Writing Assistant Trainer

Train the knowledge files until the GPT sounds like you.

Check weak document types, generate a test prompt, paste the Custom GPT output back here, save your improved version and notes, then update the six knowledge files for manual upload.

Workflow Knowledge Trainer GTP Training Loop Progress Run Quality Check Download Knowledge

Privacy check

Do not upload private, sensitive, financial, legal, credential or personal identification documents into a Custom GPT unless you are comfortable doing so.

Knowledge Trainer

Generate a prompt, test it, then save the better version.

Choose a weak document type, copy the generated prompt into your Custom GPT, paste the GPT output back here, then save your improved version and notes so the knowledge files get stronger.

Update knowledge after saving

Required document type Date added

Scenario

Selected scenario and context

Prompt to copy into Custom GPT

Paste Custom GPT output here Paste your improved version below Notes, corrections or banned patterns

Analysis Preview

What will be added to the knowledge files

Choose a document type and type a draft to see the analysis.

Direct Writing Samples

Improved versions saved into knowledge

These are counted in GTP Training Loop Progress.

Job Application Rules

Main rulebook for CVs, cover letters and applications.

These rules come before VoiceClone. VoiceClone can improve tone and clarity, but it must not override CV structure, cover letter formatting, ATS needs, file naming or factual accuracy.

Build a prompt

Job application rules override VoiceClone tone rules whenever there is a conflict.

If VoiceClone suggests a longer answer but CV rules require one page, keep the CV one page.
If VoiceClone suggests a natural paragraph but CV rules require bullet points, follow the CV rule.
If VoiceClone suggests extra detail but the documents do not support it, do not add it.
If VoiceClone suggests a softer tone but job rules require ATS keywords, keep truthful ATS keywords.

Use with my Custom GPT

Build a job application prompt.

Output type Use VoiceClone tone layer

Job description Relevant experience or CV notes

Rule Conflict Check

Job application rules override tone preferences.
Factual accuracy overrides style.
CV structure overrides natural-flow preferences.
File naming rules are mandatory.
Cover letter formatting rules are mandatory.

Generated prompt preview

Feedback Loop

Save GPT test results and turn better outputs into examples.

Capture the original GPT output, your feedback and the improved version, then re-export the writing system so the Custom GPT can learn from the correction.

Reusable Example

Superstore Power BI LinkedIn Post Example

Use this as a ready-made feedback pair when testing the Custom GPT with the Superstore Power BI LinkedIn prompt.

Save GPT Test Result

Store the test, feedback and improved version.

Test prompt Original GPT output My feedback Improved version

Category Content type Date added

What made the improved version better

Writing Samples

Saved improved outputs

Improved versions saved as writing samples will appear in the export and category filters.

GTP Training Loop Progress

Track enough examples without over-training.

Monitor feedback pairs by content type, see where the writing system is strong enough, and choose the next useful test.

Download Knowledge

Next best action

What to test next

Save feedback pairs to unlock training guidance.

Training Loop Instructions

Repeat this loop until each document type is sufficient.

Check which document types are lacking.
Choose that type in Knowledge Trainer.
Copy the generated prompt into your Custom GPT.
Paste the Custom GPT output back into Knowledge Trainer.
Paste your improved version and notes.
Save and update the knowledge files.
Manually upload the updated knowledge files to your Custom GPT.

Minimum Useful Loop Target

Aim for enough evidence, not endless tweaking.

LinkedIn posts: 3 to 5 feedback pairs
Interview answers: 3 to 5 feedback pairs
README: 2 to 3 feedback pairs
Emails: 2 to 3 feedback pairs

You do not need to loop forever. Stop once the content type reaches Good enough, unless future outputs are still bad.

Starter Training Tasks

First tests to strengthen the tone system.

Reusable Template

Clean KPI Card LinkedIn Post

Save this polished feedback pair as a strong LinkedIn post example for the training loop.

Content type

LinkedIn post

Focus

KPI card formatting under slicers, custom formatted DAX measures and what the fix taught you.

Ask ChatGPT to Review This Output

Generate a review prompt for another ChatGPT chat.

Save ChatGPT Review Result

Paste the review result and save it into the loop.

Quality Check

Review export safety before downloading.

Scan the generated markdown for weak content, sensitive data patterns, writing quality risks and Custom GPT readiness. Warnings never block export.

Safe checks 0

Needs review 0

Risky checks 0

Run the quality check before you download or upload files.

Setup Health Check

Check whether the export has enough saved data.

Review the writing samples, style rules, prompt templates, feedback pairs, job rules and tone layer before downloading Custom GPT files.

Checking setup...

Preview

tone-guide.md

Custom GPT Setup Guide

Create, upload and test the GPT carefully.

Step 1

Download the export files

Preview the files first.
Use excerpts only if the writing examples include personal or sensitive details.
Download all files.

Step 2

Create the Custom GPT

Go to ChatGPT.
Open Explore GPTs.
Click Create.
Open the Configure tab.
Name it Writing Assistant Trainer Demo or a similar name.
Open custom-gpt-instructions.md.
Copy everything inside it.
Paste it into the GPT Instructions box.

Step 3

Upload the Knowledge files

Upload these markdown files into the GPT Knowledge section:

tone-guide.md
style-rules.md
prompt-library.md
writing-examples.md

Optional: custom-gpt-setup-checklist.md. This checklist is mainly for you and is not essential for the GPT.

Step 4

Test the GPT

Test 1

Write a LinkedIn post about my Superstore Power BI project. Make it sound like me.

Test 2

Rewrite this project description in my style: [paste a rough project description]

Test 3

Write a 115-word interview answer for "Tell me about yourself" for a Junior AI Associate role.

Step 5

Improve the GPT

Feedback example

This is too polished. Make it more direct, more natural, and less like a LinkedIn influencer.

If the GPT improves after feedback, save that feedback back into the web app under Feedback Loop. Then re-export the files later.

After re-export

Custom GPT update checklist

Open your Custom GPT settings.
Replace old Knowledge files with updated files.
Upload tone-guide.md.
Upload style-rules.md.
Upload prompt-library.md.
Upload writing-examples.md.
Update Instructions if custom-gpt-instructions.md changed.
Test again with the same prompt.
Save the new result back into Feedback Loop if it improves.