Workflows that work.
Five copyable sequences of prompts, three to five steps each, for writing, research, coding, planning and learning.
Most people use a chatbot the way they use a search box: one question, one answer, done. People who get consistent results do something different. They run the same short sequence of prompts every time they do a certain kind of work, the way a cook follows the same order of steps for a stock.
Here are five sequences you can copy today. Each has three to five prompts. The words in brackets are the parts you change.
Writing: brief, outline, draft, edit for a reader
The mistake is asking for the finished piece in one go. Separate the thinking from the writing.
1. Brief.
I need to write [a short update to parents at my child's school about the new pickup process]. Reader: busy parents skimming on a phone. Goal: they know what changes on Monday and what they need to do. Tone: calm and friendly. Before writing anything, ask me what you need to know.
2. Outline. Give me a short outline: three or four points in order, one line each. Reorder or cut before moving on. It is cheaper to fix an outline than a draft.
3. Draft. Write the draft from this outline. Under 150 words.
4. Edit for a named reader. Now read it as a parent who is late and reading this in the car. What would they miss or misread? Fix those things and show me the final version.
The last step is the one people skip. Naming a specific reader for the edit pass changes what the model looks for.
Research: questions, sources, synthesis, gaps
Research is where models invent most, so this sequence builds the checking in.
1. Questions first.
I want to understand [whether a heat pump makes sense for an older house in a cold climate]. Before answering, list the five questions I would need answered to decide. Nothing else yet.
2. Sources. For each question, find sources I can open. Give the title, who published it and a one-line summary. Say which ones you are least sure about. This works best in a tool that can search the web. If yours cannot, ask it what kinds of sources to look for and search yourself.
3. Synthesis with claims tied to sources. Write a one-page synthesis. Every factual claim gets a number in brackets pointing to its source. No claim without a source.
4. Gaps. What is missing? List what you could not find, where the sources disagree, and what I should check myself.
The synthesis will sound complete even when the sources were thin. Step four is how you find out.
Coding: explain, write with tests, debug with the error pasted
You do not need to be a programmer for this one. The same sequence works for a spreadsheet formula, a short script or a snippet someone sent you.
1. Explain. Explain what this code does, line by line, for someone who does not program. Paste the code underneath.
2. Write, with tests.
Write a [Python] function that [takes a list of file names and returns the ones changed in the last seven days]. Include a few test cases that show it working on normal input and on an empty list, and tell me how to run them.
A test case is a small check that runs the code with a known input and confirms the output. Asking for tests up front makes the model think about edge cases before you hit them.
3. Debug with the error pasted. I ran it and got this error: [paste the exact text]. What went wrong, and what is the fix? Change only what is needed. Paste the error word for word, not a description of it. The exact wording is what the model matches against.
4. Understand the fix. Explain in plain words why it broke, so I recognize it next time.
Planning: constraints, options, a plan with the first step
Planning prompts fail when the constraints live in your head. Get them on the page first.
1. Constraints.
I am planning [a move across town in six weeks]. Constraints: [a budget of X, must be out by the 30th, no car, two cats]. Repeat the constraints back to me and tell me which one will be hardest to satisfy.
2. Options with trade-offs. Give me three different ways to approach this, one paragraph each, with the main trade-off of each. A trade-off is what you give up to get something else; asking for it stops the model from pretending one option is free.
3. The plan. I will take option two. Turn it into a plan with dates. For each step, say what could go wrong.
4. The first step. What is the single thing I should do today?
Ask for the first step on its own. Plans fail at the start far more often than in the middle.
Learning: explain, examples, quiz, what did I get wrong
This is the sequence that turns a chatbot into a patient tutor.
1. Explain for my level.
Explain [how compound interest works] to someone who [is comfortable with percentages but has never had savings]. Keep it to one screen.
2. Examples. Give me three examples from everyday life, from simplest to hardest.
3. Quiz me. Quiz me with five questions, one at a time. Wait for my answer before showing the next. Answer honestly. Guessing teaches the model nothing about you.
4. What did I get wrong. Which ones did I miss, and what does each mistake suggest I have misunderstood? Re-explain only those parts.
The fourth prompt is the whole point. Instead of re-reading the same explanation, you get one shaped around your particular gaps.
Make them yours
Copy these into a notes file and swap the bracketed parts. After a few runs you will notice the step you always change, or the one you always add. That is where your own version begins. Keep the shape (set up, produce, check) and let the wording drift toward how you actually talk.
Try it
Pick the learning sequence and run it on something you half understand. Start with Explain how a mortgage amortization schedule works to someone who has never had a loan. Keep it to one screen. Then Quiz me with five questions, one at a time. Answer each one without looking anything up. Finish with Which did I get wrong, and what does that suggest I misunderstood? Re-explain only that. Look for whether the final explanation is actually different from the first one. If it is, the sequence worked.
Key terms
- Workflow
- A fixed sequence of prompts you run the same way each time for a certain kind of task.
- Brief
- A short statement of what you need, who it is for, what it should achieve and what tone it should take, given before any writing starts.
- Synthesis
- A short piece that pulls several sources together into one account, with each claim tied back to where it came from.
- Test case
- A small check that runs code with a known input and confirms it gives the expected output.
- Trade-off
- What you give up in one option to gain something in another.
- Edit pass
- A separate step where a draft is reviewed and fixed against a stated reader or goal, rather than written and sent in one go.
Check yourself
Stop asking one question; run a short sequence you trust, keep the sequence, and improve it each time you use it.