50 Mistakes That Give You Bad Results from ChatGPT and AI

Most people who complain about weak AI results don't have a problem with the model itself — they have a problem with how the request is written. This list covers 50 actual mistakes, grouped into 4 clear categories so you can quickly diagnose your own.

First: Vagueness and Lack of Specificity (13 Mistakes)

1. Not specifying the desired length with an approximate word or paragraph count.

2. Leaving the target audience undefined: expert or complete beginner?

3. Asking for 'good text' or 'the best answer' without defining what quality means in that specific context.

4. Not stating the required language register: formal, colloquial, or simplified?

5. Using vague words like 'something' or 'topic' instead of naming the exact element precisely.

6. Not stating the text's final purpose: marketing, educational, informational, or personal?

7. Asking for a comparison between options without specifying the actual comparison criteria.

8. Not stating the platform the content will be published on — a tweet is nothing like a formal email.

9. Not specifying the required tone: serious, friendly, sarcastic, or strictly formal.

10. Leaving the number of requested examples completely open: is one enough, or do you want five?

11. Using the phrase 'make it better' without specifying which specific aspect needs improving.

12. Not clarifying whether the question is purely theoretical or asking for actionable, practical steps.

13. Asking for 'analysis' of a topic or data without specifying the angle: financial, marketing, or technical?

Second: Missing Context (12 Mistakes)

14. Starting a follow-up question without reminding the model of the conversation's original goal.

15. Not mentioning the project's or company's background when requesting content tailored to it.

16. Failing to mention who will actually read the text or where it will be used after it's written.

17. Not informing the model of constraints or decisions agreed on earlier in the same conversation.

18. Assuming the model already 'knows' internal terminology specific to your team or field.

19. Not stating the time context when relevant, such as noting whether data is recent or outdated.

20. Failing to mention the cultural or geographic audience the content is aimed at.

21. Not clarifying the audience's prior level of knowledge about the topic at hand.

22. Jumping to a completely new task within a long conversation without clearly separating it from what came before.

23. Not referencing prior examples that succeeded or failed with the model on a similar task.

24. Ignoring legal or regulatory constraints tied to the field, such as medical or financial domains.

25. Not clarifying whether you want the model's opinion or strictly objective facts.

Third: Missing Constraints and Format (13 Mistakes)

26. Not specifying the final required format: flowing paragraphs, bullet points, or a table?

27. Not stating what to avoid, like complex technical jargon or emojis.

28. Leaving the number of required sections or subheadings completely unspecified.

29. Not precisely specifying the language: formal Arabic, a specific dialect, or English?

30. Requesting code without specifying the programming language or the required library version.

31. Not stating a maximum or minimum word count for each section of the text.

32. Ignoring the required ordering of elements: most important first, or chronological order?

33. Not requesting sources or clear attribution when factual accuracy really matters.

34. Not specifying whether the answer should include practical examples or stay purely theoretical.

35. Leaving the level of detail undefined: a quick summary, or an in-depth, detailed explanation?

36. Not requesting publish-ready wording when that's the actual goal of the request.

37. Ignoring the linguistic audience type: native speakers or people learning the language as a second one?

38. Not specifying whether a closing statement or a clear call to action should be included at the end.

You don't need a long, complex prompt — you need a clear one.

Fourth: Not Iterating and Refining (12 Mistakes)

39. Expecting a perfect final result from the very first attempt, with no editing or iteration.

40. Judging the entire tool's capability based on a single failed attempt.

41. Deleting the conversation and starting from zero instead of fixing one specific point within it.

42. Giving vague feedback like 'this is wrong' without pinpointing exactly where the error is.

43. Repeating the exact same prompt verbatim while expecting a completely different result each time.

44. Not asking the model why it chose a particular phrasing before judging it.

45. Skipping a side-by-side comparison of two different answer versions to determine which is actually better.

46. Not testing a prompt on a small, cheap example before applying it to a large, costly task.

47. Giving up at the model's first refusal instead of clearly reframing the actual intent.

48. Not saving successful prompts to reuse as templates in the future.

49. Changing several elements of the prompt at once, making it hard to know which change actually made the difference.

50. Not requesting a shorter, longer, or differently styled version to compare options before settling on the final one.

An Example Prompt That Avoids These Mistakes

Write a 150-200 word LinkedIn post for small business owners, in a direct and friendly tone without complex jargon. The goal is raising awareness about data backup importance. Use no more than two emojis. End with a question that encourages engagement.

Conclusion

Review the last 5 prompts you wrote and compare them against this full list of 50 mistakes. You'll likely find most problems trace back to one or two of these simple mistakes — not any real weakness in the tool.