Train Your AI to Write Like a Human With Just 3 Files
A HubSpot marketing exec builds a three file Creative Context OS, creator.md, audience.md, taste.md, live on camera to stop his AI drafts from sounding like everyone else's.
Posted
yesterday
Duration
Format
Tutorial
educational
Views
3.3K
63 likes
57 · 43
Big Idea
The argument in one line.
AI models already hold more raw knowledge than any writer needs, so what makes AI content sound generic is a missing layer of the writer's own judgment, audience understanding, and taste, which can be captured in three text files and fed back into any AI assistant.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
A solo creator, marketer, or founder already publishing on LinkedIn or Substack who wants AI to draft in their actual voice instead of generic AI writing.
Someone building a Claude Project, CLAUDE.md, or custom GPT for content generation who wants a concrete file structure to start from.
A marketer trying to explain, in plain terms, what AI still cannot replace in a writing workflow.
SKIP IF…
You're hoping AI can generate your voice and taste from scratch with no existing writing or opinions to draw from.
You want a fully automated content pipeline with no editing. The video is explicit that AI drafts still need human rewriting.
TL;DR
The full version, fast.
AI writing sounds like everyone else's AI writing because the model has knowledge but no judgment, audience, or taste, the three things only the writer can supply. Kieran Flanagan's fix is a Creative Context OS: creator.md, built from a writer's actual posts (not adjectives) with one-off viral hits filtered out, audience.md, built from dream-reader profiles, filtered comment analysis, and outside research, and taste.md, built by having AI interview the writer until it produces a checklist and a single test: can a reader explain the idea in 10 minutes? None of the files write final content. They're for ideation and editing AI drafts, with the writer still doing the real work of thinking, drafting, and deciding what's worth saying.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
AI already knows enough. The missing piece is the writer's judgment, audience, and taste, and that gap is what makes most AI content interchangeable.
01:50 – 03:40
02 · If you can't write, AI won't save you
AI amplifies an existing writer. It can't manufacture perspective or point of view for someone who doesn't already have one.
03:40 – 04:40
03 · The three files: creator.md, audience.md, taste.md
The whole system is three short files, not a giant second brain: how you create, who you create for, and what you consider good.
04:40 – 05:50
04 · Taste is the only thing with a price
With production nearly free, the differentiator shifts from making things to choosing and judging what's worth making.
05:50 – 08:50
05 · Building creator.md from your real work, not buzzwords
Feed the AI your actual posts instead of adjectives, filter out one-off viral hits, and separate constants from platform adaptations.
08:50 – 10:40
06 · No audience yet? Use copywork
With no existing content, hand copying an admired writer's work, the old copywork method, builds a starting style.
10:40 – 12:20
07 · What creator.md actually contains
A finished creator.md covers how you build ideas, recurring creative modes, tone, and what makes your work distinct.
12:20 – 16:30
08 · Building audience.md: dream readers, comments, research
Audience.md is triangulated from five dream reader profiles, filtered comment analysis, and external research on the same audience.
16:30 – 20:00
09 · taste.md: you can't outsource taste to AI
Taste has to come from the writer. A curated list of admired writing, with reasons why it works, seeds the taste file.
20:00 – 22:10
10 · AI interviews you to find your quality bar
AI asks direct questions about what makes writing good, turning the answers into a concrete do/don't checklist and a publish test.
22:10 – 23:40
11 · How Kieran actually uses the system
The files power ideation and sparring, never a finished draft. Writing itself stays a human task.
23:40 – 26:00
12 · Live: the 'unbundling the marketer' draft, before and after taste.md
Running a live draft through taste.md sharpens the hook and focuses it on one idea instead of several.
26:00 – 27:50
13 · Live: fixing a LinkedIn post that flopped
Taste.md diagnoses a real underperforming post: a soft hook, an over-explained paragraph, and too many mixed ideas.
27:50 – 29:02
14 · That's the whole system
The close: start writing for real before leaning on the files, and keep the whole system simple enough to build in an afternoon.
Atomic Insights
Lines worth screenshotting.
AI models already hold enormous knowledge, so the thing missing from AI written content isn't information. It's the writer's own judgment, audience, and taste.
Content built the same way by everyone using the same prompts becomes commodity content: technically fine, but impossible to tell apart from anyone else's.
When production is free, judgment and taste become the only scarce, valuable things left in marketing.
Telling an AI 'my tone is professional and a little conversational' fails. Feeding it your actual published posts and having it extract the patterns works.
Before training an AI on your best posts, strip out the one-off viral hits, like role changes or single launches, since they're not repeatable patterns.
A writing style file should separate constants that never change from platform adaptations, like how the same voice compresses differently on LinkedIn versus Substack.
If you have no existing audience or body of work, the old school 'copywork' method, hand copying a writer you admire line for line, builds a starting style to iterate on with AI.
An audience profile built only from your own follower list misses people. Cross reference dream reader profiles, filtered comment analysis, and outside research on the same audience.
An estimated 60% of LinkedIn content is AI generated right now, so audience research built from comments has to filter those out or it just reflects other bots.
Taste can't be delegated to AI. The model can interview you about what you value in writing, but the judgments have to come from you or the output collapses into the same generic taste everyone else's AI has.
A usable taste file distills down to one test: could a reader explain the core idea to someone else in under 10 minutes?
Applying a taste file changes a draft's actual language, not just its polish, turning a soft claim like 'nearly all marketers' into a sharper, falsifiable one.
The most durable use of these three files isn't generating finished posts. It's using AI as an ideation and sparring partner while the writer still drafts and edits everything.
As AI collapses the cost of producing things, marketing work is splitting into three separate skills: judgment, systems, and cleanup, the maintenance work every AI system quietly creates.
Takeaway
Three files that stop AI writing from sounding generic
WHAT TO LEARN
AI already knows enough. What it's missing is your judgment, your audience, and your taste, and those can be written down and fed back to it.
01Why AI writes commodity content
AI models already contain enormous knowledge, so what makes their writing generic is a missing layer of judgment, audience, and taste that only the writer can supply.
Commodity content is writing that could have been said by anyone. It follows the same regurgitated patterns because it comes from the same models with no distinct input.
02If you can't write, AI won't save you
If you're not already a decent writer with real perspective, AI can't manufacture that for you. It can only amplify what you already have.
Watching how the taste and judgment layer works is itself a lesson in what to improve as a writer, even before building any files.
03The three files: creator.md, audience.md, taste.md
The system is only three files: creator.md for how you think and write, audience.md for who you create for, and taste.md for what you consider good.
None of the three files are meant to replace a big second brain or knowledge base. They're meant to be short enough to use as working context.
04Taste is the only thing with a price
For a century the hard part of marketing was creating things at all. Now that production is nearly free, taste and judgment are what's left to differentiate.
The real shift is from how do I make this to how do I choose what to make, and that choice is what a taste file is meant to encode.
05Building creator.md from your real work, not buzzwords
Skip vague self-descriptions like professional but conversational. Feed the AI your actual LinkedIn and Substack posts and have it extract the real patterns.
Filter out one-off viral posts, like a role change announcement or a single big launch, before training on your body of work. They're not repeatable style.
The file should separate creator constants, true everywhere, from platform adaptations, how the same voice compresses differently per platform.
06No audience yet? Use copywork
With no existing content, hand copying a writer you admire word for word, the old copywork method, builds a starting style you can then adapt with AI.
You still have to do the work of writing and getting better. AI cannot skip that step for you, even with a completed creator.md.
07What creator.md actually contains
A working creator.md includes how you build ideas, recurring creative modes, tone and sentence style, and what makes your work distinct.
The goal is for the AI to preserve how you see a problem, not average your voice out into something generic.
08Building audience.md: dream readers, comments, research
Start by naming five real dream readers, people whose profiles you'd hand to AI so it can extract concrete characteristics instead of guessing.
Analyzing who actually comments on and engages with your content adds a second data source, but filter out AI generated comments first.
External research on your audience's priorities and frustrations should enhance, not replace, the dream reader and engagement data.
09taste.md: you can't outsource taste to AI
Taste has to come from the writer. Asking AI to define your taste just returns the same generic taste it gives everyone else.
A strong starting point is a curated list of 12 to 15 pieces of writing you admire, with a clear explanation of why each one works for your audience.
The real test of taste is being able to explain why a piece of content is good, not just pointing at it as an example.
10AI interviews you to find your quality bar
Letting AI interview you with direct questions, like what makes contrarianism good, forces you to articulate judgment you already have but haven't written down.
The output is a concrete quality bar: what to always do, what to never do, and a single publish test, can a reader explain the idea in 10 minutes or less.
11How Kieran actually uses the system
The three files are used for ideation and as a sparring partner, never to generate a finished, publishable draft.
Writing is treated as how the creator thinks, not a task to hand off, so even a good AI draft still gets rewritten and finished by hand.
12Live: the 'unbundling the marketer' draft, before and after taste.md
Running the same draft through taste.md after a first pass sharpens the hook and forces a single focused idea instead of several mixed together.
As AI collapses the cost of producing things, marketing work is splitting into three skills: judgment, systems, and cleanup, the maintenance every AI workflow creates over time.
13Live: fixing a LinkedIn post that flopped
Running an underperforming post through a taste file catches specific fixable problems: a soft hook, an over-explained paragraph, and too many ideas crammed into one post.
A redrafted hook that names an exact behavior, like marketing teams aren't just getting smaller, they're changing shape, beats a vague claim like nearly all marketers.
14That's the whole system
If you're just starting out with no audience, do the actual work of creating content first. The three files only get powerful once you have real data to feed them.
The entire system is meant to be simple enough to build in an afternoon, not another complex infrastructure project.
Glossary
Terms worth knowing.
Creative Context OS
A three file system, creator.md, audience.md, and taste.md, that teaches an AI assistant a writer's voice, audience, and quality bar so it drafts and edits more like that specific person.
Commodity content
Writing so generic in structure and voice that it could have been produced by anyone, the default result of most AI generated content with no added context.
Copywork
An old, pre-AI training method of hand copying an admired writer's work word for word to internalize their sentence structure and style.
“When the cost of production is free, judgment and taste is the only thing with a price.”
Sharp, standalone thesis statement with a clean contrast.→ TikTok hook↗ Tweet quote
01:35
“Commodity content means it could have been said by anyone.”
Tight definition that reframes the whole AI content problem in one line.→ IG reel cold open↗ Tweet quote
16:50
“You never outsource your taste to AI. You instill AI with it.”
Punchy rule that doubles as the video's core instruction.→ newsletter pull-quote↗ Tweet quote
19:40
“Could the reader explain the core idea to someone in 10 minutes or less?”
A concrete, testable rule readers can apply immediately.→ newsletter pull-quote↗ Tweet quote
24:10
“AI isn't just changing marketing jobs. It's unbundling the marketer.”
Reframes a live edited draft into a quotable thesis.→ TikTok hook↗ Tweet quote
24:25
“AI makes producing things cheap. It doesn't make deciding what's worth producing cheap.”
Clean parallel structure, easy to remember and repeat.→ IG reel cold open↗ Tweet quote
The Script
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
metaphorstory
We have had an avalanche of new models from Fable 5 .1 to GPT Astra to Opus 5 .5 and that is becoming a problem for marketers because these models are actually pretty good at creating content. Now you have hundreds of thousands of people all creating content with the very same models. In this video, I am going to show you how you can build a creative context OS.
It's three tiny files and it's going to teach the AI, how you create content, who you create for, and what you think is actually good. In this 20 minutes, you're going to be able to build the very first version for yourself.
And at the end of the video, we'll give you a singular graphic that teaches AI how to do this for you. All of that and more on this episode of Marketing Against the Grain.
The problem with creating a lot of content with AI is actually not that AI is not a good content creator. It just creates commodity content. And that is a great line from a writer called Nicholas Cole.
Commodity content means it could have been said. by anyone. It's not distinctly used.
So it's using the same regurgitated writing patterns and producing the same commodity content, content that could be said by anyone, lacks perspective, lacks point of view, doesn't say anything about how I see the world. The best content is really created by that person from their perspective, right? You don't really want AI creating content for you.
You want AI to be able to repeat what you say in a multitude of different ways. And that is so important. And I want to be really transparent and honest about this.
If you are not a good content creator, then you probably are going to struggle to use AI. AI is really great if you already have perspective, points of view, things to say, ideas to bring to life, and you already can craft content.
Now, I will say, if you watch this entire video and you want to improve your writing, it will teach you what you need to get better at to make AI a great content assistant for you. On this... What I want to show you is how you as a writer can actually create a creative content OS and actually enrich and instill the things that make you distinctively you into AI.
And that's super important. But as I said, if you are not a great content creator or you want to become better, even just watching this video will show you the areas that you need to improve on. The thing to keep in mind is today's LLMs already have an extraordinary amount of content.
And you don't need to build another big giant knowledge base or you don't need to build a really complex second brain. You just need some simple context files that we're going to give you in this video that instill. your judgment, your experiences, your beliefs, your audience, your quality bar into AI.
So it creates those things for you. So really when people are using AI to create content, it's just missing these kind of context files that will actually bring your content to life and make it distinctly you. The models already have an extraordinary amount of knowledge.
The model is missing the stuff that makes you, you. It's missing that context. And that is what we're going to build.
We're going to build three files that can help you elevate your content to really stand out from all of the other things being created, but also feel like it's really authentically coming from you. So this is the finished creative context OS. It has a folder.
called Creative Context. And inside there, there's three files. There is a creator .md file.
And that really explains how you think and how you create the way you do. It has a audience .md file that explains exactly who I create for and what they care about. And then really importantly, it has a taste .md file.
And that teaches AI. what I think is actually good. And that is so important.
And taste is actually going to come from you. It's not going to come from AI. People don't want to do any work anymore.
They just want AI to tell them, well, what is my taste? Who is my audience? You need to actually do some work to make these things really good.
But the great thing about this is I've made it so simplistic that the AI can guide you along the way. For a hundred years, the hardest part of marketing was creating things. That was the hard part.
How do I create? this blog post, this social post, this landing page, all of these things I would have to create to attract an audience and then turn that audience into customers. Now it's so much easier to create things.
I often say when the cost of production is free, when it's so easy to make things with AI, judgment and taste is the only thing with a price. It's the only way you're going to stand out. And so now the problem has shifted from how do I make things?
to how do I choose what to make and how do I have taste to make the things that I make distinctly me and different from everything else. And this system helps you do that. And it actually helps you to instill these into your AI assistant so it creates things with these in mind.
You can put them into Codex, you can put them into Cloud, ChatGPT, any AI environment that works for you. Okay, let's start. And we're going to get in and do some live work.
I'm using Codex and ChatGPT to do this right now. But first, we're going to teach AI how we create. And there is a bad way to do this.
You could sit down and you could write, my tone is professional. I'm a little conversational. I like short sentences.
I'm insightful. I'm sometimes humorous. Like basically just telling AI these kind of buzzwords.
the best thing that you can do is you can give it your actual work. So the example here is I give AI my actual work. So I exported my LinkedIn post.
I exported my Substack archive. I'm going to do YouTube. I didn't do it for this example, but I also have a lot of content on YouTube and other places.
And then what AI will do is it will actually extract all of the best posts. best themes. And what I'm going to give you here as part of the file at the end is the exact prompt that I use, because the important part is that it doesn't just verbatim choose the highest engaged post.
We will give you an image and that image will be able to give you all of the things I went through. You can get it right here in the comments. You can click the link in the description or scan the QR code to get them all delivered to your inbox.
And that infographic, you can just give it to your AI assistant and it can start to build that you're a creative context OS. It actually works with you to say, is this the kind of posts that you believe your audience will really resonate with them? And there are some reasons to do that.
So some of my top posts are really once -off things when I've changed roles or there's been a specific launch or I've done a specific post with a bunch of satire in it. And they're just not repeatable patterns. They're really more once -offs.
And so you want to extrapolate all those out of the things that you're going to extract patterns from. So here again, your performance is really finding the best candidates. Like these are the kind of posts that are performing really well, but your taste is actually decided on the training set.
And that's so important because you, again, are doing some of the work. So when I went through it, I selected a small number of things from my LinkedIn. An example, we had a...
post about the decline of the CMO role that did really well. I did a post about the modern B2B marketing playbook that was done through a bunch of conversations I had with leaders. I did some posts about how you reframe the idea of AI slop.
And then I did this post that was kind of humorous about how product marketing might work at the companies like Antropic. I kind of went through it and decided that these were distinctly me. They worked really well for my audience.
I did the same thing for Substack. And that's kind of where I went to for the creator MD file. And let me show you how this worked in practice by sharing my actual screen.
Okay. So you can see here, basically you can play around with this and we'll include this in the tutorial that we give away. We'll give away just an image that contains the full tutorial.
And then you can just upload that image to your AI assistant and kind of replicate this for you. So you basically say, I'm going to give you five to 10 pieces of the best content. You can use any platforms.
Now, one of the pushbacks will be, well, I don't really have an audience. I haven't really started. And I'm trying to figure out how to use AI to do this.
As I said, much easier if you just start to write. If you start to write, you get better at writing. You have some success, have some data to actually start to use with AI.
But if you really want to get started and you just want to start to use AI straight away, you could for 100 % just swap in posts. that you believe would represent the style you want to use. And so there's a old methodology.
of how people got better at copywriting called copywork. And copywork was basically taking a copywriter or a writer that you really loved and you would like to be able to mimic some of their style. And the style is really structure.
It's how they structure the way they write. And everyone has learned from other peoples and how they do that. Very few people are very unique.
They've all kind of inherited their writing style from someone else. And so there's nothing wrong with doing that. And the way copywork would work is you would literally just write out there.
words. You would take something they wrote and you would write it word for word, line for line. I did that.
That's one of the ways that I got better at copy is I just picked a bunch of writers and I would just sit there with a pen and I would write out their words. And this is pretty similar. If you don't have any content and you're really at the start of doing this for you is you could take someone's content and you could basically use that.
to start to figure out what your style would be and then work with AI to like edit it and craft it for your unique taste. Now you still have to get good at creating content. You cannot skip that part.
So even if you have creator .md created, you're still going to get back something that you yourself are going to have to edit. You yourself are going to have to elevate and make better. So it will focus on how I think, how I build arguments, recurrent ways I explain ideas, tone and sentence style, creative models, and what makes my work distinct.
Now, these things here, focus on should be unique to what you want to focus on. This is what I want to focus on. So again, the worst version of using AI is going to these YouTube videos and taking exactly what they do and just regurgitating it.
You actually want to customize it and make it distinctly you and map to what you want to do. So then it looks for things that stay true everywhere. And then if I'm doing stuff for LinkedIn or I'm having doing stuff from Substack and then I'll add in YouTube, it has a little line to tell me how I would amend or adapt the writing style for those platforms.
You can just have one creator .md file and it's applicable for all platforms. And then this is kind of what we'll do, right? It will have the core DNA, like your writing DNA.
It will have how I build ideas. It will have my recurrent creative modes. It will have my voice.
This is an abbreviated version, which is what I asked it to do. And then it will actually have, you know, how these work in LinkedIn or how they work in Substack. And Northstar AI should preserve the way I see the problem, not average me out, right?
It should not make me into AI. Now, again, I'll go through at the end how I use this system. It is not to create.
final versions of content. And you can use it for however you want to use it, but I don't use it for final versions of content. So that is the creator .md file.
So next is your audience .md file, right? So you want some real context about who your audience is. And there's three core ways you could do that.
One of it is pretty interesting. You can actually just start with five dream readers. You can give AI their LinkedIn profile or their X profile, actually put a real face to who you want to read your content.
If you wanted me to read your content, you could give AI my LinkedIn profile, and then it can start to turn those profiles into some characteristics. of a real audience. Now there's some other cool thing that I was able to do.
So this one in the middle is basically how you can analyze comments. And I want to show you a quick way that I was able to do this and shout out to Dharmesh who's been on the show. So the cool thing about uSpot is uSpot has access to all of my data feeds.
And so anytime I'm doing things, it can actually start to update its information about me. And one of the things I've been doing is enriching my audience .md file. with the comments from LinkedIn because it has access to all of my content and everyone who's liking my posts and everyone who's commenting on my posts.
And what I ask it to do is build a profile from the people who are engaging with my content. Now, I say that's like one input. It should not override the other two things that I'm going to show you.
The first one was giving it real profiles so it could get characteristics of those folks. The second is... If you have content, being able to get the people who are engaging with your content and have an AI extract characteristics from them because they're already the people who are liking your content, right?
That's a pretty great thing to do. And you can see YouSpot is able to give me some pretty cool information on the people who are actually... engaging with my content and provide a lot of characteristics.
The one thing I did here in the prompt that I give you spot was I made sure that it would actually ignore any kind of comments that looked like AI 60%, I think of all content on LinkedIn right now is created by AI. So you have to be pretty. careful about that.
So that's the little bonus. And the second one is external research. And again, I will show you quickly how that looked on my screen.
So this is the prompt for external research. Obviously for the first one, if you're asking for it to look at existing profiles, you can just use the same prompt, but ask it to actually use existing profiles versus going and looking for external research. And then for the second one is pretty unique.
It's either that you can get the data. of people consuming your content, engaging your content some other way, or you can do it through use spots. So I can't really give you a cookie cutter way to do that.
So here, research people who match my target audience and help me understand what they currently care about. I already had a audience .md file that I give it to enhance with external research. If you're starting from scratch, you want to give it some paragraph or two of who your ideal audience is.
You want to write it out if you don't currently have it written out anywhere. And so it looks at their current biggest priorities, recurrent frustrations and problems, all of this information. And then it goes to public sources, practitioner discussions, LinkedIn posts, Reddit, all of these different things.
Now, this is the most simplistic version of this. If I did this in Codex or Cloud Code, it would have access to APIs. And so you'd give it a bunch of APIs and we have to pull this information pretty readily, but I'm keeping this super simple.
And then it's an abbreviated version. So it will show you who you're create for and what they care about. Again, if you give it better access to APIs, you will find it gives you incredible information.
What they already know, what they are tired of, what earns their trust, what feels novel to them. And then the North Star, do not explain why AI matters. Explain what excellent marketing looks like because AI exists.
I think that's a pretty good one. So that's it. Three steps.
Really simple. You're dream readers. You can grab their profiles.
A little bonus step if you're able to get the comments or DMs. And then that external research. And that goes into this audience .md file.
So now you have your creator .md file. And now you have your audience .md file. And let's go into the last one that we're going to create.
Okay, let's do the final version. And that's the taste .md file. And this is the most important part because you have to have taste and you have to figure out what your taste is.
And how do you teach AI taste if you have none? AI is not going to give you some. You never outsource your taste to AI.
You instill AI with it. You cannot outsource it because AI will give you the same as hundreds of thousands of other people who are using AI to create similar things. And this system can really help you understand what your taste is.
So you have to do the work, but AI can walk you through it. And let me show you what that means. Okay, so the first thing is...
If you have a list of great writing, that's even better. So if you already know what are great writing examples or great content examples that really have resonated with your audience and work for your audience, you can just use those and give AI a clear articulation of why they really matter, why they work for your audience, why they would resonate with the people that you're creating content for.
There's a great example. There's a website that has famous business memos. And there's another website that does famous positioning statements.
And there's a website that does the best ad copy that's ever been created over time. And I used to always read through those. And so you would bookmark those.
And you would have clear examples of what taste meant for you, why these resonated with you, and why you were able to explain them being so good. Obviously, you can use some of your own content as well. If you don't really have that list, this can make it really easy for you.
So you already have a audience .md file. And so what AI can do, and this again is tailored for me. So make sure when we give you the file at the end, you want to adapt your prompts for the writer that you want to be the content creator that you want to be.
And it will basically create a list. of potential writing that appeals to your audience. Now, again, this is abbreviated output.
It can create as many of these as you want. I chose 12 to 15 in the giveaway that we're going to give to you at the end, but you can increase that number to 20, 25. So for example, it would come back and it would say positioning.
This is one of the best positioning series of all time and give you a little bit of... Reason why your audience would love this. Kevin Kelly, 1 ,000 true fans.
Why would my audience at MD5 love this? Because it has a simplistic framework that is really easy to understand and is repeatable. And that's something I love.
Very similar to Paul Graham, like sharp operator, real reasoning, an idea that sticks. Ben Thompson, he explains a phenomenon that actually was happening, but no one had put a name to it. And then Claude Hopkins is just one of the most analytical.
data back old school copywriters comes from a scientific background. And so these were the ones I actually picked myself. And so the abbreviated version is just kind of showing you.
The ones I had already picked. And then again, this isn't what I wrote. I wrote a bunch more, but you want to say why you love these pieces of content, right?
A thousand true fans. One simple idea would hide your ability. Curation and algorithms by Ben Thompson explains something happening and breaks it down and gives it a name and makes it easy to digest.
This is what people miss. You yourself have to be able to explain this to AI. You can't just say, please, you do it for me because you're not going to get a great outcome.
then the AI can actually interview you. And so it will actually start to like ask you a bunch of questions and you can tailor this to whatever you're trying to do. And better yet, you can tell AI what your mission goals are and it can tailor the output we're going to give you and the tutorial we're going to give you to your needs.
And so it will come down and we'll ask, again, this is abbreviated, but like, are you really focused on an original idea or like really great writing? And I'm like, no, the idea matters way more than the writing. What makes contrarianism good?
A strong argument or the mechanism underneath it, even if it's not fully proven. And the answer I gave was really the blend of these together, right? A strong argument with some validation, but sometimes you can't validate everything.
Sometimes some things are just an idea. When is a framework useful? When it compresses a pattern and makes it easier to remember.
So it actually interviews you to start to understand the writer you want to be and the kind of taste and judgment that you have. And then it will produce your... taste .md file, and it has some of these things.
What is your core taste? One sharp mental model beats a broad summary. Simplicity really matters.
Always make sure you teach one useful idea. You explain the mechanism. You compress complexity.
It has all these things. Never do these. Polish write -in with no original idea.
Fake contrarism. AI smooth write -in that loses personality. You have to have personality.
You have to have a quality bar that it runs through. And then what's the published test? Could the reader explain the core idea to someone in 10 minutes or less?
Because that makes it shareable. If not, simplify again. And so that is the taste .md file.
You basically create a list. Either you have that or you have AI created from your audience .md file. You tell AI what are the kind of pieces of content that really resonate with you, your audience, and why.
And that why matters. That why is actually, I understand why this piece of content was so great. I can actually explain it.
You know, make this really easy. Turn on WhisperFlow and just talk to the AI or turn on voice mode and talk to the AI. And then you basically have AI create that taste .md5 for you.
One of the best things that this is good for, and I think this is the way most people who are using AI for writing find value, inclusive of me, is basically as a quick ideation machine. And then as a quick kind of sparring partner with your content, never to create the content. I just can't have AI create content for me.
It's just an alien concept for me. I've never been able to have anyone create content for me. I could never have someone write something on my behalf.
And so that could just be uniquely me. But I just don't understand why I would want to do that because I love writing. Writing is how I get thoughts out into the world and myself, how I internalize thoughts.
And so I just would never outsource my taste to AI and I would never outsource my thinking to AI. I need ideas. And sometimes I just write something that's in my head and I just write it out.
And sometimes I'm just like, well, what would be like, what's interesting? You know, I need to like have some ideation, have someone to go back and forth with. So here it will read your audience .md file.
And again, we can give you these prompts in the creator .md file. And then it will start to research what's happening across a couple of subjects. Again, this is a pretty simplistic way of doing this.
In my actual version, I have a bunch of API set up, but I didn't want to make this too complex. And I just want to show you some of the ideas it would bring back. So again, these are all pretty mapped to my audience.
And so it brings back an idea. It says, what is the signal? Why my audience cares about it?
What's my angle? It reads my creator .md file. And then basically says, what's the thesis?
So I took this one. The new marketing jobs are really ownership jobs because we're talking a lot right now about how marketing teams will change. And I won't get into that.
We're going to do a whole episode on that, but I think that's a pretty interesting one. And so I asked it to create a first draft. Hey, let's just create a first draft and it's fine.
It's looking at the different three files, but really it hasn't really adhered to taste .md because it's loading all at once. And so it's better when you actually ask it to like read one file and then really crystallize it. And so it has like a good skeleton of a post because I do like things that are.
broken out and really easy to understand concepts. And so this breaks marketing into judgment systems and cleanup and has a pretty good frame on why marketing would look way. Now I wanted to run it through taste.
So run it through the taste of MD file and show me the revised draft. And it does so much better, right? AI isn't just changing marketing jobs.
It's unbundling the marketer. Pretty good, right? Unbundling the marketer.
That's a great hook right at the start. And then a lot of that work that used to sit inside one role is being split into three different roles. Judgment, what should we do?
What is actually worth automating? Is this actually any good? AI makes producing things cheap.
It doesn't make deciding what's worth producing cheap. That got that from me. That's a line that I've used.
If you go all the way back to what I said, Nicholas Cole said is, AI should repeat the things you say. The best content creators have a couple of bits, and they do those bits for a couple of months. And they do them in a multitude of different ways.
You don't want to have AI write content. You want to have it repeat the things you already believe and say in different ways. And this is a great example of that through the taste .md file system.
Someone has to turn that judgment into something repeatable. You give the right context. You connect the right tools.
You build the right workplace. You put quality gates in place. I'm pretty sure this research might come from a post that I wrote already.
And then cleanup, this is the part nobody puts in the AI transformation deck, prompts drift, data gets stale, integration spray, quality slowly degrades, all true. And that gives you what changed. I think that's a really, really, really, really, really good first draft.
Again, I would be able to improve it. I would add some stories. I would actually finesse it a little bit.
I would see if I actually agreed with it. But as a person who can go back and forth with it and give me a great idea that fits the way I think about writing, that's pretty great, right? I want to show you a different way to use this.
So I had a LinkedIn post that didn't perform very well. And I'm doing this live. And I would love to see how it would make that better.
Can you run through my taste .md file and tell me what edits you'd? make so i think this is actually a good post it just didn't actually hit the way i thought it would hit now part of it is because linkedin has changed her algorithm again but i'm actually curious about what my taste .md file would have said because this is one i just read out and sometimes i do that i just like i have a thought i just write it all and then put it out straight away yeah so thinking is the core idea is strong The strongest parts, what I changed, make the hook more decisive.
I think that's true. Nearly all marketers, you should never say nearly. Marketing teams aren't just getting smaller.
They're changing shape. It's way better. Compress the Gartner paragraph.
I totally agree with that. There's just too much here. It's like word salad.
And so Gartner says, yeah, it compresses that. It doesn't look like something done. It looks like redesign.
That's better. Be careful with AI removes that limit. Okay.
Yeah. So let me ask it to redraft. Oh, he redrafted already.
Marketing teams aren't just getting smaller. They're changing shape. Nearly every marketing leader I talked to has been asked to do more or less, but the data suggests something more interesting than simply downsizing.
It's cut this time for me. That doesn't look like shrink period. It looks like resign.
I know that people are going to say, oh, it's done the, it's not this, it's that. You don't copy and paste this. It's just for ideas.
Three changes are happening. The zero to one marketer. One person can now take an idea from a blank page to a ship V1 research, blah, blah, blah.
I mixed and matched a couple of ideas here. And I think that was the biggest problem with it. I introduced too many ideas and it's actually trying to focus me on one because if you remember my taste .md file, that was a big part of it.
So that's the system. I think this is like all you need. If you can just do this, you don't need to overcomplicate it.
You don't need to do all these other complex systems. You can basically just build this creative context layer and then actually use it for. your writing.
But again, you have to do some of the work. And if you are just at the beginning of content creation, you should just do a bunch of content creation to find out what works for your audience. And maybe AI can help you a little bit, but I don't think it gets really good until you've figured out some stuff and you have some data to give it.
And you saw the day that I could give it. Now, again, if you want to start from scratch and you don't have an audience, do the copy work, use someone else's content that you're inspired by and use that to follow along this tutorial. You too can elevate your writing not get lost in the CS sameness and not be referred to as AI slop.
I hope this was a useful episode. I will see you all on the next episode of Marketing Against the Grain. This data is wrong every freaking time.
Have you heard of HubSpot? HubSpot is a CRM platform where everything is fully integrated. Whoa, I can see the client's whole history.
Calls, support tickets, emails. And here's a test from three days ago I totally missed. HubSpot.
Grow better.
The Hook
The bait, then the rug-pull.
Every new model release makes AI content easier to produce and easier to spot. Kieran Flanagan's answer is not a better prompt. It's three plain text files that teach an AI assistant how he thinks, who he writes for, and what he considers good, built live in this episode.
Frameworks
Named ideas worth stealing.
03:40model
The Creative Context OS
creator.md
audience.md
taste.md
Three markdown files that instill a writer's thinking patterns, audience, and quality bar into an AI assistant so its output reads as authentically theirs instead of generic AI content.
Steal forAny CLAUDE.md, Claude Project, or custom GPT built for content generation.
25:00list
Judgment, Systems, Cleanup
Judgment: what's worth automating and what's actually good
Systems: the right context, tools, workflows, and quality gates
Cleanup: catching prompt drift, stale data, and quality decay
As AI collapses the cost of producing content, marketing roles split into three distinct skill sets instead of one generalist marketer job.
Steal forFraming how a team or agency should restructure roles around an AI content workflow.
CTA Breakdown
How they asked for the click.
VERBAL ASK
06:40link
“click the link in the description or scan the QR code to get them all delivered to your inbox”
The lead magnet is the tutorial itself, a free downloadable graphic of the exact prompts, pitched mid-video rather than as a separate ad read. Fits naturally since the show is HubSpot's own podcast.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
A HubSpot podcast interview that turns into a live Claude Code build session, wiring a Facebook ads agent, a Google Ads agent, an SEO agent, and a cold-email workflow into one voice-driven marketing stack.
A HubSpot marketing lead runs a free Claude Code "marketing team" against a real company's live site and grades every output: a generic headline, a genuinely great one, and the AI audit that faked its own homework.
On The Millionaire Real Estate Agent Podcast, Gary Vaynerchuk lays out why five posts a week is barely a warm-up, why one post can outrun 30 years of cold calling, and why AI search just made content non-optional.
Neel Dhingra sits down with Think Media CEO Sean Cannell to break down the one long-form video a business owner can make once a week that teaches, sells, and closes clients for you.
00:27
00:49
01:17
02:01
02:23
02:50
03:07
03:50
04:10
04:51
05:13
06:01
06:23
06:52
07:05
07:42
07:53
08:16
09:01
09:22
09:45
10:07
10:50
11:02
11:30
11:50
12:08
12:51
13:12
13:35
13:53
14:20
14:35
14:56
15:17
15:32
16:19
16:40
17:06
17:22
17:43
18:02
18:25
18:46
19:07
19:28
19:54
20:32
20:48
21:15
21:36
21:58
22:43
23:06
23:51
24:15
24:38
24:53
25:25
25:42
26:33
27:04
27:17
27:45
28:02
29:01
28:14
21:46
26:03
05:02
25:55
43:03