Prompting
Common AI prompt mistakes to avoid
Most disappointing AI output traces back to a few small mistakes in the prompt. Fix these and your default quality jumps — without changing models or tools.
1. Vague audience
"Write a marketing email" produces a generic email. "Write a 90-word email to busy ops managers at 50–200 person SaaS companies, mentioning a 30-minute setup time" gives the model something to work with.
2. No format constraint
Always specify length, structure, and tone. "5 bullets, max 12 words each, no emojis" is more useful than "concise and professional."
3. Skipping examples
Two examples of the output you want will outperform any abstract instruction. Paste one good and one bad example and ask the model to match the good one.
4. One mega-prompt instead of a chain
Break the task into stages. Each stage gets its own prompt and its own review:
- Outline the structure
- Draft each section
- Tighten the draft
- Format and export
5. No voice sample
For anything that should sound like you, paste two or three real samples and ask the model to match them. This single change removes most of the 'AI smell.'
6. Polite suggestions instead of constraints
"Try to be brief" is a wish. "Maximum 80 words" is a constraint. Models follow constraints far more reliably than preferences.
7. Not saving what works
Every prompt you run twice should live in your dashboard. Saving prompts is the difference between getting good at AI and re-typing the same instructions for years.
8. Asking for opinions instead of options
"What headline should I use?" produces a guess. "Generate 10 headline variants in the PAS framework, all under 9 words" produces something to choose from.
9. Trusting the first draft
The first draft is rarely the best. Run a critique pass: "List 3 weaknesses in this draft and rewrite to fix them." Quality jumps without changing the prompt.
10. Treating AI like a search engine
AI is a thinking partner, not a citation. Verify any claim, statistic, or quote before publishing. Use AI for structure and speed; use sources for facts.
Recommended workflow
Tools used in this workflow
Frequently asked questions
Why does my AI output keep sounding generic?
You're giving generic input. Specific audience, specific format, specific constraints — change those three and the output stops sounding like a press release.
Should I write longer or shorter prompts?
Longer when context matters (audience, voice, examples). Shorter when the task is simple. The wrong move is medium-length prompts full of vague adjectives.
Is it worth giving the AI examples?
Almost always. Two or three examples of the output you want will improve quality more than any clever instruction.
How do I stop AI from adding unnecessary preambles?
End the prompt with 'Reply with only the [output]. No preamble, no explanation.' Direct constraints work better than polite requests.
What's the single biggest prompting mistake?
Asking for one big result instead of running a small chain. Break tasks into outline → draft → edit → format. Each step is dramatically better than one mega-prompt.
Continue reading
Recommended stack for this workflow
Tools we'd reach for next — value first, never a popup.
Claude Pro
Long-context reasoning and analysis when the workflow gets dense.
Copy.ai
Workflow-based AI writing built around GTM teams.
Jasper
Brand-aware long-form drafting when one output isn't enough.
Some links on this page are partner recommendations. If you sign up we may earn a small commission — at no cost to you. We only recommend tools we'd use ourselves.