It’s estimated that over half of UK adults use artificial intelligence for financial advice according to Lloyds Bank, including guidance on cryptocurrency. As Artificial Intelligence continues to rise in popularity, the public’s reliance will naturally follow suit.
But which AI model can you rely on for your investment advice? Experts at psyfi money conducted a study looking at which of the most popular AI models is the best for providing investment advice.
The study evaluated each response based on several key criteria: the amount of jargon used, beginner-friendliness, accuracy, relevance, length, price discrepancy of cryptocurrencies, and the inclusion of up-to-date legal and regulatory information. These factors were combined to calculate a total index score out of 100 for each provider.
Which AI model performed the best?
| AI Model | Jargon Count | Beginner Friendliness Score | Accuracy Score | Relevancy Score | Length Classification | Average Price Discrepancy (%) | Law and Regulation Score | Total Index Score / 100 |
|---|---|---|---|---|---|---|---|---|
| ChatGPT | 33 | 7.6 | 8.6 | 9.2 | Clear and Detailed | 3.09 | 7.5 | 82 |
| Claude | 36 | 8 | 8.6 | 8.4 | Clear and Detailed | 17.75 | 7.0 | 72 |
| Perplexity | 35 | 7.4 | 8.8 | 8.6 | Clear and Detailed | N/A | 5.0 | 64 |
| Gemini | 39 | 6.8 | 8.6 | 8.6 | Overcomplicated | 765.09 | 6.0 | 54 |
| Google AI Mode | 38 | 6.4 | 8 | 7.6 | Clear and Detailed | 0.71 | 5.5 | 36 |
| Grok | 44 | 5.4 | 8 | 7.8 | Overcomplicated | 0.48 | 4.5 | 16 |
The strongest performers still have serious downfalls
ChatGPT ranked highest overall, with a score of 82 out of 100 for its balance of clarity, relevance and accuracy.
Claude followed with 72, while Perplexity placed third. However, even the top-performing models showed issues, particularly on legal and tax nuance.
One of the most serious problems the study revealed concerned regulatory guidance. When asked about the safest way for a beginner to invest in cryptocurrency in 2026, Claude incorrectly described Binance as registered with the Financial Conduct Authority (FCA). In reality, Binance was ordered to cease regulated activities in the UK in 2021.
“Recommending a non-FCA-registered exchange as the safest option for a beginner is about as concerning as it gets,” said Michele Tieghi, founder of psyfi money. “Investors could be left without asset protection and exposed to market manipulation.”
While other models mentioned Binance more cautiously, the error highlights the broader risk of AI systems presenting outdated or incorrect compliance information with confidence.
Model strengths by investor type
The study also identified how each AI model would work more efficiently for different types of investors, though none were without risk.
Beginners: Chat GPT and Claude emerged as the most beginner-friendly model, scoring 7.6 and 8 out of 10 for accessibility. However, Claide’s incorrect claim that Binance is FCA registered significantly shows the information it gives can’t always be trusted.
Intermediate Investors: Perplexity offered strong source-backed guidance and achieved the highest accuracy score (8.8 out of 10), making it more suitable for users who already have a basic understanding.
Advanced Investors: Gemini offered the most detailed, technical analysis, but its heavy use of jargon and inaccurate pricing and regulatory guidance mean investors could face costly mistakes if they rely on the model without careful scrutiny.
The answers given use overly technical language and missing nuance
Beyond outright errors, the study found that AI-generated responses were frequently over-complicated and jargon-heavy. Grok produced the most technically complex and least beginner-friendly answers, while Gemini also ranked highly for jargon use.
Legal and tax guidance raised another serious red flag. While most models correctly stated that cryptocurrency airdrops may be taxable, they failed to explain the distinction made by HM Revenue & Customs between tokens received in exchange for services and unsolicited distributions, a distinction that could significantly alter an investor’s tax liability.
“AI models can summarise broad principles, but they often miss the finer regulatory detail,” Tieghi said. “In areas like tax and compliance, those details matter, and can cost investors thousands in fines if not handled correctly.”
Pricing errors put investors finances at risk
The study analysed responses from six of the most popular AI models in the UK, asking each model to price 50 different cryptocurrencies, and shockingly found that no AI model could accurately price every single one. One AI model missed the mark by over 700 per cent.
- Gemini misses the mark on crypto prices by over 700 per cent.
- No AI model could accurately price all 50 cryptocurrencies that were tested, with the highest being Google AI Mode which correctly priced five.
- Grok had the lowest price discrepancy, with it being 0.48 per cent on average away from the correct price.
- AI models particularly struggled to price Shiba Inu, with four of the six models having it as their highest price discrepancy.
AI Models ranked on cryptocurrency price accuracy
| Ranking | AI Model | Average Price Discrepancy (PCT) | Highest Price Discrepancy | No. of Accurately Priced Coins /50 | No. of Accurately Priced Coins Within 1 Per Cent /50 |
|---|---|---|---|---|---|
| 1 | Grok | 0.48 per cent | Shiba Inu (+7.03 per cent) | 3 | 45 |
| 2 | Google AI Mode | 0.71 per cent | Shiba Inu (+7.39 per cent) | 5 | 41 |
| 3 | ChatGPT | 3.09 per cent | Shiba Inu (+8.82 per cent) | 0 | 28 |
| 4 | Claude | 17.75 per cent | Shiba Inu (-100 per cent) | 1 | 9 |
| 5 | Gemini | 765.09 per cent | Pi (+33,915.53 per cent) | 1 | 10 |
| 6 | Perplexity | Perplexity was unable to return any prices at the time of research | N/A | N/A | N/A |
Michele Tieghi, financial expert and founder of psyfi money commented on the research:
“The lack of precision when it comes to cryptocurrency prices should raise serious concerns for anyone who is thinking about using these AI models to assist in their cryptocurrency trades. The average discrepancy from Gemini is especially alarming as even a small discrepancy in price information can be extremely costly when trading cryptocurrencies, let alone a discrepancy as extreme as 765 per cent. To put that into perspective, that would be like pricing Bitcoin at over £380,000, when the real current price is £50,300.
“Anyone buying or selling cryptocurrencies should rely on the exchange’s listed price to make the most accurate decisions.
“AI is advancing rapidly, but this study makes it clear that it still has a long way to go when it comes to cryptocurrency advice. Its limitations, from inaccurate pricing to incorrectly labeled exchanges, make it unsuitable as a standalone guide for investors.
“AI should be used to complement careful, expert-led research, not replace it. The lack of nuance, detail, and accountability highlighted in this study is a serious warning for anyone considering relying solely on AI for crypto guidance.”
Methodology
The six most popular AI models in the UK (ChatGPT, Gemini, Claude, Grok, Perplexity and Google AI Mode) were asked 25 questions regarding advice around investing and cryptocurrency, 10 questions about laws and regulations surrounding cryptocurrency and for the prices of 50 popular coins.
These answers were then analysed by a panel of experts at psyfi money and scored across the following categories:
- Jargon Count: Number of jargon words and terminologies used throughout the answers provided regarding advice.
- Beginner Friendliness Score: How accessible and understandable the answers are for a complete beginner in crypto.
- Accuracy Score: How factually correct, balanced, and responsible the information is based on widely accepted crypto and financial knowledge.
- Relevancy Score: How directly and completely the model answers the specific question asked.
- Length Classification: Each model’s answers were classified into one of the following:
- Oversimplified: Too brief, lacks key explanations, missing context.
- Clear and Detailed: Appropriate length, thorough but digestible.
- Overcomplicated: Excessively long, repetitive, overly technical, or unnecessarily dense.
- Average Price Discrepancy: The average percentage difference between the correct price at the time of asking versus the price given by each model.
- Law and Regulation Score: How precise and detailed the information is based on the necessary information someone would need regarding the questions asked.
Sources:
- Cryptocurrency prices: Prices correct according to https://coinmarketcap.com/ at 16:00 on the 9th Feb 2026.
