What this covers
Most “bad” ChatGPT output is a missing-instruction problem, not a model problem. Daily users hit a plateau where every response feels mid because they’re sending the same 30-word ask and hoping. This guide is the 10-minute upgrade: four levers you can add to any prompt, two cargo-cult ideas to drop, and a template that survives across tasks. The goal is not “perfect prompts” — it is going from 1-in-5 usable responses to 4-in-5 on your real work.
Who this is for
Anyone using ChatGPT daily — writers, PMs, marketers, students, devs — who feels outputs are mid and can’t tell whether the problem is the model, the prompt, or the task.
When to reach for it
- You keep tweaking the same response without much improvement.
- You’re getting fluent-but-wrong output (generic, off-tone, or technically incorrect).
- You can’t articulate why a “good” answer would be good — that’s the actual gap.
The four levers (add these, in this order)
- Audience: who is reading this output and what do they already know? “Explain to a CTO who already knows Kubernetes” gives radically different output from “Explain to a marketer.”
- Format: bullets, table, short paragraph, JSON, headline + 3 sentences. The model defaults to flowing prose, which is almost never what you need.
- Constraint: length cap, tone, vocabulary to avoid, what to NOT do. “Under 120 words” and “no marketing adjectives” are the two highest-leverage constraints.
- Example: paste one paragraph of “good” output (yours or someone else’s). One real example beats two paragraphs of abstract instruction.
Before and after
WEAK PROMPT
Write a launch email for our new feature.
UPGRADED PROMPT
Write a launch email for our new feature.
Audience: existing users who already pay us $20/mo.
Format: subject line + 3 short paragraphs + 1 CTA button label.
Constraint: under 150 words, no exclamation marks, no "excited to announce".
Example tone (one paragraph from our last launch):
"We pushed a small change today: ..."The upgraded version routinely produces output you can ship after a 5-minute edit.
Step by step
- Write your first draft prompt as you normally would.
- Read it back and ask: “Could this have been written by 50 different people with 50 different audiences?” If yes, you’re missing audience.
- Add audience, format, constraint, example — in that order. Stop at “good enough,” not “perfect.”
- Run it. If the output is wrong, change ONE variable and re-run. Changing three things at once teaches you nothing.
First-run exercise
- Take a task you ran last week with mid output. Save the original prompt and the original response.
- Rewrite the prompt with all four levers. Run it. Save the new response.
- Diff the two responses — what got better, what got worse, what stayed the same.
- Run the new prompt again 24 hours later on a different but similar task. If it generalizes, lock it in as a template.
Quality check
- Does the output answer the actual question, or does it answer a more-flattering version of the question? LLMs love restating the prompt charitably.
- Are facts verifiable, or is it confident-sounding mush?
- Would your reader actually use this, or just nod politely?
How to reuse this workflow
- Maintain a
prompts/folder in your notes. One file per recurring task. The 4 levers go in a header comment so you can edit confidently. - For team work, share the prompt file alongside the deliverable. New hires get up to speed faster.
- Re-test your templates quarterly. Model updates change defaults; what worked on GPT-5.5 may need adjustment on a newer model.
Recommended workflow
Audience + format + constraint + example → run → iterate one variable at a time → save the winning template with a comment explaining why it works.
Things to drop
- “Please” and “thank you”: neutral on output quality; harmless.
- “You are an expert in X” without specifics: the model is already an “expert”; the persona only helps when paired with concrete behavior (“respond like the author of the book Refactoring would”).
- 5-paragraph context dumps: past 500 words of preamble, model recall on the actual question drops. Move details into Project files or a paste-in attachment.
- Multiple goals in one prompt: “summarize AND translate AND format as a table” produces something that does none of the three well. Chain them.
Common mistakes
- Dumping all constraints in one wall-of-text paragraph — the model averages them rather than satisfying them.
- No format specified — you get prose when you needed a table.
- No example of “good” output — abstract style guidance loses to one concrete sample every time.
- Asking for “best practices” — vague request, vague response. Ask for “the 3 trade-offs most teams get wrong” instead.
- Treating prompt-engineering as a one-shot problem — the actual workflow is iterate-and-save, not write-and-pray.
- Copying viral “magic prompts” from Twitter without adapting audience and constraints — they were tuned for someone else’s task.
FAQ
- Does prompt length help?: Up to a point. ~300-600 words is the sweet spot for most tasks. Past ~1000 words, the model’s recall on early instructions degrades.
- Should I use a “system prompt” / Project Instructions?: Yes, for anything recurring. Put role, tone, and constraints there. Keep per-chat prompts focused on this specific ask.
- Do “tip the model $200” tricks work?: They were always marginal; on current models they’re noise. Skip them.
- Will newer models make this obsolete?: Better models reduce some prompt sensitivity, but audience and format will always matter — the model can’t read your mind.