You ask for help drafting a difficult email and get three cheerful paragraphs that say nothing useful. You ask for a summary and get something longer than the original document. The tool is capable; the request was vague. That gap explains most of the frustration people feel in their first weeks of working with chat assistants.
The fixes below need no settings, no technical setup, and no special vocabulary. They are editing decisions. Here are eight common mistakes, each with a before-and-after version and the smallest change that solves it.
Mistake 1: A goal too broad to hit
A topic is not a task. This mistake is the easiest to spot and the most common.
Before: Write about customer churn.
After: Explain the three most common reasons small subscription businesses lose customers in their first year, in 250 words, for a founder who already knows the basics. Skip the definition of churn.
The second version names a reader, a length, a level, and something to leave out. "Write about churn" gives no way to be wrong, so the answer drifts toward the safest, flattest version of the subject: definitions and generalities you already knew.
Notice what the fix cost: about one extra sentence. Beginners often assume that a longer, more detailed prompt takes longer to write than the time it saves. In practice it is the reverse. Thirty seconds spent naming your reader and your limit replaces three rounds of reading answers that miss.
When you genuinely don't know what you want, say that and ask for options. "Give me three possible angles on this, one line each, and I'll pick one."
Mistake 2: The context stays in your head
You know the client is annoyed, the deadline moved twice, and the last version was rejected for sounding too formal. None of that is in the prompt, so none of it is in the answer.
Before: Rewrite this email so it's better.
After: Rewrite this email for a client whose invoice was two weeks late. They have already complained once. Keep it under 120 words, apologetic without grovelling, and end with one specific next step and a date.
A quick test: anything you would tell a new colleague before they started the task belongs in the message. Situation, reader, constraint, deadline, what went wrong last time.
The word "better" is the trap here. It means "closer to what I have in mind", and nobody can see what is in your mind. Replace it with the two or three qualities you actually care about, and the answer stops guessing.
Mistake 3: Four jobs jammed into one message
"Summarise this report, turn it into a slide outline, suggest questions for the meeting, and keep it light." That is four requests, and the last one changes the tone of the other three. When jobs are bundled, you cannot tell which part failed, and fixing one means regenerating all of them.
Send them one at a time. Read the summary before you ask for slides. Read the slides before you ask for questions. Sequential requests also let you catch a wrong assumption early instead of inheriting it through three rounds of output.
Mistake 4: No shape specified for the answer
Say nothing about format, and you get a wall of friendly prose. Most assistants default to a few paragraphs, which is rarely what you need. Put the container in the prompt:
- Length: "250 words", "five bullets", "one paragraph, no preamble"
- Format: table, numbered steps, headings only, plain prose
- Voice: "direct, no adjectives", "warm but professional", "as if explaining to a curious fifteen-year-old"
- Exclusions: "no bullet points", "don't repeat my question", "no closing summary"
If you are unsure which shape works, ask for two: "one version as five bullets, one as a single paragraph."
Skip the private shorthand
"Fix the Q3 onboarding flow" assumes a shared vocabulary. Acronyms, project names, and internal labels mean nothing outside your team, so the answer guesses. Define it once, in brackets: "the onboarding flow (the five welcome screens new users see)".
Mistake 5: Expecting memory you never provided
Two things get confused here. Inside a single conversation, earlier messages usually carry over, but long threads get compressed and details fade. Across separate conversations, nothing carries over unless a memory feature is switched on and has stored it. Documents you did not paste or attach are invisible, however well known they are.
So paste the source text, or write two lines of background yourself. "Here is the meeting note I'm working from:" followed by the text costs you a copy and a paste. Describing the note instead is where errors creep in, because your description quietly drops the awkward details that make the summary useful.
If a detail matters, put it in the current message. "Remember the budget is capped at £4,000" is worth repeating even if you said it twenty messages ago.
Mistake 6: Starting over instead of correcting
When an answer is nearly right, most people throw the whole prompt away and rewrite it from scratch. That discards the parts that were working.
Before: No, that's wrong. Try again.
After: The second paragraph says we ship in 48 hours; the correct figure is five business days. Keep everything else exactly as it is and change only that.
"That's wrong" tells nobody which sentence to touch. Point at the specific line, say what should stand in its place, and say what to leave alone. Corrections are cheap; restarts are not.
Mistake 7: Reading fluency as if it were evidence
Confident phrasing and accurate content are two different things. The same smooth rhythm appears whether the answer is drawn from a document you pasted or invented to fill a gap. A precise-sounding number with no source behind it is the clearest warning sign.
Two habits help. First, keep the facts in the conversation: paste the source, and ask for an answer based only on it. Second, make uncertainty visible. "If the source doesn't state a figure, write 'not in the document' instead of estimating." Then read the answer with the same suspicion you would apply to a stranger's first draft. You are still the one who signs it off.
Mistake 8: Quitting after one bad answer
One vague prompt produces one vague answer, and many beginners conclude the tool simply cannot do this kind of work. The first reply is a draft of your request as much as a draft of the answer. If it misunderstood, that is information about what your prompt left out.
Name the miss and ask again once. "You gave me general advice; I need the specific steps for our five-screen flow." Two rounds of narrowing usually beats one round of over-describing, because you can only see what was missing after you have read something.
The five-line pattern that prevents most of these
Once you have fixed a few prompts, a shape emerges. Almost every effective request answers five questions, and you can answer them in any order:
- Task: one verb, one job
- Reader: who it is for and what they already know
- Shape: length, format, tone
- Source: the text, notes, or figures to work from, pasted in full
- Exclusions: what to skip, and what not to change
Filled in, it reads like this: "Draft a 150-word follow-up email to a customer who reported a broken hinge. They are practical, not sentimental. Use the order number and replacement date below. No apology paragraph, no exclamation marks."
It is not elegant prose, and it is not meant to be. It is a work order.
Before you hit send: a thirty-second check
- Could a stranger complete this task from my message alone?
- Is there more than one job in here?
- Did I say how long, and in what shape?
- Did I paste everything I referred to?
- Did I say what to leave out?
Five questions, thirty seconds, and most of the frustration in this list disappears. Prompting is not a secret skill you either have or lack. It is the ordinary work of saying clearly what you want, which is difficult for humans in every medium, and which improves the same way everywhere: by noticing what you left out and adding it next time.
Photo: Kampus Production / Pexels
