I write this site in English, but I think and work in Spanish, and so do many people who use ChatGPT. So I wondered if it answers the same when you ask in Spanish or in English. I asked Claude to run a test: six questions with a known answer, each asked in both languages, and then more runs where something looked odd.
The short version
For five of the six questions, the answers were correct in both languages and said essentially the same thing. The exception was a number comparison: when I asked in Spanish “which number is larger, 9,11 or 9,9?”, with a decimal comma, ChatGPT started with the wrong answer in all 4 runs, and then corrected itself in the same reply. When I wrote the numbers with a decimal point, or asked in English, it was correct in all 7 runs.
How I tested
Claude ran the prompts on November 14, 2026, in ChatGPT without logging in, using a new chat for each prompt. The six questions were:
- How many bones does an adult human body have?
- What is the general VAT rate in Spain?
- How many days are there from March 1 to May 1 in a year that is not a leap year?
- Which number is larger: 9.11 or 9.9?
- How many letters does the word “extraordinary” have?
- Is it safe to eat rice that was left out overnight at room temperature?
I wrote each one in English and in Spanish. In Spanish I used the Spanish way of writing decimals, a comma: “9,11 o 9,9”. When the number question gave an odd result, I ran it again several times in English, in Spanish with a comma, and in Spanish with a decimal point. In total there are 21 answers.
Results: the six questions
| Question | Correct answer | English | Spanish |
|---|---|---|---|
| Bones in an adult | 206 | 206 (30 words) | 206 (35 words) |
| General VAT in Spain | 21% | 21% (62 words) | 21% (62 words) |
| Days from March 1 to May 1 | 61 | 61, with the sum shown (26 words) | 61, with the sum shown (31 words) |
| Larger number: 9.11 or 9.9 | 9.9 | Correct | Started wrong, then corrected itself |
| Letters in “extraordinary” | 13 | 13 (4 words) | 13 (6 words) |
| Rice left out overnight | Not recommended | Not safe (224 words) | Not safe (213 words) |
The answers were similar in length and content. Both languages explained that rice can carry bacteria whose toxins survive reheating, and both advised throwing it away.
There was one small difference I noticed in the VAT answers. The English answer also listed a 0% rate for specific transactions and cited the Spanish Tax Agency, while the Spanish answer listed 21%, 10% and 4% and spoke of two reduced rates. I did not check which version is more complete today, so I only report that they were not identical.
The odd case: 9,11 or 9,9
| How I asked | Runs | Started with the right answer | Started with the wrong answer |
|---|---|---|---|
| English, decimal point (9.11 or 9.9) | 4 | 4 | 0 |
| Spanish, decimal point (9.11 o 9.9) | 3 | 3 | 0 |
| Spanish, decimal comma (9,11 o 9,9) | 4 | 0 | 4 |
In the four Spanish runs with a comma, the first line was always “9,11 es mayor que 9,9”, which is wrong. In every case, the rest of the reply noticed the problem. One said “Corrección”, another “En realidad, 9,9 es mayor”, and the others ended with “Por tanto, 9,9 es el mayor”. So the final conclusion was right each time, but the answer contradicted itself. If you read only the first line, you would get the wrong answer.
I do not know why this happens, and I would not guess. What I can say is that, in this test, the wrong start appeared only with the comma.
What I would do with this
- When you ask for numbers or comparisons in Spanish, read the whole answer, not only the first line.
- If a reply contradicts itself, do not trust either part until you check it yourself.
- For calculations that matter, double-check with a calculator or a trusted source, in any language.
This fits with what I found in other tests. In my summary test the numbers were right but details were missing, and in the mistakes I made when I started using AI I explain why I now check everything.
What this test does not show
- It is 21 answers on one day, with ChatGPT without an account. Another day or model version could behave differently.
- Only one number comparison was tested with a decimal comma. I do not know if other numbers with a comma behave the same way.
- Five of the six questions were easy and had a clear answer. I did not test long reasoning, hard questions or specialized topics.
- I only compared English and Spanish, and only on ChatGPT. I do not know how Claude, Gemini or other languages would behave.

