Why does ChatGPT give different answers to the same question?

If you ask ChatGPT the same question twice and get two different replies, nothing is broken. This is built into how language models work.

When ChatGPT writes an answer, it does not pull a fixed reply from a database. It generates text one small piece at a time, and at each step it has several reasonable options for the next word. Instead of always choosing the single most likely option, it picks with a bit of controlled randomness. This setting is often called temperature. A little randomness makes the writing feel natural and varied instead of robotic, but it also means each attempt can take a slightly different path.

Other things change the answer too. The conversation history matters, because ChatGPT reads everything earlier in the chat before replying. Asking the same question in a new chat versus halfway through a long discussion will give different context. Memory and custom instructions can also shape replies. If the model has been updated in between, or if you are switched to a different model because of usage limits, the style and content may shift as well. Answers that use web search can change because the search results themselves changed.

For creative tasks, this variety is useful. Ask for ten Instagram captions and you want ten different ideas. For factual questions, though, two different answers can be a warning sign. If the facts disagree between attempts, the model is probably unsure, and you should verify with a reliable source.

Here is how to get more consistent results:

1. Be specific. "Explain GST registration rules for a freelancer in India with annual income under 20 lakh" leaves less room for drift than "tell me about GST."
2. Give a format. Ask for a table, a numbered list or a fixed word count.
3. Provide the source material. Paste the document or data and ask the model to answer only from it.
4. Use custom instructions to set your preferred tone and level of detail once.
5. Developers using the API can lower the temperature setting for more predictable output.

Think of it like asking a knowledgeable friend the same question on two different days. The core idea should stay the same, but the wording will change. If the core idea itself changes, that is your cue to check further.