Common Prompt Engineering Mistakes to Avoid

Even good prompt engineers make mistakes that silently degrade their results. This guide catalogs the 15 most common errors we see across thousands of prompts — from beginners writing their first query to experienced developers building production AI systems. Each mistake includes a concrete example of what goes wrong and how to fix it.

1. The Vague Request

The mistake: Asking the AI to do something without specifying what good looks like.

Write a summary.

Why it fails: The AI has no idea what length, depth, audience, or format you want. You will get a generic description that misses your actual need.

The fix:

Write a 3-paragraph summary of the following article for a non-technical executive audience. Focus on business implications, not technical details. Use bullet points for the 3 most important takeaways.

2. Overloading a Single Prompt

The mistake: Asking the AI to perform 10 different tasks in one prompt.

Write a blog post, generate 5 social media variants, create an email newsletter, and design a landing page for this product.

Why it fails: Quality degrades with task quantity. The model has a fixed reasoning budget, and spreading it across 10 tasks means each gets shallow attention.

The fix: Chain prompts. Do one task, use the output as context for the next.

3. No Output Format Specification

The mistake: Expecting structured output without defining the structure.

Extract the key information from this contract.

The fix:

Extract the following from this contract and return as JSON:
{
  "parties": ["string"],
  "effective_date": "YYYY-MM-DD",
  "termination_clause": "boolean",
  "liability_cap": "string or null"
}

4. Negative Instructions Without Positive Alternatives

The mistake: Telling the AI what not to do without saying what to do instead.

Do not be too technical.

The fix:

Use analogies from everyday life. Define any technical term the first time you use it. Assume the reader has no prior knowledge of this topic.

5. Forgetting the Audience

The mistake: Writing prompts as if the AI knows who will read the output.

Explain machine learning.

The fix: Always define the audience, their background, and what they need to do with the information.

6. No Constraints on Scope

The mistake: Letting the AI decide how comprehensive to be.

Tell me about JavaScript frameworks.

The fix:

Compare the 3 most popular JavaScript frameworks for building single-page applications as of 2026. Limit to React, Vue, and Svelte. Focus on learning curve, ecosystem maturity, and hiring market. Max 400 words.

7. Assuming the AI Remembers Earlier Context

The mistake: Starting a new conversation and referencing information from a previous one. Unless the platform has memory enabled, the AI cannot see past conversations.

The fix: Paste relevant context into each new conversation, or use a platform with persistent memory.

8. Not Providing Examples

The mistake: Describing a desired format when an example would be clearer.

The fix: When you need a specific format, provide 1-2 examples. Few-shot prompting is one of the most reliable ways to control output style.

9. Too Many Adjectives

The mistake: Loading prompts with subjective qualifiers.

Write an amazing, incredible, super engaging introduction.

Why it fails: The AI does not share your taste. "Amazing" is meaningless to a language model.

The fix: Replace adjectives with constraints and examples.

10. Not Iterating

The mistake: Accepting the first output as final.

The fix: Every AI output is a draft. Ask for revisions: "Make this shorter," "Add an example," "Rewrite for a technical audience."

11. Hallucination Blindness

The mistake: Accepting facts, statistics, and citations from AI without verification.

The fix: Treat all AI-generated facts as provisional. Verify names, dates, numbers, and sources independently before publishing or acting on them.

12. Missing Error Handling in Code Prompts

The mistake: Asking for code without mentioning edge cases.

Write a function to parse a CSV file.

The fix:

Write a Python function to parse a CSV file. Handle: empty rows, malformed lines, files with or without headers, and Unicode encoding issues. Include error handling and logging.

13. Ignoring Temperature and Parameters

The mistake: Using default model settings for all tasks.

The fix: Adjust temperature (creativity) based on the task:

• Low (0.0-0.3): Coding, data extraction, factual summaries
• Medium (0.4-0.7): General writing, brainstorming
• High (0.8-1.0): Creative writing, fiction, ideation

14. Not Defining the Role Explicitly

The mistake: Skipping the role assignment.

The fix: Starting with "Act as a [specific expert]" anchors the model's perspective and tone. The more specific, the better.

15. Giving Up Too Early

The mistake: Abandoning a prompt after one bad result.

The fix: Good prompting is iterative. Change one variable at a time: add an example, clarify the audience, tighten constraints. Document what works so you can reuse it.

Building a Prompt Improvement Checklist

Before sending any prompt, run through this checklist:

• Have I defined the role?
• Have I described the audience?
• Have I specified the output format?
• Have I added constraints (length, tone, scope)?
• Have I provided examples if the format matters?
• Is there one primary task, or have I split it into multiple prompts?
• Have I asked myself what a bad output would look like, and prevented it?

Next Steps

Dig deeper into prompt structure theory with our Zero-Shot vs. Few-Shot Prompting guide.