Ever been frustrated because ChatGPT was not giving you what you thought you wanted? I’ve been there.
You type what feels like a clear request. The answer comes back polite, general and not quite it. You try again, and it is the same story. At some point, you start blaming the tool.
Most times, the problem is not the AI. It is the prompt. The request was clear in your head, but only half of it made it onto the screen. AI is clever, but it cannot read your mind. It is not the Holy Spirit.
The good news is that prompts are easy to fix. Here are five common reasons AI misses what you meant, with a prompt you can copy for each. (A prompt is simply what you type to the AI.)
1. Your request was too thin
Suppose you typed: “Write a report on staff training.”
The AI does not know who you are, who will read it, how long it should be or what was in the training. So it guesses, and it guesses safe and general.
Anthropic, the maker of Claude, describes the right mindset well: treat the AI as a brilliant new employee who does not yet know your workplace. The more precisely you explain what you want, the better the result. Its guide even offers a test: show your prompt to a colleague who knows nothing about the task and ask whether they could do it. OpenAI’s advice for ChatGPT says the same thing: be clear and specific, and give enough context.
The fix: say who you are, who it is for, what you need and what shape it should take.
Try this: “I am an HR manager at a 30-person firm in Abuja. Write a one-page summary of this week’s staff training for our managing director. Use plain English and five bullet points. Training topics: [paste your notes here].”
2. You stopped after one try
Hardly anyone gets a perfect result from the first message. A first answer is a draft, not a verdict.
OpenAI’s own guide says getting good results is often iterative: start with a prompt, read the answer, then adjust the wording or add context. It also suggests describing the tone you want with simple words such as formal, friendly or professional.
The fix: reply to the answer, the way you would reply to a colleague.
Try this: “Good start. Make it shorter. Keep the second bullet as it is. Use a warmer tone, and start with the most important point.”
Two or three follow-ups usually take you further than a perfect first message ever will.
3. You asked for the whole job at once
“Write my annual report” is too big for one message, just as it would be for a person. Large jobs go better in steps, and Anthropic’s guide lists breaking a task into a chain of smaller prompts as a standard technique.
The fix: split the job. Ask for an outline first. Approve it. Then ask for one section at a time.
Try this: “I need to write a report on our year. Before you write anything, ask me five questions you need answered, one at a time.”
That last move is useful whenever you are not sure what to include: let the AI interview you.
4. You trusted it, or it made something up
This one cuts both ways. Some people stop using AI after it gets a fact wrong. Others never notice the mistakes until it is too late.
OpenAI describes hallucinations as plausible but false statements: the model gives a confident answer instead of admitting it is unsure. The company says this remains a hard problem even as models improve. So a wrong answer does not mean the tool is useless. It means you need a habit.
The fix: give it your own material and tell it to stay inside that. Give it permission to say “I don’t know.” Anthropic’s guidance on reducing made-up answers recommends exactly this, and also suggests asking the AI to find a supporting quote for each claim and to drop any claim it cannot support. Then check names, numbers and dates yourself before anything leaves your desk. (We walk through the checking step in how to turn messy meeting notes into clear minutes with AI, and the staff rule behind it in the one-page AI rule for your team.)
Try this: “Use only the notes I paste below. If something is not in the notes, say ‘not in the notes’ instead of guessing. After your answer, list any claim you are unsure about.”
5. You tried it on the wrong kind of task
If your first question was a trivia question, a very recent news item or something only you could know, you were testing the weakest part of the tool. AI is strongest at the everyday work of drafting, rewriting, summarising, organising and planning, using material you give it.
The fix: pick one real task from this week. A message you have been putting off, notes that need tidying, a plan you need to structure, a long document you need summarised.
One honest limit: the plan you use matters too. As of 27 October 2026, OpenAI’s help pages say that on the free plan, file uploads, data analysis and image creation have separate limits that can be stricter than on paid plans. Other tools differ, and plans change often. If a free tool disappointed you, the plan may be part of the reason, but start with the five fixes above before you spend money.
Try this: “Here is a message I need to send to a parent about a late fee. Rewrite it so it is polite, clear and firm, in under 120 words.”
A five-minute retry
Before you give up for good, try this once:
- Pick a real task you did this week.
- Write a prompt that says who you are, who it is for, what you need and what shape it should take.
- Reply to the first answer at least twice: shorter, warmer, clearer, whatever it needs.
- Check every name, number and date before you use anything.
If it still does not help after that, you have lost five minutes and learned something real. Most people find the second try is a different experience.
What I cannot promise
These prompts are standard, widely recommended ways of writing, not a secret formula. I have not tested the same prompts across every tool, and results vary by tool, plan and day. AI will still sometimes be wrong. That is why a person stays in charge.
The takeaway
A first try that fails usually says more about the prompt than about the tool. Give context, ask again, go step by step, check the facts and choose a real task. That is how AI becomes a thinking partner, not a shortcut.
Would you like your team to learn this properly, with practice on your own real work? Hillward can help. See how at hillwardconsults.com.
One useful idea, every Thursday.
The Hillward weekly note: a short tip, tool or lesson in plain English. Join below.
Sources
- OpenAI Help Center, “Prompt engineering best practices for ChatGPT” (checked 27 October 2026): https://help.openai.com/en/articles/10032626-prompt-engineering-best-practices-for-chatgpt
- Anthropic, “Prompting best practices” (Claude Platform Docs): https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices
- Anthropic, “Prompt engineering overview”: https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overview
- OpenAI, “Why language models hallucinate”: https://openai.com/index/why-language-models-hallucinate/
- Anthropic, “Reduce hallucinations”: https://platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/reduce-hallucinations
- OpenAI Help Center, “ChatGPT Free Tier FAQ” (checked 27 October 2026; this page changes often): https://help.openai.com/en/articles/9275245-chatgpt-free-tier-faq



