IntermediateGeneralProductivity
Five prompt patterns that get consistent results in any AI
Once you use AI every week, the problem is usually consistency rather than getting an answer at all. These five patterns work in any chat app and turn hit-or-miss prompts into ones you can reuse.
- Time
- About 15 minutes
- Checked
If a button name on the screen does not match this guide, follow the action. These products rename buttons often.
Steps
Pattern 1: Goal, audience, and limits
Most weak answers come from missing context. Say what the result is for, who reads it, and the limits.
Prompt to paste
Goal: [explain our new refund policy]. Audience: [customers who are annoyed]. Limits: [under 100 words, no legal jargon, one apology max].
Pattern 2: Show an example
One good example beats a paragraph of description. Tell the AI to match its style, not its facts.
Prompt to paste
Here is an example I like: """[paste example]""" Write a new one about [topic] in the same style and length. Don't copy its facts.
Pattern 3: Ask for a fixed format
Name the exact structure: a table with named columns, numbered steps, or JSON with given keys. Consistent format makes results easy to compare and reuse.
Prompt to paste
Return a table with columns: Option | Cost | Biggest risk | Best for. Exactly 4 rows. No text before or after the table.
Pattern 4: Ask it to think through, then answer
For decisions or tricky problems, ask the model to list its considerations first. Newer models also have built-in reasoning or "thinking" modes; turn them on for hard tasks.
Prompt to paste
Before answering, list the 3 to 5 factors that matter most for [decision]. Then give your recommendation and the one thing that would change it.
Pattern 5: Add a self-check
Ask the model to check its answer against your rules before it finishes. It doesn't catch everything, but it catches a lot.
Prompt to paste
After drafting, check: Is it under [100] words? Does it mention [required point]? Does it avoid [banned word]? Fix anything that fails, then show only the final version.
Save your best prompts
Keep winning prompts in a note with brackets for the parts that change. Or put them in a Claude Project, a ChatGPT Project, a custom GPT, or a Gemini Gem so they apply automatically.
Tips
Change one thing at a time when a prompt isn't working, so you know what helped.
Put long pasted material in quotes or between clear markers like three quotation marks.
If you need facts, give the source text and ask it to answer only from that.
Common questions
Do these work in every AI app?
Yes. They are plain instructions, not product features, so they work in Claude, ChatGPT, Gemini, Grok, and others.
Do I need 'act as an expert' role prompts?
A role can help set tone, but clear goals, examples, and formats usually matter more.
Why does the same prompt give different answers?
Models generate fresh text each time. Examples, fixed formats, and self-checks narrow the variation.
More guides
IntermediateClaudeProductivity
Set up a Claude Project so Claude always knows your context
Create a Claude Project, add instructions and reference files once, and get on-brief answers in every chat without pasting context again.
About 15 minutes using Claude Projects
IntermediateChatGPTProductivity
Build a custom GPT for a task you repeat
Use the ChatGPT GPT builder to set instructions, add knowledge files, test in Preview, and share a custom GPT with your team.
About 20 minutes using ChatGPT GPTs
AdvancedGeneralBuilding
Write simple evals to test your LLM app before you ship changes
Build a small eval set, score outputs with code checks and an LLM grader, and rerun it after every prompt or model change to catch regressions.
About 40 minutes using Claude API docs, OpenAI API docs
Occasional notes
Hear when a new guide is up
A short note when a guide is added. No account, and no emails until this form is connected to a real list.