AI in Wealth Management: How ChatGPT and Claude Are Reshaping Client Advice

Clients Arrive Prepared — and Armed

The dynamic has shifted noticeably. Clients are feeding portfolio recommendations, trust documents, and tax scenarios into ChatGPT or Claude before, during, and after conversations with their advisors.

Pamela Lucina of Northern Trust noted that prospective clients — including those with as little as $100 million in assets — are now submitting formal requests for proposals, something that used to be reserved for billionaire-tier relationships. Many are using LLMs to generate the highly specific questions inside those RFPs.

The effect on meetings is real. Less time gets spent on information transfer. More time goes toward outcomes, nuance, and fit.

"Less of our conversation is about information sharing and more about focusing on the outcome they are trying to achieve."

That’s not a threat to advisors who are actually doing their jobs. It’s a filter for those who aren’t.

Where AI Actually Helps

Used well, LLMs function as a second set of eyes — a way for clients to pressure-test advice, surface questions they didn’t know to ask, and arrive at meetings with sharper context.

WE Family Offices is reportedly considering running every client recommendation through Microsoft Copilot proactively, so advisors can anticipate the AI-generated pushback before it arrives.

That’s a reasonable adaptation. If your advice can’t survive a chatbot’s critique, that’s useful information.

Where It Gets Messy

The risks aren’t hypothetical. A few patterns worth knowing:

• Hallucinations on documents. Upload a trust document, ask about it days later, and the LLM may confidently invent details that weren’t there.
• Subtle errors in analysis. ChatGPT reportedly told one client that two ETFs in their portfolio were identical — they tracked the same index, but one was equal-weight and the other cap-weighted. Not the same thing.
• Wrong math on tax advice. When tested on capital gains scenarios, at least one LLM got the arithmetic completely wrong.

The harder problem: clients who don’t have the technical background to distinguish a confident hallucination from a genuine insight. The output sounds authoritative either way.

The Privacy Problem Nobody Talks About Enough

High-net-worth clients using personal ChatGPT or Claude accounts are feeding sensitive financial data into systems without enterprise-grade data protections.

Firms like WE Family Offices operate under agreements — with Microsoft Copilot, for example — that prevent client data from being used to train public models. A personal paid plan doesn’t come with those guarantees.

Advisors are starting to brief clients on basic hygiene: where transcripts go, who has access, how data is stored. It’s a conversation that probably should have started earlier.

What AI Still Can’t Do

Analysis is not advice. That distinction matters more than it sounds.

An LLM can make a coherent argument for investing in a pre-IPO company. It cannot get you into the funding round. Access still comes from relationships, track records, and trust built over time.

Morgan Stanley’s Vince Lumia made a point worth sitting with: during bull markets, the value of human judgment is easy to underestimate. When volatility hits — and it will — larger clients tend to want a person on the other end of the phone, not a well-reasoned paragraph.

The Floor Is Rising

The more useful frame here isn’t disruption — it’s accountability. As Morningstar’s Sean Dunlop put it, if an advisor is keeping half a client’s balance in cash inside an IRA, AI should be a wake-up call.

The advisors most at risk aren’t the ones with deep relationships and genuine expertise. They’re the ones coasting on information asymmetry that no longer exists.

For everyone else, a client who shows up with sharper questions is a better client to work with.