Ten everyday prompting mistakes (and fixes)

0:32 Why study mistakes?

Why focus on mistakes? Because a handful of habits cause most disappointing answers, and they are easy to fix once you can name them. It is a bit like learning to drive. You do not become safe by memorising the manual; you become safe by noticing your own recurring errors, like forgetting the mirror, and building a habit that fixes each one. In this lecture we name those errors, so you can spot them in your own prompts in seconds.

1:07 Missing information

The first group is about missing information. Mistake one, the one-line prompt, gives generic results; fix it with context, audience and format. Mistake two, assuming the AI knows your world, produces the wrong product, company or currency; fix it by stating who you are and where you are. And mistake seven, not saying what to do when information is missing, produces invented deadlines, prices or names; fix it with if you don't know, write unknown, or ask me instead of guessing.

1:42 Example 1: 'make my CV better'

A simple example. Someone asks: make my CV better. The AI rewrites everything, adds skills they do not have, and uses phrases they would never say. Which mistakes happened? Vague feedback, no context, and no instruction about missing information. The fix: here is my CV and the job advert; improve the wording of the experience section only; keep every fact accurate; do not add skills I have not listed; and if something would strengthen the application but you do not know it, ask me. Now the edits are targeted, honest and useful.

2:22 Example 2: property listings (illustrative)

Now a business scenario, with illustrative numbers. A property agency in Lahore used AI to write listing descriptions for about one hundred properties a month. Complaints started arriving. Some listings mentioned parking or a garden that did not exist. The mistakes: one-line prompts and no rule for missing information. The team changed their prompt template: include only the features in the property sheet; if a feature is not listed, do not mention it; list any questions for the agent at the end. They also added a quick human check of three facts per listing: size, price and parking. Within a month, complaints about inaccurate listings dropped sharply. The numbers are illustrative, but misleading property descriptions can breach consumer protection rules, so this fix matters beyond customer happiness.

3:17 Conversation mistakes

The second group is about the conversation. Asking for everything at once means half the requests are skipped; number them or break the job into steps. Vague feedback like make it better produces random changes; say what is wrong and why, for example too formal for Instagram, my followers are students. Never starting fresh leaves you in a tangled conversation; start a new chat with an improved prompt. And giving up too early, after one bad attempt, means missing results that usually arrive by the third message.

3:55 Trust and safety mistakes

The third group is about trust and safety. Trusting facts without checking, especially statistics, quotes, sources, and health, legal or financial information. Pasting private information carelessly, such as customer data, passwords or confidential plans; remove or replace personal details and follow your workplace policy. And copy-paste without reading, publishing placeholders like your name in brackets, odd phrasing or claims you cannot support. You are the editor, and you are responsible for the final result.

4:27 Three newer mistakes

Now three newer mistakes. Mistake eleven: using the wrong mode, like waiting two minutes for a thinking mode to rewrite a caption, or asking a quick mode to solve a complex budget. Match the mode to the job. Mistake twelve: trusting an uploaded file blindly. Long or scanned documents, tables and images inside PDFs can be partly missed, so ask for quotes with page numbers and check the key answers. Mistake thirteen: letting memory or old instructions steer you wrong. A formal tone you set months ago may still be shaping today's answers. Review saved instructions and memory every few months.

5:11 Before and after

Here is a before and after. Before: write a product description. After: write a sixty-word product description for our online shop, based in Dubai, prices in dirhams. Product: handmade oud-scented soy candle, two hundred grams, burn time about forty hours from our supplier's spec. Audience: gift buyers aged twenty-five to forty-five. Tone: elegant, not over the top. No claims about health benefits. If you need details I have not given, list them as questions instead of guessing. Then the human step: check the burn time against the label.

5:49 The prompt doctor

Now the prompt doctor. Paste a weak prompt and a short excerpt of the answer you got, and ask the assistant to act as a prompt coach: diagnose which problems it has, such as missing context, vague words, no format, no audience, too many tasks, or no instruction for missing information; rewrite it; and say which single change will make the biggest difference. Compare its diagnosis with your own. Over time you will spot the problems before the coach does.

6:24 Diagnose in 30 seconds

Let us diagnose one prompt together. Write a product description. Which mistakes can you spot? It is a one-line prompt, it assumes the AI knows your shop, and it says nothing about missing information. Now the fix: a sixty-word description for an online shop based in Dubai, with prices in dirhams, the real product details, the audience, the tone, a rule against health claims, and an instruction to list questions rather than guess. Three mistakes fixed in about thirty seconds. That is the speed you are aiming for.

7:02 Recap

To recap. Most poor results come from missing information, messy conversations, unchecked trust, and now the wrong mode, blind trust in uploads, and stale memory. The habit that fixes most of them: before sending, ask, if I gave this to a smart new colleague, would they have what they need? Try this now: for one week, tag each disappointing answer with the mistake number that caused it. Fix your top two first. Next lesson: checking AI work and building your prompt toolkit.