Custom GPT Prompt Engineer
Turn proven prompt templates into reusable assistants that interview and draft for you.
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 98%
Take a prompt that works reliably, separate its reusable structure from its one-time content, and place that structure inside a custom GPT or similar assistant. Instruct the assistant to follow the template exactly and to collect whatever inputs are missing for each new task. The user can then describe an objective in ordinary language, answer clarifying questions, and receive a complete prompt containing the required goal, format, safeguards, and context placeholders. The generated prompt can be refined conversationally by asking for more detail or changing selected sections. This converts prompt engineering from a memory-dependent craft into a repeatable interface: the template holds the expertise, the assistant performs intake and assembly, and the user supplies task-specific judgment.
Origin
Extracted from Marketing Against The Grain, where Kieran demonstrates custom GPTs trained to reproduce his preferred O1 and deep-research prompt templates.
Core principles
- 01Preserve a proven prompt structure in a reusable template.
- 02Let an assistant gather inputs instead of recalling every field manually.
- 03Use conversation to clarify the task before generating the final prompt.
- 04Iterate on the generated prompt rather than treating version one as final.
How to run it
- 1
Select a proven template
Start with a prompt structure that has already produced useful results. Identify its stable sections and quality requirements.
Pro tip Choose a template with explicit fields rather than an unstructured block of prose.
Watch out Automating a weak prompt only reproduces weak results more consistently.
- 2
Encode the template
Create a custom assistant and instruct it to reproduce the template for every relevant request. Preserve required sections, ordering, and constraints.
Pro tip Explain why each required section matters so the assistant can handle unusual requests.
Watch out Do not let optional conversational flourishes replace mandatory fields.
- 3
Define the intake
Tell the assistant which task-specific details it must obtain before drafting. Have it ask clarifying questions when essential information is absent.
Pro tip Keep the initial intake short, then ask follow-ups only where answers materially affect the prompt.
Watch out Generating before collecting critical context creates polished but generic prompts.
- 4
Describe the task naturally
Give the assistant a plain-language account of the desired output, audience, constraints, and available inputs. Let it translate that conversation into the formal template.
Pro tip Include rough ideas even if the wording is incomplete; the assistant’s role is to structure them.
Watch out Do not assume the assistant knows private organizational context.
- 5
Generate and inspect
Ask for the full prompt and check that every required template component appears. Correct inaccurate assumptions before using it in another model.
Pro tip Review the generated prompt as a specification, not merely as prose.
Watch out A long prompt is not necessarily a complete or effective prompt.
- 6
Iterate the prompt
Request specific changes such as greater detail, narrower scope, or stronger verification. Save durable improvements back into the reusable template when appropriate.
Pro tip Name the sections you want retained as well as those you want changed.
Watch out Unfocused requests to “make it better” may add length without improving utility.
In the wild
An executive tells the custom GPT that they need a two-page memo about how AI will change software sales and that internal and external data will be provided. The assistant converts those sentences into a structured prompt with an outcome, return format, scope, warnings, and context requirements.
→ The executive receives an advanced prompt without reconstructing the template manually.
A marketer describes a need to compare AI SDR tools for open-rate improvements. The assistant asks for industry, target customer, and integration requirements, then generates a detailed research prompt.
→ A short conversation becomes a reusable, structured research specification.
Common mistakes
Automating before validating
A custom assistant should reproduce a prompt known to work. Otherwise it institutionalizes untested assumptions.
Skipping clarifying questions
The assistant cannot create a tailored prompt if it never gathers the audience, objective, constraints, or context.
Relying on it without learning
Beginners should still draft and inspect prompts themselves so they can recognize omissions and improve the generated version.
Is it for you?
Best for
It is best for individuals or teams that repeatedly build prompts from the same underlying structure.
Not ideal for
It is not ideal when the underlying template is untested, unstable, or too task-specific to reuse.
From the transcript
“I'm constantly taking prompts and then turning them into templates and then creating custom GPTs to replicate that template for me.”
“It's making sure you follow that template every single time.”
“I don't think you want to rely solely on AIs, but if you are like a novice at Prompton, doing what I just showed you…”
From the episode
Use This AI Trick To Get 10x Better Results Every Time