The Exact Claude Setup That Writes My Documents Automatically

I haven't written a proposal from scratch in probably five months.
Now, I just drop my call transcripts into Claude.
It then gives me back a polished proposal in about 30 seconds.
And it matches my exact style every single time.
So in this video, I'm going to walk you through that three-step system that I built
that makes this happen.
We'll use proposals as the example, but it's important to note that this works for any
document you're creating from conversations, such as contracts, SOPs, follow-up emails,
really any type of document.
So let me show you how I built it.
This use case, being able to take a conversation and turn it into a document or an action is
probably one of the lowest hanging fruits that I've seen with any of my clients.
And it's the beginning of where real businesses start to see leverage through AI automation.
As I mentioned previously, this is not just for proposals.
You can do this process for all types of document creation.
So in this example, I'm going to walk you through proposals, but you could do this for
contracts, SOPs, follow-up email campaigns, any type of reporting.
And the process is simple.
There's three primary steps.
The first step is figuring out what good looks like and then reverse engineering that.
So we already have documents that we use with our clients, our prospects, and our employees.
In this step, the goal is to reverse engineer those documents that are already good, setting
it as a benchmark.
So in the future, we can embed that into an AI.
After we've extracted out what the good looks like, we're going to then embed this into a skill.
So the aesthetics of our reports look perfect.
And then after this, we're going to embed that into a project so we can use this on a
recurring basis.
And all we need to give this project is a transcript.
And that's the simple flow.
The first step is starting with what looks good.
So this is going to be the document they're already using.
So you already have contracts you send to clients.
You already have SOPs you give to employees.
You already have proposals you send up to the prospects.
All of these define what good looks like.
Now, what we need to do is we need to extract out of that the structure, the styling, the
fonts, the spacing, all the things that make the aesthetics, the aesthetics for our documents.
So we can ensure that when an AI writes it on our behalf, they're not just writing it in
a good way, but they're making it look like it's on brand.
And this is where skills come into play.
It's important to note that we're going to use Claude in this case, because skills are still
only native to Claude.
But like I said in previous videos, I can almost guarantee in the next couple of months,
ChatGPT and Gemini will both adopt this specific feature.
And they'll have it there as well.
All right.
So if you're enjoying this, you're probably going to enjoy two other things.
First off, Blow is a 30-day AI insight series.
Completely free.
You'll get 30 insights in your inbox if I can apply AI to your business and your work.
The second thing is if you'd like to work with me, Blow are a series of offerings to
see if there's a good fit between the two of us, such as a private AI community for business
owners and leaders or one-on-one AI coaching.
But that being said, let's get back into the video.
Let's look at the prompt of how we can extract out of the document what good looks like.
And actually, as you can see, this is a very basic prompt.
It's not rocket science, just a few lines.
So at the very beginning, we're asking the AI to methodically reverse engineer the document
we've given it to ensure that we can recreate that document going forward with an AI.
Now, two things I'll call it here.
I've mentioned methodically and I mentioned reverse engineer.
These are two terms that I've seen that works more effectively with AI when doing the specific
process.
So I recommend using those terms in your prompt.
After that, we're explicit about what we wanted to extract.
But we're also giving it some wiggle room because we're saying extract everything.
So if there's something that I've not mentioned below, the AI is going to take the power to
do what it needs to do to pull out whatever else I couldn't think of.
And that at the very end, we're being specific about the fact that this needs to be pixel
perfect.
Again, this is another term that I found that works well with AI in this specific process,
a reverse engineering, a document to get out something that we can recreate on a consistent
basis.
So this is our basic prompt.
Let me actually show you what this looks like inside of Claude.
Now we're inside of Claude.
And this is the exact conversation I just had with an AI.
So what I did is I dropped in an example.
So this is a template of a specific type of document that I wanted to mimic.
And in this case, it's a synthetic or fictional proposal that I would send to a client.
So you can see it has a very specific style to it.
Certain colors, certain spacings, a little symbol here in the middle.
And then also when you scroll through here, you'll see there's a very specific aesthetic
to this proposal.
Now what we want to do is we want to reverse engineer that.
So I gave it the aesthetic both in a PDF format and a doc X format.
That's important.
I want to give the AI as many file formats as possible for it to be successful in reverse
engineering the aesthetics of the file.
So that's why I've given it those two.
I then pasted in the prompt that I just showed you.
And then it went off and worked for about, I think 10 minutes.
Yeah, 10 or 15 minutes.
And at the very end, it has its breakdown of the spec.
So it broke down everything that's included inside of that document.
So that's the first step.
We're reverse engineering what good looks like.
After we reverse engineered the document and we have the aesthetics, we want to embed that
into a skill.
And that's what we're going to lock this in permanently for a skill that we can call on
a recurring basis.
Now a cloud skill, all that is, is a bunch of prompts inside of a folder that the AI can
call at any given time.
And the way that we're going to create a skill is very simple.
In that same thread of that conversation, we're going to put in a prompt like this.
Again, not rocket science, just a few lines.
So here we're saying now, after everything that you've learned in that same thread, I want you
to use the skills creator to create a skill.
And then I'm stating when the skill is called, it's important that the output should match
exactly a pixel for pixel perfect match of the original document that we've based this
learning off of.
Now, before you run this prompt, we need to ensure that skills are turned on inside of
cloud.
And the way that we can do this is by going to our settings.
So if you click your little profile here, you then go to settings, you'll land on a page
kind of like this in general.
You then want to go to capabilities and our capabilities, you want to then scroll down to the point where
you get the skills.
And here at skills, you want to make sure that this isn't turned on.
And also if you go to example skills and this toggle here, if I scroll down, you'll see
there's a skill here called a skill creator.
So this is a skill that creates skills.
I know very meta.
We want to make sure that this is switched on so we can use it in the prompt that I just
showed you.
So assuming you have this turned on, let me go back to the chat so I can show you what
it looks like.
And then right here is the prompt that I just showed you.
So I copy and pasted it into that same thread.
So this is the extraction where I extracted everything methodically.
And after it extracted it, I pasted it in that specific prompt and it's going to then
use the skills creator to create a dedicated skill.
So anytime I call that skill and it creates a proposal for me, it's going to be a pixel
perfect match to what my original document was.
And if I scroll through here, this took probably five to 10 minutes again.
And at the very end, it has a skill here.
So it's important to know that you can select this button here and it says copy to your skills.
When you copy to your skills, it's automatically going to save that to my skill section
so I can use it anytime I want.
And if you want to share this with other people, all you have to do is download this, send it
to them and they can just upload it into their skills platform as well.
And if I quickly go back to my capabilities, you can see that I've already selected copy
to my skills because right here I have the proposal creator for gradient labs, my company.
And that's the second step, which is embedding that learning into a skill so we can use it
on a recurring basis.
The third and final step is putting this all into a cloud project.
So anytime that we come to it, all we have to do is drop in a transcript and everything
is done for us.
And to do that, there's a few different components we want to combine.
So first is we have reference files.
So these are going to be files that the AI can reference if we need them.
This is optional.
But in this case, I'd recommend if you're having a draft, a proposal, report, a contract,
whatever else, it should have inspirational files to reference.
So either previous proposals you drafted that have the same exact structure and tone that
you expect to have the AI draft, or it could have maybe some of your core offerings that
you have.
So when it drafts a proposal for you, it's pulling from the specific offerings that you already provide
your clients or a few other documents that it could reference for inspiration.
Once we've added those, we then want to draft our system prompts.
So we're going to spend most of our time here to ensure that we're telling the AI exactly
what to do and how to do it.
And then finally, inside of that system prompt, we're also going to call the skill so we ensure
that the aesthetics of the report that we give back is perfectly like we would like it
to be.
So those are the three pieces that we really need there.
Before you draft your prompt, I always recommend writing a base prompt.
So the base prompt is something very basic.
And after that, you can then extend it and really add superpowers to it by using a prompt
improver.
So both Anthropic and OpenAI both offer prompt improvers, where once you provided a base prompt,
they're automatically going to inject all the best practices for prompt engineering their
models and give you back a much better prompt that you can get higher quality responses from
the AI in.
But before we get to that, we have to use the WWH framework.
And that's how you write your base prompt.
You want to answer three questions.
What, why, and how?
First, we have the what, which is simply explaining to the AI what we want it to do.
Second, we have the why, which is telling the AI why we're doing this in the first place.
And the reason it's important is that oftentimes there's a lot of stuff in our head that we can't
communicate effectively.
But if we tell the AI why we're doing this in the first place, it can start to infer from
that other things that we may have forgotten to say that could give us a higher quality response
from the AI.
And then finally, we have the how, which is all around constraints.
So if there's very specific things that matter to you when this report's drafted, you can add that
here, such as maybe the output format, the structure, the file type, et cetera.
And those are the three questions we want to answer.
And what I'll show you is a base prompt that I drafted for this before I improved it with
the anthropic prompt improver.
I won't run through everything, but I'll walk you through kind of the sections and why I added what.
So at the very top, we're answering that first W.
So this is going to be the what.
We're first giving it a persona saying you're an expert proposal writer.
Then we're giving it a task saying that you're going to receive a meeting transcript.
And then it's your job to then draft a proposal based off of that.
I then state that there, it's going to have some examples that I can reference inside of
its knowledge base.
And then most importantly, I'm saying, then I need you to use this very specific Claude
skill to ensure that the report that you provide back to me is aesthetically pixel perfect.
Now here I've said X within parentheses.
This is actually for me going to be the name of the skill and the same thing should be for
you.
So if you copy and paste this prompt and use it for your own project, which you can, because
I'll share the presentation.
So you can copy and paste the prompt, you want to fill that in with the skill name that
you have.
So for me, it's gradient dash proposal for you, it might be something else.
So make sure you're naming the skill effectively here.
So that's the first portion of the what.
Down here, I'm talking more about some of the what, but also some of the why, and I'm
including some context around my offering.
So I'm saying here are the offerings that I have to offer.
And I also mentioned in here as well, that I'll give you a more detailed breakdown inside
of a markdown file that's within your knowledge base.
So noting to the AI that this is just a one sentence summary.
Or actually two sentences for two offerings.
And then I want you to look at these.
And then after you look at those, you can also go to this file as well to get more understanding
of what the offerings are.
And then finally down here, I'm giving it more of a how.
So this is around the structure of what it should give me back when it comes to the proposal
itself, saying that it needs to have these six components inside of the proposal when it
gives it back to me.
And then finally, another how is stating that I want this very specific file type given back
to me with the pixel perfect aesthetic.
So I can then just download it and use it.
And if you're not aware, this specific file type is for Word documents.
So that's our base prompt.
But we can run this through a prompt improver and get back a much more effective prompt
to get higher quality responses from our AIs.
And the way that you're going to do that, if you're going to use Claude specifically,
is you're going to go to, actually, it's a new URL now.
It's platform.claude.com.
And in here, you're going to drop in your prompt, the base prompt, and you're going to
say generate.
So let me actually go back because I've already generated this for myself.
You can see once you land on this page and they close this too, you're going to have
these options.
You can go generate prompt here, and that's where you paste in that base prompt.
So the base prompt that I just walked you through over here, I'll copy and paste this
in there.
I'll select generate.
But then, actually, before I select generate, I'm going to paste that in.
And I want to check this box here that says, is this prompt being used by thinking model?
Since we're using a reasoning model, Opus 4.5 with extended thinking on, we're going to
say generate and make sure that's checked.
And then you just wait for a few seconds, and then it regenerates that prompt in a more
effective way.
So instead of walking you through this, I'm going to go to the one that it's already generated
for me.
And I won't walk through the entire prompt because it is a similar prompt to what I've
just walked you through.
I'll just call it the areas that are new based off the prompt improver.
So first off, you can see that it's saying that you're an expert proposal writer with the emphasis
that it's writing for the specific company.
In addition to that, it added these two sentences here, calling out explicitly what it needs to
look for inside the knowledge base, including previous proposals, as well as the offerings
associated to this company.
And then if I scroll through here, you'll see that the format of the prompt itself has changed.
So it's added these headers here.
So these headers are allowing the AI to know what each segment of the prompt is dedicated
to.
So it knows that this is kind of the opener for the prompt.
Here, it knows that this is associated to the offering because of the header.
Here, it knows that this is associated to the proposal structure.
And you can see that it added it in here a variety of ways.
So it knows what part is which when it looks at it.
And the last two things I'll call it here is if you scroll down here that you'll see that
there's a segment called scratch pad.
So this is oftentimes this will be included when you use a prompt improver from Claude.
Reason being is that Claude's models, they do a pretty good job at actually thinking internally
and critiquing itself as it works to ensure that it's meeting your expectations.
And it's using the scratch pad as a way to think internally before it responds back to
you.
And then right after this, if I keep going, you'll see that it added a format requirement
section and then added the Claude skill like I mentioned previously.
So that's our improved prompt.
And now we're ready to actually embed this into the project and use it.
So let me go to the project and show you what that looks like.
So here, let me close this for a second.
So here at the very top, all I did was I pasted and I typed in this one line, draft a proposal
from the below transcript.
That's it.
I then pasted in a transcript, pushed enter, and then the AI gave me back a proposal that
looks identical to the template I provided previously.
So it is a pixel for pixel perfect match to my expectations and actually have it here already
open inside of Google drive.
And you can see this tab here is the template.
This tab is the output.
And if I go back and forth between the two, and I've already checked this pretty thoroughly,
it's identical to the template I provided.
So it's doing the job as expected.
And please note that this is all fictional and synthetic data.
It's not a real client.
So that's the end result.
Let me quickly show you how to create this project in the first place.
So if I go back here, you can see this is the starting project.
But if I wanted to create one from scratch, I would go over here to the project section.
I would click projects.
I would select new.
And then here I would name it.
So here we'll just name it test.
And then we'll do create project.
This is going to create the base project.
Then we want to add a few files.
The files I want to add here for my specific use case are going to be my core offerings, as well
as the template that the AI is going to be basing its references off of.
And that's exactly what you can see in the one that I've already drafted.
So here we have the offerings that I offer, as well as the template that it's going to reference
its writing off of.
And then in the instructions, these are the exact instructions I just showed you in the presentation.
I just copied and pasted them in here.
I updated the skill to call the right skill when it needed to be called.
And then we got that output back that I just showed you.
And that's how it works.
But I want to leave you with one thing.
So anytime you use an AI, there's a good chance that the first shot you get back from the AI
is not going to be perfect.
And that's where one of the most important skills when it comes to any client that I work with,
if they have this skill or this mentality, they're going to get very good at using AI.
They're going to create serious leverage within their business.
And that's persistence and being able to test effectively.
And I want to emphasize the importance of testing here.
And the reason being is that oftentimes we humans, we have a certain expectation when we start
using AI.
We say, this is what good looks like.
And when we start using AI, it's really over here.
So there's a huge gap between what good looks like in our eyes and what AI is giving back
to us.
And our job as AI native employees, leaders, and business owners is to get better at adjusting
the AI over time by giving it better context, giving it better system prompts, et cetera,
et cetera, iterating over time.
And as you iterate, eventually the AI starts to get closer and closer and closer to what good
looks like for a human.
And eventually these start to converge.
And once they converge is when you can actually outsource that specific task to AI completely.
And again, as I said at the beginning of the video, that's where real leverage begins to
occur within a business.
So it's important to note that you should have high quality testing and persistence when using
these different AIs.
Now, as a quick recap, as we stated previously, we're going to start with reverse engineering
a document that already is good for us.
So it's a contract, it's an SOP, it's a proposal, whatever else.
And we're going to reverse engineer the aesthetics of that first thing with an AI.
It needs to be a strong AI too.
After we've done this, we're going to then embed those learnings into a skill specifically
within Claude.
So we can call it at any time and get the pixel perfect match aesthetic back from the
AI.
And then we're going to embed that skill inside of a project that has a series of files,
system prompts, et cetera.
So we can come to it.
And all we have to do is drop in a transcript and it's going to then give back to us the document
that we care about.
And that's it.
So as a reminder, two quick things.
First off, blow is a 30 day AI insight series completely free.
You'll get 30 insights in your inbox if I can apply AI to your business and your work.
The second thing is if you'd like to work with me, blow a series of offerings, see if
there's a good fit between the two of us.
Now, you know how to turn one call into a polished proposal, but what if you have 50 calls and
you want to go across all those calls to find patterns across them?
Most people drop all of those files into an AI all at once.
And then it breaks.
I made a video that shows you how to process huge amounts of data without the AI losing
track.
You can check it out right here.
So go ahead, click that video and figure out how to give AI unlimited memory.
See you next time, internet.