-bracket tax model for each country, so its after-tax numbers are estimates at best.
- Memory between sessions. Each new chat starts fresh. There is no saved scenario to come back to, compare against, or share.
None of that makes ChatGPT less useful for thinking out loud about retirement. It just means the numbers are better produced by something built to produce them.
How to connect Retirement Lab to ChatGPT
You need a Retirement Lab account; a free account works, and anonymous access is not available through ChatGPT.
- Open the Retirement Lab ChatGPT setup page and select Install in ChatGPT, or go straight to the Retirement Lab listing in ChatGPT.
- Install it, then connect your Retirement Lab account when prompted. You sign in and approve access in your browser.
- Start a chat with @Retirement Lab and describe your situation.
The app asks for anything it is missing, such as your age, savings, annual spending, or country of tax residence, before it runs anything. Calls use your account's tier and its current simulation quotas. You can disconnect ChatGPT at any time from Connected apps in your Retirement Lab account.
If you use Claude, Gemini, or another client that supports remote MCP connectors, the same setup page has instructions for those too.
Prompts to try
Paste any of these into a chat with @Retirement Lab and swap in your own numbers. Each one asks for a hypothetical projection based on the inputs you give it.
Retiring early at 55
I'm 42, live in the US, have $850,000 invested, and save $25,000 a year. I'd like to stop working at 55 and spend $90,000 a year after that. Run a simulation to age 95 and tell me the failure probability and the range of ending balances.
Will my money last?
I'm 63 and retired in the UK with £600,000 in a pension and £150,000 in an ISA. I spend £38,000 a year and start my State Pension at 67. What does the simulation show for how likely the money is to last to age 95?
Comparing countries
Same household as before: $1.2M saved, $70,000 a year of spending, retiring at 60. Compare the simulated outcomes if my tax residence is Portugal versus Spain.
Each country in a comparison is a separate hypothetical projection with the same inputs under that country's modeled taxes. Free accounts compare up to two countries at a time; a comparison with three or more countries, such as Portugal versus Spain versus the US, requires Pro or a 30-day Pro pass. On a free account you can still run the third country as its own comparison.
Previewing a Roth conversion's tax
I'm 58, married filing jointly in the US, with $1.1M in a traditional IRA and $40,000 of other income this year. Preview the tax on converting $60,000 to a Roth this year.
A conversion preview is a quick tax calculation for one year, based on your inputs. It does not run a full simulation, and it does not say whether converting is a good idea for you.
Testing spending levels
Using my saved scenario, test annual spending of $80,000, $90,000, $100,000, and $110,000, and show the simulated success rate for each. Which spending levels keep simulated success above 90%?
This runs a sensitivity sweep: one simulation per spending level, side by side. The answer describes what the simulation shows at each level you asked about, given these assumptions; it is not a safe spending number. Free accounts can test a limited number of values per sweep.
Coming back to a saved scenario
Open my saved retirement scenario and compare it with a version where I retire two years later.
ChatGPT can list your saved scenarios, create a variant, and compare the two, without overwriting your original unless you ask it to.
What the app does and does not do
It does:
- Run Monte Carlo simulations, with 10,000 simulated market paths on a free account, using bracket-based tax modeling for your country of tax residence.
- Report the failure probability, percentiles of the ending balance, and year-by-year percentile bands the simulation produces.
- Compare countries, test one assumption across several values, and preview the tax on a Roth conversion or a relocation exit tax.
- Save runs to your Retirement Lab account. Each completed run comes with a link to its record on the website, where you can inspect every assumption and change the inputs. A public share link is only created if you ask for one.
It does not:
- Pick stocks, funds, or allocations for you, or tell you what to buy or sell.
- Use live market data. Returns are drawn from modeled distributions, not today's prices.
- Give financial, tax, investment, or legal advice. Every result is a hypothetical projection based on your inputs and the model's assumptions, and the methodology page explains those assumptions and the tax model's limits.
- Make ChatGPT's own reasoning more accurate. When ChatGPT adds commentary around a result, the numbers from Retirement Lab are the simulation output; the commentary is ChatGPT's.
A note on privacy
When ChatGPT calls Retirement Lab, the simulation inputs and results are stored under your account like any other scenario you create on the website, and they follow the same retention terms. For measuring how the app is used, each tool call records an internal user id, the tool name, the result status, your tier, and a coarse authentication method. Retirement Lab does not store your prompt text, email address, raw credentials, or financial payloads for that measurement.
Your conversation itself happens in ChatGPT, so what you type there is handled by OpenAI under its own terms. Retirement Lab receives the inputs ChatGPT sends to the simulation, not the rest of the chat. The Privacy Policy has the full details.
Getting started
Connect the app from the setup page, then ask your first question in plain language. If you would rather fill in the inputs yourself, the same simulation runs on the Retirement Lab website, no account needed.