8. Common prompting mistakes and how to avoid them

Even with all this knowledge, it’s easy to fall into a few traps when writing prompts. Let’s highlight some common mistakes that product managers (and others) often make when starting out with prompt engineering, along with tips to avoid them. Recognizing these pitfalls will help you troubleshoot when a prompt isn’t giving you what you need.

Mistake 1: Being too vague – “Tell me something about our product.”

• What goes wrong: A vague prompt yields a vague answer. The AI has to guess what you want, often resulting in generic or irrelevant info. For example, “Explain our product” might lead to a basic description, but it might miss the angle you cared about (performance issues? selling points? etc.).
• How to avoid: Be specific about what you want and about what aspect of the topic. Instead of “tell me something,” ask something like “Summarize the top 3 benefits our product offers to e-commerce customers.” In a PM context, don’t just say “Give me feedback,” but “Give me feedback on the new onboarding flow, focusing on potential usability issues.” The more clearly you define the request, the more targeted the response.

Mistake 2: Asking too much at once – “Explain our pricing strategy, generate three campaign ideas, and do a SWOT analysis of competitors.” (all in one prompt)

• What goes wrong: Overloading the prompt with multiple tasks can confuse the model or lead to a superficial handling of each part. It might do one task and neglect the rest, or do all poorly.
• How to avoid: Break complex or multi-part requests into separate prompts or at least separate clearly within one. If it’s logically sequential, do one at a time. For example, first ask for the SWOT on competitors. Then in a new prompt ask for campaign ideas (maybe even using insights from that SWOT). This ensures each output is focused. If you must combine, delineate sections: “Part 1 …; Part 2 …” but it’s safer to separate. Think of it like how you’d ask a colleague – you wouldn’t dump five requests in one breath if you want clear answers, you’d sequence them.

Mistake 3: Wrong tone or style – “Write the release notes” (and you expect a formal tone, but you get a casual one, or vice versa).

• What goes wrong: If your prompt doesn’t specify style, the AI might choose one that doesn’t fit your audience. Or if your wording implies a certain tone accidentally, you might get that. For instance, phrasing something very formally might lead the AI to continue formally even if you wanted a friendly tone.
• How to avoid: Explicitly state the tone or style if it matters. E.g., “Draft a release note in a friendly, informal tone that non-technical users can understand.” or “Provide a professional and concise summary for executive stakeholders.” Also, consider the examples you give: if you say “Dear user, we are thrilled…” the AI will likely continue in that excited tone. If that’s not desired, adjust your prompt phrasing. Essentially, lead by example and/or instruction.
• Note: There’s also the case of asking for a tone that mismatches the content – e.g., “Make a joke about a serious security breach” – which is a bad idea. Not exactly a prompt mistake, but a judgment mistake. The AI might comply in making a joke, but that’s obviously not appropriate. So remember common sense – don’t use AI to generate content in an unsuitable tone for the context, and if you do specify an unusual tone, double-check if that’s wise.

Mistake 4: Not providing enough context or details – “How big is it?” (with no reference to what “it” is). Or “Summarize the feedback” (which feedback? whose feedback?).

• What goes wrong: The AI lacks information to give a good answer. In the worst case, it might guess incorrectly (like assume “it” refers to something from earlier in the conversation incorrectly, or just not answer usefully). Even if you had a prior conversation, if you start a new chat and say “continue from earlier” – it won’t know. It’s not persistent across sessions.
• How to avoid: Always imagine you’re explaining the situation to a new team member or an intern. Don’t assume the AI knows your product or the specifics of your project (unless you described them). For example, instead of “Summarize the feedback”, say “Summarize the user feedback from our last mobile app beta test. We received feedback like users want easier onboarding, etc. Focus on common requests.” Similarly, always define pronouns: “it”, “they”, etc., should be clarified (e.g., “the feature” instead of “it” if multiple things were discussed). It can help to re-read your prompt and see if any word could be ambiguous. If yes, clarify it.

Mistake 5: Unrealistic expectations from AI – “What should our company’s strategy be next year?” or “How should I feel about this product failure?”

• What goes wrong: You might be asking the AI to do something beyond its capability or role. It’s not a fortune teller or your therapist (well, it can mimic one, but it doesn’t truly know your feelings). Questions that are too broad (“what should we do?”) or too human (“how would you feel…”) can lead to either generic advice or misleading answers. For example, it might give generic business advice that sounds plausible but lacks insight into your unique situation.
• How to avoid: Use AI for what it’s good at: analyzing given data, generating options, summarizing info, providing general knowledge. But for decisions that require deep domain knowledge or emotional intelligence, use AI as input, not the oracle. Instead of “What should our strategy be?”, you could ask, “What are some strategic options for a company in [domain] given [certain data]?” Then you evaluate those options. If you catch yourself wanting the AI to make a tough call for you, reframe it to help you make the call (like pros/cons, scenarios). And for emotional or subjective matters, maybe avoid making AI the judge (“how would a typical user feel” is okay, but “how should I feel” is weird).

Mistake 6: Not iterating or learning from poor outputs – You use the same prompt wording over and over even if it’s not quite giving the right result.

• What goes wrong: Insanity is doing the same thing expecting different results – if a prompt was unclear and gave a bad answer, using it verbatim again likely won’t change much. Or sometimes people give up after one bad try, assuming “AI can’t do this”, when maybe a rephrased prompt could.
• How to avoid: Always iterate. If the output isn’t what you want, analyze why. Did I miss context? Was I too vague or too broad? Try a revision. This guide basically encourages that at every step. Each “failure” is feedback to refine your approach. For instance, if you got an overly long answer and you wanted just a summary, next time explicitly say “in a brief summary”. If the answer veered off topic, double-check if your prompt accidentally introduced something. Also, use the conversation: say “That’s not what I needed, focus more on X in the next answer.” ChatGPT usually will correct the course. The key is to not see a subpar output as final – it’s a draft. Prompting is an iterative process, and learning from what didn’t work is part of getting better.

Mistake 7: Ignoring the audience in the prompt – “Explain feature Z.” (but you didn’t specify to whom – a user? developer? exec?)

• What goes wrong: The explanation might be too technical or too simplistic depending on default assumptions. A common oversight is not telling the AI who the output is for. If you just say “explain X”, it might assume a general audience, which might not fit your needs (maybe you needed an explanation for novices, but it used jargon).
• How to avoid: Include the target audience or level of detail in the prompt. E.g., “Explain feature Z to a non-technical customer who has never used our app before.” Or “Explain feature Z to our engineering team, focusing on the technical implementation details.” These two prompts yield very different answers about the same feature, as they should. Always tie the task to the perspective of the end reader if relevant.

Mistake 8: Over-complicating the prompt – Writing an essay in the prompt describing what you want in 10 different ways.

• What goes wrong: Sometimes in an effort to be precise, people write extremely lengthy prompts that may confuse more than clarify (or hit token limits). If you provide too much extraneous info or repeat yourself a lot, the AI might get mixed signals or waste effort on irrelevant details. It might also truncate what it was going to say because it spent half the output repeating your lengthy instructions back (some models do that).
• How to avoid: Be concise and clear. Include all necessary info, but no need to keep rephrasing the same instruction. Once you’ve stated the point, additional sentences should add new info, not redundancy. For example, don’t do: “It is very important that you only list the items, do not number them, just list them one after another in a comma separated format, and please ensure the list is comma-separated without numbering, and do not put them in a vertical list but in one line separated by commas.” That’s overkill and actually could confuse the model. Instead: “List the items in one line, separated by commas (no numbering).” That’s it. If your prompt is well-structured and maybe uses bullet points for different parts (which is allowed in input), it can be long in content but still clear. Just avoid unnecessary flourish or circular descriptions.

By staying mindful of these pitfalls, you can diagnose issues quickly. If an output disappoints, scan through this list in your head: Was I vague? Did I accidentally bundle requests? Did I give context? Is the tone off because I didn’t specify the audience? This mental checklist will often reveal the cause, and then the fix is straightforward – apply the best practice.

Remember, even experienced prompt engineers slip up sometimes. The difference is they realize it and adjust rapidly. With time, you will too.