---
title: '99% Of Claude Users STILL Don''t Know These 7 Official Prompting Rules'
source: 'https://youtube.com/watch?v=GKTkPnqhnf0'
video_id: 'GKTkPnqhnf0'
date: 2026-06-18
duration_sec: 0
---

# 99% Of Claude Users STILL Don't Know These 7 Official Prompting Rules

> Source: [99% Of Claude Users STILL Don't Know These 7 Official Prompting Rules](https://youtube.com/watch?v=GKTkPnqhnf0)

## Summary

The video presents the seven most important prompting rules from Anthropic's official guide for Claude, aimed at improving AI output quality. It emphasizes clarity, providing intent, using examples, focusing on desired formats, being direct with actions, and leveraging Claude's strengths in research and document creation. The rules are applicable to modern models like Claude Opus 4.5, Gemini 3, and GPT-5.2.

### Key Points

- **Be Clear and Specific** [0:31] — Vague prompts lead to generic 'AI slop' outputs because models default to common patterns (e.g., purple gradients in UIs). Specificity instructs the AI to escape this pull and generate tailored results.
- **Explain the 'Why' Behind Your Request** [3:08] — Adding the reason why you're asking something (e.g., 'formal tone for the board') helps Claude infer missing details and produce a higher quality response aligned with your true intent.
- **Give Good Examples (and Watch Your Own Prompt Style)** [4:24] — AI follows examples exactly. A poorly crafted example or even the style of your own prompt (playful vs. formal) will dictate the response style. This can be a trap if you're not careful.
- **Ask for the Desired Format (Positive Instruction)** [5:19] — Instead of saying what you don't want (e.g., 'don't use markdown'), tell the AI what you do want (e.g., 'smooth paragraphs with headers'). This is more effective with modern models.
- **Be Direct About Actions** [6:22] — Avoid suggestive language like 'suggest' or 'think about.' Use direct action verbs like 'change' or 'edit' to make the AI actually perform the task rather than just discuss it.
- **Use the Metaprompt for Research** [7:55] — Anthropic provides a 'metaprompt' for research: it instructs the AI to develop competing hypotheses, track confidence levels, self-critique its findings, and break complex questions into manageable parts. This dramatically improves research quality.
- **Leverage Claude for Document Creation** [9:43] — Claude is particularly good at creating formatted documents like presentations, reports, and Excel sheets. Bad prompts like 'write a report' can be improved by specifying sections, charts, action items, and formatting requirements.

## Transcript

Most people learn how to prompt Claude
from random tips on the internet. But
Anthropic, the company that actually
built Claude, released their official
prompting guide. Almost no one's read
it. I did. And here are the seven most
important rules straight from the people
who created the model itself. Let's get
into it. So, as I mentioned, these tips
came directly from Anthropic, the
company that made Claude. And this here
is the blog post that I'm referring to.
So, it walks through a series of tips,
not just seven, many more. But I wanted
to call out the seven that I feel like
are applicable to the broadest audience
where people can get direct impact of
using these tips. And we'll start with
tip one, which is to be clear. Now, this
is more critical than ever for all the
models that are relevant today, such as
Gemini 3, GBT 5.2, Claude Opus 4.5. All
of these models are very good at
following instructions. So, when you're
clear with the model, you're going to
get a higher quality response, and the
AI is going to work in the direction
that you care about. Vague prompts tend
to lead to common patterns that are kind
of considered AI slop today. So it's
generic outputs that aren't very useful.
Now why does this happen? It happens
because most models have a gravitational
pull towards what's most common. So here
we have gravity pulling in all the
commonalities of the internet. And
here's the default state. An example
here is UIs. So often times when an AI
creates a UI for you, it's going to have
purple gradients and it's going to have
rounded corners. Reason being is the AI
was trained on millions of examples that
had that type of aesthetic. So the AI
thinks that if it provides you that
aesthetic, you're going to be satisfied
as a user if you don't provide it very
specific instructions. And that's the
key here. We have to be specific to
escape the gravitational pool that is
all those examples the AI was trained on
historically. And that's what this tip
highlights. So I'm going to walk you
through a few examples of what good
looks like. So on the left hand side we
have a bad prompt. This is weak. So
we're basically stating create an
analytical dashboard. We we gave no
specificity here. A better prompt is
create an analytical dashboard. Include
as many relevant features and
interactions as possible. Go beyond the
basics to create a fully featured
implementation. And there's a few key
things I want to call out here. So first
off, we're stating I want you to include
as many. So be as inclusive as possible
when it comes to the features and
interactions relevant to the dashboard
we're trying to create. Another thing
we're making explicit is the AI needs to
go beyond the basics. So, it's going
beyond what's normal and creating
something that's fully featured that's
relevant to the initial dashboard we're
trying to create. And this is what a
good prompt looks like that's more
clear. And here's another example for
presentations. So, here we have a week
example that says create presentation.
An improved version of this is to create
a professional presentation in relation
to our quarterly results. I want you to
include thoughtful design elements,
visual hierarchy, and engaging
animations where appropriate. And as you
can see, we're being very clear on what
we want the AI to include in the visuals
here, and that's key. So, this is our
first tip, which is be clear, quick
pause in your regular programming. This
video is brought to you by me, as
always. So, two quick things. First off,
below is a 30-day AI insight series,
completely free. You'll get 30 insights
in your inbox how you can apply AI to
your business and your work. The second
thing is, if you'd like to work with me,
there's a variety of ways you can do
that, either through a private AI
community or directly one-on-one. You
can check that out below as well. Let's
get back into the video. Our second tip
is often something that a lot of people
skip, which is explain why. And what do
we mean by this? Well, when you give the
AI some sort of context, so you ask it
to do something. If you also add the
intent as to why you're asking the AI to
do this and why it's important for the
AI to take that task on, it can actually
infer a lot from what you've stated,
even if you've not made it explicit. In
other words, Claude can figure out
things that you didn't even say as long
as you're explicit about why you're
asking the AI to do this in the first
place. And that often usually deres a
higher quality output for you. Now let's
look at some examples. So this first
example here is a weak example. Not a
great prompt. We're saying write this in
a formal tone. It can be improved. A
better way of saying this is I want you
to write this in a formal tone because
remember we're emphasizing the why here.
It's going to our board of directors and
we need to look credible and
professional. By providing just this
context, the AI is going to be able to
create something that's more
aesthetically pleasing and aligned with
the quality you're seeking by knowing
this specific intent. Now, here's
another example. So, here we have the
weak prompt of saying, "Keep it short."
A better version of this is stating,
"Keep it short because, again, we're
highlighting the Y. I'm sending this to
my team via text message, and longer
messages don't get read." Again, we're
highlighting the Y, which is going to
get better quality outputs from the AI.
That's our second tip. And now, we'll
move on to tip three, which is giving
good examples. So the thing with
examples, especially in relation to the
state-of-the-art models today, such as
Opus 4.5, Gemini 3, GPT5.2,
all of these models, they follow the
examples to the tea. So if you include
an example in your system prompt, the AI
is going to follow that very
specifically. So it's important that you
don't misdirect the AI by accidentally
including something in your example that
you don't want in its output. And a meta
lesson in relation to this tip is not
only is the example that you explicitly
call in the prompt dictating what the AI
gives you, but also the way that you
write your prompt determines how the AI
gives you an output. What does that
mean? In the sense that if I write my
prompt in a very playful, fun, and
simple way, the AI is likely going to
respond and give me an output in that
way. If I might write my prompt in a
very formal and structured way, the AI
is going to do the same. So even the way
that I write my prompt is going to
dictate how the AI outputs its response
to me. So it's important to note not
just the example but also your prompt
matter. Now on to tip four which is
asking for the format. Not saying
something that you don't want. You want
to say what you do want because often
when you look at system prompts people
have really huge negating terms of
saying never do this or don't do that or
whatever else. Well with AI today since
it follows instructions so well all you
have to do is say what you want instead
of what you don't want and the AI is
likely going to do that more
effectively. What are some examples of
this? Well here's an example. There's
first one is a weak example. So we say
do not use markdown in this response.
The reason this is weak is because we're
using a negating term. So we're saying
do not. A better version of this is your
response should be composed of smoothly
flowing pros and paragraphs. This is
going to give us what we want instead of
avoiding what we don't want. Another
example is here we have a weak version
saying make it look nice, not very uh
not very clear. And a better version of
this is saying use clear headers in each
section. Bold the key takeaways. Add a
summary at the top. We're very explicit
about what we want. The AI is likely
going to achieve that because it's very
good at following instructions. So
that's tip four. Ask for the formats
that you want instead of saying what you
don't want. Now on to tip five, which is
being direct about actions. What does
this mean? Well, often times when you
want AI to do something for you, we need
to avoid suggestive language. So instead
of saying suggest, think about,
consider, you need to be clear and
saying I want you to change this now,
edit this thing, or make whatever. by
being explicit about it taking action,
it will then take the action. But
oftentimes the AI can default to being
more oriented towards not taking action
to make to ensure it's not breaking
anything or doing anything it shouldn't.
So if you say, "Hey, I want you to
suggest this." It will suggest
something. It won't change anything. If
you want it to change something, then
you explicitly state that you want it to
change it. So be clear on the action
verbs that you're using with the AI. And
here are just some examples. So on the
left hand side we have a weak prompt
which we're stating can you suggest some
changes to improve this function. The
reason this is weak is we're stating
suggest. We're being more careful with
our language which means the AI will be
careful with its actions. It's not going
to do anything. It's only going to
suggest what we need to do to the
function. This is in relation to code. A
more explicit way of doing this that's
more actionoriented is saying change
this function to improve its
performance. That's when the AI goes off
and does the action for you. Another
example that's unrelated to code is
proposal prompts. So a weak version of
this is what do you think about this
proposal? This is just looking for
feedback, no actions. If you want an
action to occur from the AI, we would
say something that's more strong, which
is edit this proposal to make the
benefits clearer and add a call to
action at the end. This means the AI is
going to edit. It's going to do a thing
because there is an action verb inside
the prompt. So be clear about actions.
If you wanted to suggest something or
act acts on something, you need to be
clear about that. Now, on to tip six,
where Claude is actually extremely good
at research. It's gotten much better
over the last iteration of versions that
that Anthropics released with Opus 4.5.
Now, let me show you a weak and good
example. So, a weak example here for
research in relation to AI is saying,
"Research my competitors." This is not
good because it's vague and it's not
clear on what you want. A better version
of this is research my top three
competitors in the home services
industry. For each one, I want you to
find their pricing, their main services,
and their customer reviews. Compare them
to my business and tell me where I have
an advantage. So here we're explicit
about the type of research we wanted to
do, what we wanted to do with the
research after it's gotten back and a
series of other things. So this is what
a good research prompt looks like. And
something that's really interesting
about the guide that was provided for
anthropic is they actually provided a
metaprompt. So this specific prompt
right here can be applied to almost any
research use case that you have and it's
going to improve it. I've been using it
for the last couple of days and it's
night and day in the quality of the
research that I'm getting getting back
from the AI by using this prompt. So, I
highly recommend just copying and
pasting this from their guide and using
it inside your research. Now, why is
this so good? Well, there's a few things
as to why this prompt is so useful.
First off, we're asking the AI to
research in a structured way. So, be
more formal in your process. But also,
we're asking as you gather data, I want
you to develop competing hypotheses. So,
multiple opinions against what's
happening here. In addition to that,
we're also saying, can you uh track your
confidence levels over time and adjust
them as you learn new things? because
you're going to regularly self-critique
your confidence levels as well as the
hypotheses that you build over your
research process. And then we're also
noting that the research needs to be
broken down from a complex ask to more
manageable asks that you then
consolidate over time. So there's a lot
of really interesting information in
this prompt as to why it's useful, but I
highly recommend just using this for
your research going forward because
you're going to get much better
responses back for things that truly
matter. And then our final tip is
Claude's really good at making documents
specifically because it uses a claude
skill or variety of claude skills that
allows it to create documents that are
more compelling. So this is going to be
in the relation to presentations,
animations inside of presentations
and/or landing pages like this or any
type of visual documents that's relating
to a PDF, an Excel sheet, etc. So it's
generally better at creating these types
of documents and they tend to be
formatted more effectively instead of
what we used to get back in the day with
other models. So what are some examples
of this? Well, the first one here is a
presentation. So a bad prompt is make me
a presentation. A better prompt, a
stronger one is create me a professional
presentation on topic. You fill in the
topic, including thoughtful design
elements, visual hierarchy, and engaging
animations when appropriate. So we're
asking specifically about certain design
elements and asking it to think hard
about those elements. So when it
provides us back a presentation, it's
more compelling. So it's a better
prompt. Another example here is a
report. So instead of saying, "Write me
a report." Bad prompt. A better prompt
is create a monthly report for my team.
Include a summary at the top, sections
for each department, charts showing our
progress, action items for the ne for
the next month, and use clean formatting
that's easy to scan. So, we're asking
for a lot of things here. We're asking
explicitly what should be included in
the prompt. We're being very uh specific
on the inclusion. We're also being very
explicit on the formatting as well
because we want be people to be able to
scan this instead of having to read it
or struggle to read through it. And
that's our seventh tip. And as a quick
recap, here are just seven rules that I
pulled from Anthropic themselves on how
you can get most the most out of their
models. So the first one here is about
being clear. You want to be clear with
the model because it's very good at
following instructions. So be clear in
your prompt. The second thing is always
provide your intent. Explain why as to
why you're asking the AI to do this
thing for you. It can infer more from
that and give you a higher quality
response. After that, you should give it
high quality examples if you decide to
give it examples because the AI is going
to follow them to the tea. that includes
the prompt itself of how you're writing
it. So, be very careful on the prompt
that you're drafting. After that, if you
care about formatting, you want to be
explicit about that and ask for certain
types of formats from the AI. And that
means we're asking the AI what to do
instead of what not to do. After this,
we have the importance of being direct
with the AI. So, if we want it to take
action, we need to be clear on it taking
that action. If we wanted to suggest
something, we need to be clear on that.
Also, it's very good at research. And
that metaprompt that I showed you, you
should use that probably for good on all
your research going forward with the
specific model because it's night and
day compared to what I've gotten in the
past. And then finally, it's very good
at creating documents and it's good at
following different types of formatting
um for generalized reports, Excel
sheets, presentations, etc. And that's
it. So if you enjoyed this, reshare with
your friends. And also, as a reminder,
two things. First off, below is a 30-day
AI insight series, completely free.
You'll get 30 insights in your inbox how
you can apply AI to your business and
your work. Second thing is if you'd like
to work with me, there's a variety of
ways you can do that either through a
private AI community or directly
one-on-one. You can check that out below
as well. Okay, so now we know that cloud
can create great documents. Anthropic
even says so. But here's the thing. It
still can't match your formatting, your
fonts, your colors, your style. You
still have to copy the content and fix
it by hand in the document. I found one
cloud feature that fixes this. Now,
every document that comes back to me is
exactly how I want it. Let me show you
right here. Go ahead, click that video
right there.