← Back to Blog

Common Mistakes People Make When Writing AI Prompts

I've made basically every mistake on this list at some point, some of them repeatedly, before they finally stuck. Most disappointing AI results trace back to one of a small handful of repeatable habits, not some deep flaw in the tool itself. Here are the ones that come up most often, based on what actually tripped me up and what I see other people run into constantly.

1. Being too vague

"A landscape" or "help me write something" gives the model almost nothing to work with, so it fills the gap with the most generic interpretation it can reach for. This is the single most common mistake, and also the easiest to fix, add a subject, a style or tone, and ideally a purpose, and the quality jump is usually immediate.

2. Cramming contradictory ideas into one prompt

Asking for "minimalist and highly detailed" or "brief but comprehensive" sends the model mixed signals it can't fully resolve. Pick one direction per prompt instead, you can always generate a second version taking the other approach if you genuinely want to compare them side by side.

3. Forgetting to specify format

For chat prompts especially, not stating whether you want a list, a table, a short paragraph, or a long explanation means the model has to guess, and it often guesses wrong for what you actually needed in that moment. This one's easy to forget precisely because it feels obvious in your own head.

4. Writing a full sentence when a phrase list would work better

This applies mainly to image prompts. Tools like Midjourney parse comma-separated descriptive phrases more effectively than grammatically complete sentences, "a cat, sitting on a windowsill, soft morning light" tends to outperform "There is a cat that is sitting on a windowsill in the soft morning light," even though the second one reads more naturally to us.

5. Giving up after one bad result

A weak first output usually isn't a dead end, it's information about what to adjust. Look at what went wrong specifically, change one or two elements, and try again rather than abandoning the whole idea after a single disappointing attempt.

A bad first result is information, not a verdict.

6. Not giving the AI a role for chat tasks

Skipping a role instruction, "Act as a...", means the model answers in a generic, average tone instead of the specific expertise or voice that would've actually suited your task better. It's one sentence that shapes the entire rest of the response.

7. Overloading a single prompt with too many tasks

Asking for a blog post, an SEO title, a meta description, and social captions all in one prompt usually produces a mediocre version of everything rather than a strong version of anything. Breaking it into separate, focused prompts consistently gets better results per individual task.

8. Ignoring length control

Not specifying roughly how long you want a response means you might get three sentences when you wanted a full page, or the reverse. A simple word or sentence count instruction fixes this immediately and removes one more thing left to chance.

Avoiding these mistakes is mostly about being specific rather than assuming the AI will correctly infer what you meant, which it very often won't. If you'd rather skip the trial and error altogether, try the PromptNest generator, it structures your idea into a properly detailed prompt automatically. New to prompts entirely? Start with our beginner's guide.