When you ask ChatGPT "what is Tesla's P/E ratio?", it generates a number from its training data. Research shows ChatGPT hallucinates financial data in up to 47% of cases. GPT-4 Turbo achieves only 79% accuracy even when given the full filing context, collapsing to 19% without it. The numbers look precise, which makes them harder to catch. A hallucinated P/E of 31.4 is more dangerous than a hallucinated claim that "Apple makes good products," because the number looks like it came from a database.
This is a data access problem, not an intelligence problem. Better prompts do not give ChatGPT access to data it does not have. Shibui Finance is a free MCP server that connects ChatGPT to 64 years of US stock market data: 31M+ daily prices, quarterly financials, daily valuations, and 56 technical indicators for nearly 10,000 companies. Once connected, ChatGPT queries a real database instead of generating numbers from memory. Research on MCP-based grounding shows that "grounding in external, verifiable sources is the primary mechanism by which [MCP] mitigates the risk of LLM hallucination," with studies reporting up to 92% reduction in financial hallucination when models have database access.
What goes wrong when ChatGPT has no stock data source?
The failures follow predictable patterns. Stale data: ChatGPT's training has a cutoff date. It cannot know that a company's P/E changed last quarter. Confident fabrication: for less-covered stocks, ChatGPT generates ratios from distributional priors. A hallucinated P/E of 31.4 looks precise enough to act on. Wrong time period: ChatGPT cannot distinguish current from historical. It may return 2024 Q3 revenue when you asked about the latest quarter, with no indication the number is outdated.
Derived metric errors: ratios like EV/EBITDA or FCF yield require multiple inputs. If any one input is hallucinated, the derived figure is wrong. Web search does not fix this: ChatGPT's browsing feature can look up a single figure from a financial website, but it cannot screen thousands of companies, run multi-period queries, or compute derived metrics from verified source data. It returns data in inconsistent formats and is too slow for cross-market analysis.
For mega-cap stocks with heavy press coverage, headline figures (revenue, market cap) are often close to correct. The problem is worst for mid-cap and small-cap companies, historical data, and any derived metric that combines multiple data points. That is most of what investors actually need.
What kinds of financial questions trigger ChatGPT hallucination?
Not every financial question is equally risky. The hallucination rate depends on how specific the question is, how commonly the data appears in training corpora, and whether the answer requires combining multiple data points. Here is how different question types break down:
| Question type | Hallucination risk | Why | Shibui covers |
|---|---|---|---|
| Current stock price | High | Training cutoff means any "current" price is stale | Yes (end-of-day) |
| Latest quarterly earnings (revenue, EPS) | High | Most recent quarter rarely in training data | Yes |
| Valuation multiples (P/E, EV/EBITDA) | Very high | Derived from price + financials, both may be wrong | Yes (daily history) |
| Technical indicators (RSI, MACD, Bollinger Bands) | Very high | Require exact daily price sequences ChatGPT does not have | Yes (56 indicators) |
| Historical comparisons ("revenue growth 2020-2024") | Medium-high | Older data may be in training, but ChatGPT mixes periods | Yes (back to 1962) |
| Cross-market screening ("P/E under 15 with 3yr FCF growth") | Extreme | Requires querying thousands of companies at once | Yes (~10,000 stocks) |
| News interpretation, analyst opinions | High | Subjective, no single data source | No |
The pattern: anything that requires current, precise, or cross-company data is high risk. If you are screening across the market, hallucinated inputs make the entire analysis unreliable.
Why don't prompts or workarounds fix ChatGPT financial hallucinations?
The most common workaround is adding a disclaimer to the system prompt: "always state when you are unsure about data." ChatGPT may add a caveat, but the underlying number is still generated from training weights. A confident disclaimer does not make a fabricated figure real.
Web browsing lets ChatGPT pull figures from financial websites. This works for spot-checking one number, but it cannot screen thousands of companies at once, cannot run temporal queries across multiple periods, and returns data in inconsistent formats. It is a workaround, not a data infrastructure.
Custom GPTs with uploaded files work for one company at a time. You download a financial statement, upload it to a custom GPT, and ChatGPT reads from the document. This is accurate but does not scale. You cannot upload financials for 10,000 companies. The document goes stale the day after you download it.
Building your own MCP server requires coding an API wrapper, sourcing data, handling rate limits, and maintaining infrastructure. Shibui Finance is already built: 64 years of data, no API keys, connect in 2 minutes.
None of these approaches give ChatGPT a database. They patch the symptom without fixing the cause.
How does connecting ChatGPT to a real database fix hallucination?
MCP (Model Context Protocol) lets AI models connect to external data sources. ChatGPT supports MCP via custom actions. Shibui Finance is an MCP server that gives ChatGPT read access to a pre-built financial database. Once connected, ChatGPT writes SQL queries against real data instead of generating from training weights. Every number traces back to a daily-refreshed data source, not a language model.
Setup takes about 2 minutes. See the step-by-step connection guide or the ChatGPT setup page.
"What is Tesla's current P/E ratio, and how has it changed over the past 5 years?"
Without Shibui, ChatGPT generates a plausible P/E from training data. With Shibui, ChatGPT queries the valuation table for Tesla's daily P/E history, returns the current value from yesterday's close, and shows the 5-year trajectory with actual dates. The difference is not the quality of the prompt. It is whether the model has access to real data.
Before and after
Simple lookup. Ask: "What is NVIDIA's P/E ratio?" Without a data source, ChatGPT says "approximately 65" (may be months stale, may be fabricated). With Shibui, ChatGPT queries the valuation table and returns yesterday's closing value with the exact date.
Screening. Ask: "Find stocks with P/E below 15 and positive free cash flow for 3 consecutive years." Without a data source, ChatGPT lists companies it remembers from training, with no guarantee the criteria are currently true. With Shibui, ChatGPT screens 9,900+ companies against actual quarterly financials and valuation data, checking each individual year.
Technical analysis. Ask: "Which stocks crossed below their 200-day moving average this week with RSI under 30?" Without a data source, ChatGPT cannot answer. It has no price data after its training cutoff. With Shibui, ChatGPT queries the technical indicators table, which is updated daily with 56 pre-calculated indicators including RSI, MACD, Bollinger Bands, and moving averages.
What does connecting a database not fix in ChatGPT?
ChatGPT may still hallucinate about things outside the database: news events, management commentary, analyst opinions, macroeconomic interpretations. Shibui gives ChatGPT financial data. It does not give ChatGPT judgment.
The data is end-of-day, not real-time. Intraday prices and after-hours moves are not available. Coverage is US equities only (NYSE, NASDAQ). No international markets, options, crypto, or commodities. Data comes from tier-3 providers, not Bloomberg or Refinitiv. It is good for screening and analysis but not institutional-grade for regulatory compliance.
If ChatGPT writes a wrong SQL query, the numbers will be real but the logic may be flawed. Shibui solves the data problem, not the reasoning problem. See the data sources page for what is covered and the feature overview for the kinds of analysis Shibui supports.
This does not make ChatGPT a financial advisor. It makes the numbers real.
Frequently asked questions
Does ChatGPT make up stock market data?
Yes, when it has no data source connected. Research shows ChatGPT hallucinates financial data in up to 47% of cases. It generates numbers that look precise - a P/E of 31.4, a revenue figure to the million - but they come from training weights, not a database. Connecting ChatGPT to a verified data source via MCP eliminates this: every number comes from a query, not a guess.
How accurate is ChatGPT for financial analysis?
Without a data connection, GPT-4 Turbo achieves roughly 79% accuracy on financial statements when given the full filing context, but collapses to 19% without it. For derived metrics like P/E ratios or free cash flow yield, accuracy is even lower because multiple inputs can each be wrong. With Shibui Finance connected, accuracy depends on query logic, not model memory. The numbers come from a database, so they are as accurate as the underlying data source (daily-refreshed US equity data back to 1962).
Can I connect ChatGPT to a stock database?
Yes. Shibui Finance is a free MCP server that gives ChatGPT access to 64 years of US stock market data - 31M+ daily prices, quarterly financials, daily valuations, and 56 technical indicators for nearly 10,000 companies. ChatGPT queries the database directly instead of generating numbers from memory. Setup takes about 2 minutes.
How do I add a developer mode app for stock data in ChatGPT?
Add Shibui Finance as a developer mode app in ChatGPT settings using the MCP server URL. The connection persists across conversations. See the step-by-step setup guide for detailed instructions with screenshots.
Does ChatGPT have access to real-time stock prices?
No. ChatGPT has no built-in stock data feed. Its web browsing feature can look up a single price, but cannot screen thousands of companies or run temporal queries. With Shibui Finance connected, ChatGPT gets end-of-day prices updated daily after market close for nearly 10,000 US equities. For intraday prices, use your broker or TradingView.
ChatGPT vs Claude for financial analysis?
Both hallucinate financial data without a data connection. Claude supports MCP natively; ChatGPT supports it via developer mode apps. Shibui Finance works with both. ChatGPT is generally stronger for code and spreadsheet generation; Claude is stronger for long document analysis. For data-grounded financial research, the model matters less than having a real database connected. See the Claude version of this guide for Claude-specific details.
How much does connecting a database reduce ChatGPT hallucination?
Research on MCP-based data grounding reports up to 92% reduction in financial hallucination when AI models query a real database instead of generating from training weights. Without filing context, GPT-4 Turbo accuracy on financial statements drops from 79% to 19%. Connecting a database addresses the root cause: data access, not model intelligence.
Do I need to build my own MCP server for stock data?
No. Shibui Finance is a pre-built, free MCP server with 64 years of US stock market data already loaded. Other approaches require you to build a server, configure API keys, handle rate limits, and maintain the infrastructure. Shibui is ready to connect in 2 minutes.