Does AI give good financial advice? We put £16,000 to the test
AI is filling a gap the advice market left open. With fees rising and capacity shrinking, 40% of Britons now ask chatbots for unregulated guidance. That's a risky mix of low financial literacy and high confidence in tools that can sound certain but be wrong.
Policy voices are blunt. "There's nothing to stop it putting out rubbish," warned Sophie Legrand-Green of TISA. FCA-approved adviser George Sweeney added a core risk: models can treat a BuzzFeed blog and a Bank of England report like equal sources. Helpful? Sometimes. Safe? Not by default.
What we tested
We asked three leading chatbots-Microsoft Copilot, ChatGPT and Google Gemini-how to invest £16,000 (roughly the UK average in savings). We then put their answers in front of Emma Wall, chief investment strategist at Hargreaves Lansdown, for a sanity check.
Her verdict in one line: the bots can outline basics (risk, diversification, tax wrappers), but they miss context, overconcentrate, and ignore practical frictions.
Copilot: strong tech flavor, weak diversification
- Produced 25 ideas across stocks, commodities, bonds, ETFs, real estate and crypto, framed as "medium risk" for the long term.
- Biggest flaw: concentration in top-tier US tech, creating correlation risk for a UK retail investor. Company descriptions were fine, but lacked discussion of key risks you'd expect from a human analyst.
- US-centric bias throughout. As Wall noted, it's an awkward fit for a British saver building a core portfolio.
ChatGPT: sensible structure, but US double-count and property/cash issues
- Opened with investing principles (risk appetite, diversification, time horizon, cash flow) and suggested using a stocks and shares ISA via mainstream UK platforms. It then split the £16k between an upfront allocation and drip-feeding the rest over 6-12 months.
- The problem: a separate US fund plus a global fund that is ~65% US created an unintended US overweight. That adds fees and forces manual rebalancing-tough for beginners.
- Property at ~10% looked high given UK market pressures. Holding ~10% cash also felt heavy if rates are falling; Wall preferred government bonds for liquidity/income and gold for diversification.
Gemini: better guardrails, patchy implementation
- Clear disclaimers, an emergency-fund reminder, and three moderate, long-term plans via a UK stocks and shares ISA. It highlighted growth themes (AI, renewables, healthcare, defence) and named sectors/regions.
- Gaps: it mentioned commodities but offered no investable vehicles, and flagged renewables without routing through specific stocks or funds. That leaves novices to guess-and often pick expensive or ill-suited products.
What this means for finance professionals
- LLMs are decent at first-order thinking: risk framing, diversification, and wrappers like ISAs. They falter on second-order work: portfolio construction, regional balance, valuation context, costs and operational friction.
- Expect US bias and thin risk commentary. Treat outputs as prompts for research, not answers.
- Watch for property overweights in UK portfolios, cash drag as rates fall, and missing shock absorbers (gilts, gold) for liquidity and downside protection-points Wall raised repeatedly.
- Operationally, piecemeal ETF/stock baskets raise rebalancing effort and trading costs for retail clients. Consider whether fewer, broader vehicles could reduce complexity while maintaining objectives.
Why consumers are defaulting to AI
- Access is the choke point. Adviser firms accepting clients with under £50k in investable assets fell from 52% to 25% in six years; those serving £200k+ rose from 11% to 30%.
- Usage skews young: 65% of Gen Z and 61% of millennials report using AI for personal finance help. Demand hasn't disappeared-it shifted channels.
Regulatory shift: "targeted support" on the way
Today, firms can give a recommendation only after a full, personalised suitability assessment-a costly service. The FCA is working on "targeted support": practical suggestions for common scenarios (e.g., excess savings, retirement income) that sit between generic guidance and full advice. The goal: broader access without the full advice price tag.
Some industry leaders see this improving economics for lower-asset clients. Others worry that perception won't change-many consumers will still default to "free" AI answers. You can read the FCA's work on the advice-guidance boundary and targeted support here: FCA Advice Guidance Boundary Review.
Practical checklist: using AI safely in client workflows
- Force assumptions into the open: ask the model to state data cut-off, regional exposure, and macro views it's implicitly using.
- Demand UK specifics: ISA eligibility, platform availability, fund domicile, OCFs, and FX/trading fees.
- Stress concentration risk: cap single-country/sector weights and ask for alternatives that reduce correlation.
- Interrogate risk sections: require a bullet list of key risks per position (valuation, earnings sensitivity, balance sheet, regulatory, liquidity).
- Map to an IPS: have the model restate portfolio fit against the policy's risk band, drawdown tolerance and rebalancing rules.
- Verify with primary sources: fact-check tickers, fees, holdings, and mandates on issuer pages before implementation.
- Keep a human in the loop: AI drafts the idea; your team validates, prices, and decides.
A note on tax wrappers
All three bots pointed to using a stocks and shares ISA to improve tax efficiency. For clients who qualify, that call is sound in principle. For rules and annual limits, see the official guidance: Gov.uk: Individual Savings Accounts (ISAs).
Bottom line
AI can help you structure thinking and speed up research. It can't carry fiduciary responsibility, judge source quality reliably, or price risk in real time. Treat it like a junior analyst who writes fast, not a portfolio manager who decides.
Further learning: If you're upskilling teams on safe, high-leverage workflows in finance, explore AI for Finance.
This article is for information only and is not financial advice.
Your membership also unlocks:
