Experiment: Same Questions to Several AIs (Method)
A protocol for asking several AIs the same financial questions and checking them against official sources. Results pending.
This experiment puts the same financial questions to several AI assistants and records each answer with date and model so it can be checked against official accounts. We publish the protocol and the template; results do not exist yet and we do not invent them. They will be added once it is actually run.
Short answer: if you ask several AIs the same financial question, they can answer differently, with different confidence and different dates. Here is the method to find out. Status: not yet run. There is no results table because we have not run the protocol with saved outputs, and we will not make one up.
Key takeaways
- Same questions, same named models, same date.
- Each figure is checked against the official source, not between AIs.
- Only what is actually run is published, with a dated capture or quote.
See the Apple analysis in 10 seconds
What you just read, applied to Apple. A preview with real data:
- Making money? Yes
- Pricey vs its sector? Pricier
See the full Apple analysis free → Already done: nothing to type.
1. Protocol questions
All about a well-known company (for example Apple) and easy to verify:
- What is the company's current P/E ratio? State the date of the figure.
- How much dividend per share did it pay last year?
- What percentage of earnings does it pay out (payout ratio)?
- What are three risks it states in its latest 10-K?
- Where can I verify these figures?
No question asks whether to buy or for a price target.
2. Running rules
- Date: the same for every test, noted at the start.
- Models: name and version as each service displays them.
- Text: the question is pasted unchanged; once per model.
- Record: the full answer is saved (a capture or a dated quote).
3. Notes template
| Question | Model | Gave a date? | Gave a source? | Matches the official figure? | Notes |
|---|---|---|---|---|---|
| 1 (P/E) | |||||
| 2 (Dividend) | |||||
| 3 (Payout) | |||||
| 4 (Risks) | |||||
| 5 (Where to verify) |
The template is empty on purpose. Filling it in requires running the tests.
4. How to cross-check with official sources
For each figure, open the primary source: the 10-K on SEC EDGAR, or the annual report and the CNMV for a Spanish company. The Apple page on Fintols, with figures as of November 9, 2026, can serve as a second reference (dated, with N/A if something is missing), but the official source wins. The step-by-step procedure is in how to verify ChatGPT's financial numbers. The comparison between a chatbot and an analyzer with sources is set out in ChatGPT vs Fintols on the same stock.
Keep reading free.
Create your free account to continue reading “Experiment: Same Questions to Several AIs (Method)”.
No card required
5. What will be published (when it exists)
Only the table with the models, the date and the discrepancies found, with links to the sources. No returns as a hook: CNMV Working Paper No. 96, *Large Language Models and Stock Investing: Is the Human Factor Required?*, looks at model performance with caution and we do not use it as a selling point.
6. Running it at home
You can run this protocol today: pick a company, note the date, ask the assistants you use and fill in the template. For a starting point with dated figures, open the analyzer. The rest of the topic is in the AI for investing hub.
Frequently asked questions
Where are the experiment results?
They do not exist yet. This piece publishes the method; results will be added once the protocol is run with saved, dated outputs.
Why is there no sample table?
Because it would mean inventing AI answers. We only publish what can be checked.
Which models should I use?
The ones you normally use. Note the name and version each service shows and the date of the test.
How do I know who is right?
With the official source: the 10-K, the annual report, the CNMV or SEC EDGAR. Not by comparing one AI with another.
Educational content; not investment advice.
Learn to analyze companies, one a week
Each week we explain a real company in simple words. Create a free account and try the analysis on a company you choose.
See it with real companies
Each of these pages shows the company's dated figures and how it compares with its sector:
Turn what you read into your first analysis.
Create a free account and run your first full analysis in a minute. No card needed.
- How it makes money
- Valuation scenarios
- What could go wrong
First full analysis free · No card
Keep learning
- How I Use AI Without Letting It Decide for Me
- Why NVIDIA stock is falling in November 2026
- What happens to a stock when the CEO leaves
- Hardware and devices stocks
- Software and cloud stocks
- Semiconductor stocks
- All Fintols articles in English
An article from the Fintols blog, adapted for US investors. Educational content, not investment advice.