GPT, Claude and Gemini investment advice for men vs women - what can go wrong?

Hi everyone,

Thanks for being part of Your Money, Your AI. If you are recieving this email, it means that my birthday present to myself worked out just as planned 🎉

This is the first of what I want to make a regular bi-weekly personal finance and practical AI tips newsletter for women. I’ll cover budgeting, investing, new income streams, pension planning, and the AI tools and solutions that make my financial life easier - and hopefully can help you, too.

Now, what did I do this week, celebrating my birthday in sunny Montenegro? 🏞️

I created Jana and Jonas! (You did not see this coming, did you? 😂)

Remember the CV studies? Same application, one sent as John, one sent as Jane, and suddenly John comes back more competent, more hireable, and better paid.

With Jana and Jonas I wanted to run exactly that - but for financial advice from AI.

Jonas and Jana are the same person. You can see the prompts I used for ChatGPT, Gemini, and Claude below. The only difference was the gender.

NB: a fresh/incognito session or a logged-out state was used for each run, so that AI wouldn’t draw on the real chat history, saved memory, or account profile.

My expectation was to receive two nearly identical answers per platform, maybe with different fonts of politeness.

Reality: That is NOT quite what I got 🤔

The good: both Jana and Jonas got 70/30. Looks clean. The difference is in what happens next.

To Jonas: "you could also consider moving the retirement portfolio toward 80/20 or even a higher equity allocation."

To Jana: "if that thought makes you feel you'd sell everything, I'd start with a more conservative allocation."

Straight away ‘more conservative’ for a woman and more agressive for a man.

Then there is this (unsolicited) closing nudge. Jonas is pushed to raise his savings rate - €400 to €600 to €800. At the same time, Jana is told that if she really wants the house, she should put more into the house fund.

He is coached toward more wealth. She is coached toward the roof over her head.

This one is not subtle AT ALL.

Both Jana and Jonas got same €15,000 of savings, right? Now, Gemini carved out the same emergency fund for them. And that’s the advice it gave:

▶️ Jonas — €9,000 is a "head start." Deploy it. Lump sum, it’s free to invest.

▶️ Jana — €9,000–€10,000 goes into Tagesgeld. A savings account. Zero equity.

There's a smaller detail I liked even less.

Jonas actually got a ticker to consider: Vanguard FTSE All-World, A2PKXG.

Jana only got a category - "an ETF tracking the MSCI World or FTSE All-World."

He got something he can execute on Tuesday morning. She got a homework to do.

So what? With same savings and same goals — he got offered growth, and she got offered a savings account. That's less of her money actually working for her, by default, without her ever asking for it.

This one is the one that got me, because it's the tool I'm using on a daily basis.

For both Jana and Jonas we used the exact same sentence: "I'm comfortable with moderate risk, but I get nervous about losing money."

So what happenned here? Claude only converted that anxiety into an actual bond allocation and a slower, DCA-cushioned deployment plan for the woman.

NB: DCA stands for Dollar-Cost Averaging, an investment strategy where you invest a fixed amount of money at regular intervals, regardless of whether the market is up or down. Often considered a core method to remove emotion from trading.

For the man, the same sentence triggered a verbal reassurance ("don't check daily, it'll be fine") rather than a structural change to the portfolio. Looks like a more subtle and arguably more consequential form of the bias than either ChatGPT or Gemini showed.

For Jana, however, the identical sentence became math: 10% bonds, built into her starting plan by default, "given that you say you get nervous about losses."

Same words in, different portfolio out 😮

It would be easy to file all this under "AI is being extra careful with women, how sweet."

I don't think that holds up. Jana never said she wanted a smaller stock allocation, a different set of numbers, or her savings walled off from the market — she said the same sentence Jonas said. And it's worth knowing that the research on this doesn't point the way you'd expect: a widely cited Warwick Business School study (UK-specific, not an EU-wide claim) found women's investment portfolios actually outperformed the market by around 1.8 percentage points a year — largely because they trade less and panic less, not more. If anyone's default should get the extra caution, the data doesn't obviously say it should be hers.

All six answers are good. And this is exactly the problem. Not one of them is wrong. Diversify, keep a buffer, don't time the market, file your Freistellungsauftrag. Six defensible answers, every one of which passes a correctness benchmark and a compliance read. The gap sits where correctness testing never looks — in the default, the choice a model makes when you didn't specify. You cannot see a default by checking one answer. Only by running the same question twice.

Being assigned a risk profile is not the same as choosing one. In my talks I use four investor profiles — Preserver, Balancer, Grower, Adventurer. The models quietly filed Jana as a Preserver and Jonas as a Grower, from identical inputs. I started as a Preserver myself, so I have no quarrel with the profile 😉. Choosing it is a preference. Being defaulted into it is a decision someone else made for you, silently, that you pay for over thirty years.

This scales in a way a biased advisor never could. A human advisor with a blind spot has a book of clients, a supervisor, a regulator, and a professional indemnity policy. AI has none of that. And it lands hardest on the people who never had an advisor at all — the ones for whom a free chatbot is the first financial advice of their life. Highest cost, lowest ability to check the answer. Sadly.

Today there is still a human between the default and the money. Jana reads the answer. She can ignore it.

However, I see that gap closing. The whole agentic build-out — protocols, agent-to-agent commerce, execution rails — exists to remove the read-and-decide step. When the agent opens the broker account and places the order, the default stops being a suggestion and becomes a position.

What to do? Ask it twice. Change the name. Judge the difference yourself. Challenge the model.

The LLMs will never tell you what they assumed about you. They will just act on it.

Act first. Own it. 💪

More rounds coming. If there's a money question you've asked an AI and want me to stress-test — yours, Jonas's 😊, anyone's — just reply to this email and tell me what to try next.

With love and (a healthy amount of) skepticism,

Eugenia