Lesson 1 of 8 · Prompt Writing That Works
Why the Same Question Gets Different Answers
Understand why AI answers change from one try to the next, and the simple habits that make results more consistent.
Ask an AI assistant the same question twice and you will often get two different answers. For a poem that is fine. For a work task, such as a formula, a policy summary or a customer reply, you want something you can rely on.
This lesson explains why answers vary and how to narrow the gap between a good answer and a random one.
Answers are generated, not looked up
As you saw in the first course, an AI chatbot writes its reply word by word, choosing what is likely to come next. A small amount of randomness is built in on purpose, so the writing does not sound robotic.
That means each reply is one of many reasonable answers. When your question is vague, there are thousands of reasonable answers, so each try lands somewhere different. When your question is precise, there are far fewer, so the answers start to agree.
The rule: the vaguer the prompt, the more the answers vary. Consistency comes from precision, not from luck.
Three reasons answers drift
1. The question leaves room to guess. "Write a summary" could mean three lines or a full page. The AI picks one, and next time picks another.
2. The conversation history changes things. Everything earlier in the same chat affects the next answer. A question asked in a long, messy chat can get a different answer from the same question in a fresh chat.
3. The tool or mode is different. ChatGPT, Gemini, Claude and Copilot are different models. Even within one app, a "fast" mode and a more capable mode can answer differently.
Make answers more consistent
| Habit | Example |
|---|---|
| Fix the length | "In exactly 5 bullet points" instead of "briefly" |
| Fix the audience | "For a new accounts assistant" instead of "simply" |
| Fix the format | "As a table with columns Step, Action, Shortcut" |
| Give the facts | Paste the actual text, numbers or policy you want used |
| Start fresh | Open a new chat for each new task |
| Ask for the reasoning | "Explain why you chose this formula" |
Try it: vague vs precise
Send this prompt in two separate new chats and compare:
Explain VLOOKUP.
Now send this one in two separate new chats:
Explain Excel's VLOOKUP to an office assistant who knows basic formulas.
Use exactly 4 short steps and one example that looks up a product price
from a product code. Under 120 words.
The second pair of answers will be much closer to each other, and both will be more useful.
When variety is useful
Sometimes you want different answers: names for a product, ideas for a social post, ways to open a difficult email. Then ask for it directly:
Give me 5 very different options, from formal to playful, and label each one.
Asking for several options in one reply is faster than regenerating again and again.
Practice task
Pick a question you asked an AI assistant recently and did not love the answer to. Rewrite it using at least three habits from the table above. Run the old and new versions in fresh chats and compare. Save the better prompt for lesson 7.
Key takeaways
- AI answers vary because they are generated with a little built-in randomness.
- Vague prompts allow many answers; precise prompts narrow them down.
- Fix length, audience and format, give the facts, and start a fresh chat per task.
- When you want variety, ask for several labelled options in one reply.
Finished this lesson?Mark it complete to track your progress on this device.