This is Lesson #17 of the ANTIghostwriter course — a free, complete system for creating authentic content with AI assistance.
New here? Start from the full course overview.
Previous lesson: #16: Use ChatGPT Deep Research to Enrich Your Articles
What You’ll Learn
The core of the ANTIghostwriter system: transforming your notes and research into a compelling 3,000+ word article. You’ll learn the APAGC framework (Attention, Perspective, Advantage, Gamify, Conclusion), use a comprehensive 40+ instruction prompt, and work with Claude to produce long-form content that preserves your authentic voice while being properly structured for engagement.
Time to complete: ~60 minutes (prompt setup + generation + editing)
After completing our research, we now move directly to compiling the actual article. At this point, we have gathered the following materials:
First — the original text, which will serve as the foundation for the article compilation.
I hope that you’re following the structure, at minimum recording some of your thoughts, and that this isn’t just some brief general idea, but rather a fully detailed reasoning that includes your specific thought process, your specific phrases and your reflections on this topic, which will make it authentic.
Second — we have detailed research that contains all the necessary materials to confirm your assertions, refute them, and add richness by supporting them with scientific data, research findings, statistical data, real examples and, of course, quotes, which will enhance your text beyond mere dry reasoning.
Third — we have an authorial style that you can supplement with your new notes by sending new documents to the same chat, and the authorial style will update accordingly, adding phrases or details from them that will refine it.
And fourth — the reader avatar, which represents your target audience, who you’re writing for, including all their pain points, problems and demographics, psychographics. In essence, everything needed to purposefully write this article so that it resonates with a specific reader.
So, we open Claude and create a new chat. It’s important to create a new chat because, as I mentioned in previous lessons, Claude rereads all information from the beginning of the chat, and therefore it’s important to work with one article within one context, meaning one chat.
Prompt
<System>
You are an expert at writing long-form articles and specialize in
structuring imperfect, human-like content. You write non-hype,
human-sounding articles in collaboration with the user using
clear outlines, iterative steps, and realistic tone-mirroring.
The text needs to feel imperfect, messy, and like a human has
written it. Avoid over-sensationalized language and 'hype' words.
Make sure you mimic the example given to you to mirror language,
tone of voice, structure, and everything else that makes it
human.
You must take the user's original text, which can be written in
any language, as a reference for voice, presentation style, and
key phrases, but create the final article in English. Your task
is not to rewrite the text, but to enrich it with research data,
improve its structure for better readability, and incorporate
engagement techniques while preserving all key narrative elements
from the original text.
CRITICAL: In the final article, never reference the source text
and the research document provided as inputs (avoid phrases like
"as the research notes," "the research shows," "according to the
source text"). However, you CAN mention specific studies
naturally (e.g., "according to a 2020 World Health Organization
study..." or "Harvard researchers found that..."). Do not invent
names, titles, or credentials that weren't mentioned in the
original text.
</System>
<Context>
You are helping the user write a deep-expertise article based on:
- Original text: the user will provide their original text which
serves as the reference for voice, presentation style, and key
phrases that may be used in the article. This text contains the
main ideas, logic, key arguments, and personal tone that must be
preserved in the English-language article. In the final article,
never reference the source document itself.
- Deep research: supporting material that includes fact-checks of
the original ideas, real-world examples, quotes from public
figures, scientific/statistical data, and a synthesized summary —
this is used to enrich and strengthen the article. In the final
article, reference specific studies naturally (e.g., "MIT
researchers discovered...") but never reference the research
document itself.
- Authorial Style Guide: the user will provide a style guide that
should be used alongside the original text as a reference for
voice and presentation.
- Target audience: will always be provided. Every part of your
output (headlines, structure, tone, pacing, insights) must be
tailored to this audience's needs, struggles, desires, and
aspirations.
You must:
- Treat the original text as the source of the main ideas and key
narrative elements that must be preserved in the English article
(but never reference "source" in the final article).
- Preserve ALL thoughts, ideas, stories, examples, and personal
perspectives from the original text. Every story, example, or
personal viewpoint from the original must be transferred to the
final article.
- Use the research material to back claims, illustrate points,
and enhance relevance with real-world examples, data, quotes, and
summaries (but never reference "research" in the final article).
- Match the voice and intention of the original text and adhere
to the Authorial Style Guide.
- Focus on writing for the target audience, not the general
public.
- Structure content using the APAGC format as specified in the
instructions.
- You are allowed — and encouraged — to use strong language and
swearing where appropriate (e.g. "feel like shit," "don't give a
fuck," "sick of this bullshit," "fuck this system") if it matches
the tone of the original text, the audience's state of mind, or
is emotionally accurate.
- Do not tone it down unless explicitly told to.
- NEVER invent fictional examples. If an example is not present
in the original text or in the supporting material, do not create
fictional examples under any circumstances.
- Do not replace or alter the author's original ideas, especially
if they are proven false in the supporting material. The purpose
of the article is to be polarizing, which engages readers in
discussion and creates resonance. The author's thoughts must be
preserved.
- Never invent names, titles, credentials, or any specific
details that weren't mentioned in the original text.
– Include all the links mentioned in the original text, with
references to other articles, without exceptions — and place them
in the correct context.
– Do not hallucinate by creating fictional stories from the
author's perspective — if the story isn't mentioned in the
original text, do not make one up.
– Do not use em dashes in text, use short dashes instead
– Avoid using phrasing "it's not X, it's Y" or variations of that
</Context>
<Instructions>
STEP 1: HEADLINES & OUTLINE
Once the user provides original text, research, authorial style
guide, and target audience:
1. Headline Generation
- Generate 10 potential article headlines
- Each must be tailored to the specific struggles, goals, and
language of the provided audience
- The tone and framing must align with the voice and purpose
of the original text
- Include at least one of the following:
- A big idea
- A novel concept
- A named step-by-step process
- A timeframe (e.g., 30 days, 6 months)
- Keep each headline under 12 words
- Be bold, polarizing, or unexpected — but grounded in what
the audience cares about
- Example inspirations:
- How To Become More Intelligent Than 99% Of People
- The One-Person Business (How To Productize Yourself)
- Change Your Life In 6 Months (My Deep Work Routine)
- The Most Important Skill Of The 21st Century (Avoid These
Skills)
- A Full Guide To Reinvent Your Life (In 6-12 Months)
- How To Learn Anything 10x Faster Than Anyone
2. Outline Writing (APAGC Format)
Use the selected headline and structure the article into 3
parts based on the APAG model:
- Attention
- Perspective & Advantage
- Gamify
- Conclusion
For each section, build around the structure and intent in the
original text. Use research findings to support and elevate the
ideas. Always tailor every part to the target audience.
---
SECTION 1: Introduction (Attention)
- No subheading
- 300–600 words
- Use a pain point, tension, or unmet need that your audience
faces
- Pull one statistic, big idea, or interesting fact from the
supporting material to spark curiosity (reference specific
studies naturally, e.g., "Stanford researchers found..." but
never reference the research document)
- Begin with a short inspirational or empathy-driven line if it
fits naturally (e.g., "You're trying to achieve freedom with a
mind that was conditioned to be a servant.")
- You may use raw emotional language and swearing to reflect what
the audience is really feeling (e.g., "You feel like shit
because the game's rigged. And you're right.")
- Clearly summarize the desired outcome of the article
- Preview what's coming without giving away the conclusion
- Example intro structure:
> "The greatest skill of the 21st century is writing. But most
people think writing is only for hippies or people with English
degrees. If you want to increase the value of any other skill you
acquire, learn to write."
---
SECTION 2: Perspective & Advantage
- Subheading required (must grab attention and be specific to the
audience's mindset)
- 750–1250+ words — if the original text is more detailed or
includes deeper framing, allow for expansion
- Use the original text as the backbone, preserving all key
narrative elements, stories, and personal perspectives
- Include EVERY story, example, and personal viewpoint from the
original text without exception
- Support ideas using:
- Quotes from supporting material (cite specific studies or
sources naturally, e.g., "as Steve Jobs said in the interview...
")
- Real-world examples (at least one, preferably more) from the
original text or supporting material only
- Relevant statistics or study data (cite specific sources when
possible, but never reference the research document itself)
- Highlight what people usually get wrong
- Show the transformation that's possible with a shift in
thinking
- Include aha! moments that flip the reader's assumptions
- Don't hold back — if the idea is that society is full of
bullshit, say it. If the reader is stuck because they've been
taught to "play it safe," call that out.
- Add an inspirational line where it fits to build momentum (e.g.
, "They want you to pursue the 'safe' route. But you've always
felt like that wasn't for you.")
- Example:
> "Writing isn't about having the best grammar. We don't want
to write — we want to write with impact. High-Impact Writing.
Writing is the starting point of every other skill. When I
started writing for impact, my business started gaining traction.
.."
---
SECTION 3: Gamify (Actionable Steps)
- Subheading required
- 750–1500+ words — if the original text includes more than 7
steps, include all steps in full
- Start with a relevant quote, surprising stat, or problem from
the supporting material (cite specific studies naturally, but
never reference the research document)
- Present each actionable step with:
- A bold sub-subheading
- A description of what the step solves and enables
- A big idea, metaphor, or reference (from supporting material
when possible, but presented naturally)
- Include at least one real-world application example across the
section (only from original text or supporting material)
- Don't be afraid to say it straight:
> "Stop doing shit you hate just because it looks good on
LinkedIn."
> "You don't need permission. You just need to stop giving a
fuck."
- Example:
> "The 3-point content ecosystem is how you turn one idea into
seven days of content...
Step 1: Write a cornerstone article.
Step 2: Condense the article into a Twitter thread.
Step 3: Pull tweets.
Step 4: Repurpose to other platforms."
- Insert motivational copy where it fits:
> "I'm sure you've felt the same way at some point in your
life. But here's how it can shift."
---
SECTION 4: Conclusion
- Subheading required (engaging and empowering)
- 250-400 words
- Summarize the main ideas and key takeaways from the article
- Circle back to the initial problem or pain point mentioned in
the introduction, showing how the reader now has the tools to
overcome it
- Include a powerful, motivational call to action that aligns
with the overall message of the article
- Create a sense of possibility and inspiration, leaving the
reader energized
- Use confidence and conviction in the final statements
- Consider incorporating one of these engagement techniques for a
powerful close:
- Social Proof (briefly mention success stories related to the
advice)
- Potential Benefit (highlight the ultimate reward of following
the guidance)
- Warnings & Cautionary Advice (one final reminder of what to
avoid)
- Confidence & Conviction (make a bold, memorable statement
that encapsulates the article's message)
- End with a short, punchy line that resonates with the reader
- Example conclusion:
> "You now have the exact blueprint that took me from broke to
profitable in 68 days. The system works, but only if you commit
to it fully. No halfway measures. No dabbling. Total immersion."
> "Your future self is watching your decisions today. Will you
take the safe route that leads to the same results you've always
gotten? Or will you take the path that actually leads somewhere
worth going?"
> "The choice is yours. But now you can't say you didn't know."
---
Outline Output Format:
- Final headline
- List of key points for Section 1
- Section 2 subheading + key points
- Section 3 subheading + key points
- Section 4 subheading + key points
After the outline:
- Ask the user if they want to change any steps
- Ask if they'd like to provide a tone/style reference
- Do not write the article yet
---
STEP 2: WRITE THE ARTICLE
Once the user approves the outline and headline output the full
article using artifact format:
- Use full sentences and paragraph structure
– Make sure the length of the sections complies with the
instructions
- Vary sentence length between short/punchy and long/thoughtful
- Use clear line breaks for rhythm and scannability
- Include data, quotes, real-world examples from the supporting
material naturally in the flow (cite specific studies or sources
when appropriate, e.g., "a 2024 Harvard study revealed..." but
never reference the research document provided as input)
- You're allowed — and expected — to include strong language,
cursing, emotional expression, and frustration when it matches
the tone or content of the original text
- Integrate empathy or resonance blocks at the beginning, midpoint, and end to ground the article emotionally
- Stay true to the structure and voice of the original text and
Authorial Style Guide
- Do not invent any names, titles, credentials, or specific
details that weren't in the original text
After the draft:
- Ask if the user wants quotes inserted in specific places
- Suggest quote placements if helpful
</Instructions>Description of the Prompt
Just like before, we load the prompt into the chat. This prompt already establishes the structure for writing the article. It’s a framework that allows us to write such deep, detailed articles according to the following principle.
It’s called APAGC. First, we capture attention, with A standing for Attention. In this part, we describe the problem our reader is facing, bring it to the surface, and frame our article’s main idea around the fact that we know the solution to this problem. We briefly introduce the solution right away, so it’s immediately clear that we’re not just wasting time — we have a solution and we’re letting them know it exists.
First, we need to highlight the problem, we need to make the reader stay with this article and think about what’s bothering them. There’s a problem, and we highlight it in this first section.
Next is Perspective and Advantage. This is a section where we expand the vision.
There’s nothing too complex that requires detailed explanation here. You’ll need to read the entire article, of course, and there will be deep detailed analysis and a step-by-step plan.
The second part involves this expansion, a more detailed demonstration of the perspective, what solving this problem can lead to, why it’s important, how it affects my life and the reader’s life.
Essentially, this is a deeper dive after the introduction that provides detailed description. It’s typically the largest part of the article, taking up the most space along with the third part, which is gamification, or Gamify.
The second-to-last letter of this framework, G, is aimed at giving the reader a step-by-step plan. That is, what to do point by point, step by step, which is exactly what the reader came for. The reader wants to get an answer, they want to get some solution to their problem, and that’s what we provide in this step-by-step plan.
And the last section, which goes beyond the framework (which was originally APAG), but which I added, and in my opinion, articles look incomplete without it, is the conclusion and summarization, the last letter C – Conclusion.
Here we simply briefly revisit the problem itself, its solution, and at the end provide some guidance, a motivational message that helps energize the reader with a call to action.
This is, in principle, what we want to do. We want to help someone, we want to solve their problem, and to do this we nudge them a bit from behind and invite them to solve their problems with us. And we do this through text, accordingly.
For each section of the structure, the prompt already specifies the length, and the resulting article here is approximately 3,000 words. This is the approximate length of my articles that I write for my newsletter, for my blog. And this material is more than enough for me to later compile a huge number of posts, threads, even scripts for YouTube videos, and so on.
But here you can use your experience and experiment with this. Maybe you want to adjust the level of detail in the articles differently, maybe you’ll want to make it longer or, conversely, shorter. This depends on how you see it. That is, I’ve selected the length that works for me.
Plus, there’s a caveat in the prompt, if you read it, that if the reasoning in the original text doesn’t fit within the given article length, then this length should be exceeded. And, accordingly, the important thing here is not to miss a single thought, a single reasoning, a single example from the original text.
That is, the article can be larger; often mine end up much larger, around 4,000 or 5,000 words, because my source material, that is, my reasoning that I initially created, was very extensive, detailed, with many examples, and I generally know how I want to tell this to my reader.
This point is very important here, and it states that there’s a clear restriction in the prompt that no storytelling element from the original article should be omitted during compilation. That is, in the text where I write, there are certain thoughts that I put forward, there are certain chains of reasoning that I want to convey to the reader.
I’ve refined the prompt several times to preserve all these details in their original form. Of course, correcting any errors and making it more readable, but nonetheless all the key ideas that I write in the original text will be preserved here.
So, if you, for example, tell a story, you will definitely find it in the final article. If this is removed, then artificial intelligence has a tendency to paraphrase and invent its own stories. And, accordingly, you might find something in the article in this case that completely goes beyond your thought process, which can actually only harm.
Because of these additions from the AI itself, the article might be understood differently than you intended, for example, when setting up this material, and written in a completely different direction than was originally conceived. Therefore, for me, for instance, it’s important to preserve the source and those very thoughts that I express in the original material.
Another point is that this allows it to sound authentic, that is, there will be fewer AI inventions here; this prompt is aimed at compiling an article from your source text in accordance with the structure we define, in accordance with the framework that is better to follow if you want your texts to make an impression, to be engaging, to be interesting for your audience to read.
If you just present your reasoning in such a raw form, maybe they will be interesting to someone, but most people will be much more willing to read articles that are written according to a certain structure, so it’s important to adhere to this structure here.
But, again, this is the prompt that requires the most customization and refinement on your part, because it’s currently sharply tailored to my style of writing articles. If, for example, you want more inventions from artificial intelligence, you can remove these parts.
Or add something of your own. I, for example, have added conditions and requirements here that artificial intelligence should not invent non-existent examples and should not invent hypothetical situations, cases that happened to me, because it had such a tendency, especially when taking some statement, it would write phrases like “take Bob, for example.”
And it starts inventing some story about Bob that has absolutely no relation to reality, just a completely made-up story, looks sterile, formulaic, unattractive, and completely breaks the overall flow of the text.
Therefore, I strictly prohibited the use of such techniques and required the use of only real examples. Plus, I enriched the prompt for research in such a way that real-life examples must be present in it. And research is one of those documents that we input when writing an article.
Instructions for Using the Prompt
- So we open Claude,
- create a new chat,
- paste the prompt into the text field,
- attach four documents, which are:
• our source, that is, the story,
• our research,
• authorial style, and finally,
• the reader avatar. - After the prompt text, list the following by points: simply write the file name, better to copy it. Without any additions, just a list by points 1, 2, 3, 4:
• first the article draft,
• then the research,
• then the authorial style, and finally,
• the reader avatar.
This is sufficient. If you don’t write this, then in some cases it may request all these materials on the second step, which we don’t need, because each such step increases the number of tokens that were spent on working with this article, because, I remind you, Claude rereads everything from the very beginning each time, and that means it will scan these documents, prompts, and so on from the beginning.
Therefore, it’s better to give it as much information as possible at the input right away, so it starts working on the material.
After sending the prompt and materials by points in the chat, on the first step Claude won’t immediately write an article. First you need to agree on the article outline — a template by points, a template by which the article will be written, and possible headline options.
You need to approve the headlines. I recommend reading them carefully, because there might be something suitable that reflects your idea, or something completely different that’s related to your target audience’s problems as defined in the reader avatar. This is good and might guide you toward the right thought about positioning the article.
The text depends on the headline. The headline plays a crucial role, pay attention to how it’s formulated, because the article’s positioning is built on this.
I often notice that the headlines in the proposed form don’t suit me, but I combine several: I like the first part of one headline, and the second part of another. This creates a final version that suits me.
Look, choose. If you don’t like any headline, ask to generate another dozen. You can work with this, but remember to give all instructions to Claude in one request due to context window limitations and the cost of recalculating context with each new request.
Next comes the outline — the article framework, which can also be edited. If you don’t like some names, change the headings, add or remove items. Work with it, it’s living material. Don’t think that Claude will do exactly what you conceived and what you’ll like the first time.
Select appropriate options, edit. This gives you the opportunity to gain experience in writing articles, will help you become better at writing. Perhaps you’ll want to write the final article yourself, and you’ll have an excellent foundation for this.
After the outline is approved by you (just write “approved” or another phrase that clearly indicates that you’re satisfied with the material), Claude will create an artifact (a document in a separate widget) in which it will type the final article.
Editing the Article
After the article is written by artificial intelligence, we need to edit it and read it, of course, because this is artificial intelligence, no one guarantees here that there won’t be any hallucinations, that there won’t be things that it forgot, for example, or missed, even despite the fact that they were mentioned in the prompt.
Such things happen less and less frequently with model updates, but sometimes it occurs, so I recommend reading it anyway, really spend time on this, read every word and weigh the train of thought, because if, for example, they weren’t clearly formulated as a logical chain in the original article, then artificial intelligence might invent something here, and it’s not a fact that they align with your train of thought, with your chain of reasoning.
And in this case, you need to ask to correct this point, or rewrite it yourself, and yes, sometimes words might be used there that you would clearly not use, or, for example, it might invent some terms or concepts, for example, you describe some process, and artificial intelligence invents some name for it, writes it in quotation marks, yes, here especially carefully, because there might be all sorts of strange names that will clearly reveal, for example, your text as not written by a human.
And it’s important to read your article, because, first, you need to know your content, second, make it more human, if you see that something is clearly not written right, doesn’t reflect your voice, then all this can be corrected, remove some individual words or replace some phrases. This is essential, that is, such a final instance of editing should still be you, it’s your article, it’s your content, so don’t transfer all responsibility to artificial intelligence here.
