Narrate-to-Prompt Delegation Method
Describe a recurring task and let AI design its reusable prompt
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 7
- Confidence
- 98%
The Narrate-to-Prompt method uses AI as a prompt designer rather than requiring the operator to build instructions manually. The user selects a recurring task, speaks through the current process, explains the desired future state, and describes relevant constraints and examples. AI converts that narrative into a structured reusable prompt. The operator tests it against a real case, optionally asks another model to critique or improve it, and updates it based on failures. Because the prompt originates in the user's own workflow, it is more context-specific than templates scavenged from social media. The approach also surfaces unspoken assumptions and may reveal that the task lacks a coherent process, providing a foundation for broader workflow redesign.
Origin
Barbara Jovanovic said she does not manually write many of the prompts used for content and agency management; she explains the job to ChatGPT and has it produce the prompt. Extracted from Marketing Against The Grain.
Core principles
- 01Users do not need to handcraft every prompt
- 02A real process is better input than a borrowed prompt template
- 03Current-state and desired-state descriptions expose requirements
- 04Cross-checking with another model can reveal missing constraints
- 05The best prompt is adapted to the user's unique workflow
How to run it
- 1
Select a recurring task
Choose work performed often enough that a reusable prompt would save meaningful time.
Pro tip Start with a task whose quality you already know how to judge.
Watch out Automating an unstable or rare task may not repay the setup effort.
- 2
Narrate the current process
Record yourself explaining each action, input, decision, exception, and output involved today.
Pro tip Speak naturally so tacit context is less likely to be omitted.
Watch out Do not hide manual judgments that the future workflow must preserve.
- 3
Describe the desired state
Explain what should become faster, automated, delegated, or higher quality.
Pro tip State non-negotiable constraints and examples of acceptable outcomes.
Watch out Efficiency without a defined quality threshold can accelerate poor work.
- 4
Generate the prompt
Ask AI to turn the narration into a complete reusable instruction for performing the task.
Pro tip Invite the model to ask for missing information before finalizing the prompt.
Watch out The first generated prompt is a draft, not a proven procedure.
- 5
Cross-check the design
Optionally ask a second model to identify ambiguity, omissions, or unsafe assumptions.
Pro tip Compare critiques rather than automatically combining every suggestion.
Watch out Multiple models can reinforce the same misunderstanding.
- 6
Test on real work
Run the prompt on a representative task and compare the result with your normal standard.
Pro tip Use difficult edge cases as well as the common path.
Watch out A single successful run does not demonstrate reliability.
- 7
Convert failures into instructions
Revise the prompt whenever testing reveals a missing constraint or decision rule.
Pro tip Preserve version history for prompts used in important workflows.
Watch out Avoid adding conflicting patches without simplifying the whole prompt periodically.
In the wild
An agency owner records how a recurring management task is currently handled, what information it consumes, and what result is needed. ChatGPT converts that explanation into a reusable prompt, and Claude is used to cross-check it before testing.
→ The owner receives a workflow-specific prompt without searching for a generic template.
Common mistakes
Searching before explaining
Borrowed prompts may omit the specific constraints, language, and decisions that make the user's workflow work.
Narrating only the happy path
A prompt built without exceptions or quality checks may fail as soon as the task varies.
Skipping real-world testing
A well-structured prompt can still produce unsuitable results when applied to actual inputs.
Is it for you?
Best for
People converting familiar, repeatable work into AI-assisted procedures.
Not ideal for
Tasks the user cannot yet explain, evaluate, or distinguish from an acceptable result.
From the transcript
“I don't write them. I tell Chat GPT what I need and then it writes the prompt for me.”
“If it's a task that you do often, that you want to outsource, automate whatever you need, sit down, record yourself talking through what your…”
“Like just tell it what you needed to do and then we'll create the prompt for you.”
From the episode
I Run a 6-Figure Agency with ZERO Employees (AI Only)
Barbara Jovanovic